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.ppat.1002337 | IFITM3 Inhibits Influenza A Virus Infection by Preventing Cytosolic Entry | To replicate, viruses must gain access to the host cell's resources. Interferon (IFN) regulates the actions of a large complement of interferon effector genes (IEGs) that prevent viral replication. The interferon inducible transmembrane protein family members, IFITM1, 2 and 3, are IEGs required for inhibition of influenza A virus, dengue virus, and West Nile virus replication in vitro. Here we report that IFN prevents emergence of viral genomes from the endosomal pathway, and that IFITM3 is both necessary and sufficient for this function. Notably, viral pseudoparticles were inhibited from transferring their contents into the host cell cytosol by IFN, and IFITM3 was required and sufficient for this action. We further demonstrate that IFN expands Rab7 and LAMP1-containing structures, and that IFITM3 overexpression is sufficient for this phenotype. Moreover, IFITM3 partially resides in late endosomal and lysosomal structures, placing it in the path of invading viruses. Collectively our data are consistent with the prediction that viruses that fuse in the late endosomes or lysosomes are vulnerable to IFITM3's actions, while viruses that enter at the cell surface or in the early endosomes may avoid inhibition. Multiple viruses enter host cells through the late endocytic pathway, and many of these invaders are attenuated by IFN. Therefore these findings are likely to have significance for the intrinsic immune system's neutralization of a diverse array of threats.
| Influenza epidemics exact a great toll on world health. Thus research to identify new anti-influenza virus strategies would be useful. Each of our cells contains antiviral factors that work to inhibit infection. A large component of this antiviral program is regulated by the interferon family of signaling molecules. Here, we seek to better understand how one of these antiviral factors, IFITM3, contributes to both baseline, as well as interferon-induced, antagonism of influenza A viral infection. We found that interferon prevents influenza A virus from entering our cells by blocking the virus' fusion with the cellular membrane. Furthermore, we learned that IFITM3 is required for this antiviral action of interferon, and that high levels of IFITM3 alone can produce a similar viral inhibition. Together, these results improve our understanding of how IFITM3 serves to defend us against viral invasion at a very early stage of infection.
| The 2009 H1N1 pandemic provided a strong reminder of the threat that influenza A virus poses to world health (http://www.cdc.gov/h1n1flu/cdcresponse.htm). The most effective means of protection against influenza is the seasonal vaccine. However, if the vaccine does not match the viral strains, its effectiveness can be reduced to 50% or less [1], [2]. Among small molecules, only two approved influenza drugs remain effective, zanamivir (Relenza) and oseltamivir (Tamiflu). Although resistance to zanamivir is rare, there has been an increase in oseltamivir-resistant flu strains [3]. Of concern, both drugs target viral neuraminidase (NA), precluding combinatorial therapy to minimize resistance [4], [5]. Thus, research to identify new anti-influenza strategies would be useful.
The influenza A virus is 50–100 nm in size, encodes for up to 11 proteins, and contains eight segments of negative single-stranded genomic RNA (3). Influenza A virus infection initiates with the cleavage and activation of the viral hemaglutinnin (HA) envelope receptor by host proteases [6], [7], [8], [9]. HA then binds to sialylated proteins on the cell surface, eliciting endocytosis of the viral particle. Endocytosed viruses are transported through the early and late endosomes, with late endosomal acidification triggering a conformational change in HA which results in viral-host membrane fusion [6], [10]. Fusion transitions from a hemifusion intermediate into a fusion pore through which the virus' eight viral ribonucleoproteins (vRNPs) enter the cytosol. The vRNPs are subsequently guided by the host cell's karyopherins into the nucleus [11], [12], [13], wherein the viral RNA-dependent RNA polymerase synthesizes viral genomes (vRNA) and mRNAs, both of which are exported to the cytosol, culminating in the production of viral progeny.
Genetic screens have identified multiple host factors and pathways which modulate influenza A virus infection in vitro [14], [15], [16], [17]. Using such a genetic screen, we identified the IFITM protein family members IFITM1, 2 and 3 as antiviral factors capable of blocking influenza A viruses [14]. We further tested the antiviral activity of IFITM3 protein using the seasonal influenza A strains, A/Uruguay/716/07 (H3N2) and A/Brisbane/59/07 (H1N1), and found similar levels of IFITM3-mediated viral inhibition [14]. IFITM3 accounts for a significant portion (50–80%) of IFN's (type I or II) ability to decrease influenza A virus infection in vitro, and IFITM3 resides in vesicular compartments that are IFN-inducible [14]. In addition, the IFITM family inhibits infection by the flaviviruses, dengue virus and West Nile virus [14], [18], as well as the filoviruses, Ebola and Marburg, and the SARS coronavirus [19]. The IFITM proteins also block vesicular stomatitis virus-G protein (VSV-G)-mediated entry, but do not substantially alter the replication of Moloney leukemia virus (MLV), several arena viruses, or hepatitis C virus (HCV, [14], [20]).
The human IFITM proteins were identified 26 years ago based on their expression after IFN stimulation [21], [22], [23]. The IFITM1, 2, 3 and 5 genes are clustered on chromosome 11, and all encode for proteins containing two transmembrane domains (TM1 and 2), separated by a conserved intracellular loop (CIL, [22]), with both termini extra-cellular or intra-vesicular [24], [25]. TM1 and the CIL are well conserved between the IFITM proteins and a large group of proteins representing the CD225 protein family. CD225 family members exist from bacteria (125 members) to man (13 members, with 156 members in chordata), with no in depth functional data available for any member other than the IFITM proteins. IFITM1, 2 and 3 are present across a wide range of species including amphibians, fish, fowl and mammals. The IFITM proteins have been described to have roles in immune cell signaling and adhesion, cancer, germ cell physiology, and bone mineralization [25], [26], [27], [28], [29], [30]. IFITM3 expression can inhibit the growth of some IFN-responsive cancer cells [31]. Genetic evidence also points to IFITM5/Bril being required for early bone mineralization [30], [32]. IfitmDel mice, which are null for all five of the murine Ifitm genes, display a 30% perinatal mortality among null pups, but thereafter grow and develop normally in a controlled setting [26]. However, cells derived from these IfitmDel mice are more susceptible to influenza A virus infection in vitro [14]. IFITM3 inhibited infection by all influenza A virus strains tested including a 1968 pandemic isolate and two contemporary seasonal vaccine viruses [14]. We have found IFITM3 to be the most potent of the IFITM protein family members in decreasing influenza A virus replication [14].
Viral pseudoparticles are differentially inhibited by the IFITM proteins based on the specific viral receptors expressed on their surfaces [14], [19]. Therefore, we have hypothesized that IFITM proteins inhibit susceptible virus families (Orthomyxoviridae, Flaviviridae, Rhabdoviridae, Filoviridae, and Coronaviridae) during the envelope-dependent early phase of the infection cycle, which extends from viral binding to cell surface receptors through the creation of the fusion pore between viral and host membranes [14], [19], [20]. In support of this notion, recent work demonstrated that IFITM protein overexpression did not prevent influenza A virions from accessing acidified compartments [19]. Consistent with its acting on endocytosed viruses, a portion of IFITM3 resides in structures that contain host cell endosomal and lysosomal proteins [19]. Furthermore, inhibition of influenza A virus infection depends on the palmitoylation of IFITM3, a post-translational modification that targets proteins to membranous compartments [33].
Here we directly test the idea that IFITM3 restricts influenza A viral infection during the envelope-dependent early phase of the viral lifecycle. Consistent with previous studies, we find that IFITM3 inhibits influenza A viral infection after viral-host binding and endocytosis, but prior to primary viral transcription [19], [20]. Moreover, using a combination of assays, we find that either IFN or high levels of IFITM3 impede influenza A viruses from transferring their contents into the host cell cytosol, and that IFITM3 is necessary for this IFN-mediated action. Therefore, we conclude that IFN is acting predominantly through IFITM3 to block viral fusion. We also find that IFN expands the late endosomal and lysosomal compartments, and that IFITM3 overexpression is sufficient for this phenotype. This study also presents data showing that IFITM3 overexpression leads to the expansion of enlarged acidified compartments consisting of lysosomes and autolysosomes. Interestingly, we observe that viruses trapped in the endocytic pathway of IFITM3-overexpressing cells are trafficked to these expanded acidified compartments. Based on these results and those of others [19], [20], we present a model whereby IFN acts via IFITM3 to prevent viral fusion, thereby directing endocytosed viruses to lysosomes and autolysosomes, for subsequent destruction. Collectively this study expands our understanding of how IFITM3 restricts a growing number of viruses by exploiting a shared viral vulnerability arising from their use of the host's endocytic pathway.
The inhibition of HA-expressing pseudoparticles by the IFITM proteins pointed towards restriction occurring during the envelope-dependent phase of the viral lifecycle [14]. Therefore we tested IFITM3's impact on the most proximal phase of infection, viral binding, by incubating influenza A virus A/WSN/33 H1N1 (WSN/33, multiplicity of infection (moi) 50) with A549 lung carcinoma cells either stably overexpressing IFITM3 (A549-IFITM3) or an empty vector control cell line (A549-Vector, Fig. 1A). Samples were incubated on ice to permit viral binding but prevent endocytosis. After incubation, cells were washed with cold media, fixed and stained for HA. When analyzed by flow cytometry, we observed no appreciable difference in surface bound HA between the vector and IFITM3 cells. There was also no difference in surface-bound virus over a series of ten-fold dilutions of viral supernatant (data not shown). We also determined that the stable expression of IFITM3 did not alter the surface levels of (α2, 3) or (α2,6) sialylated cell-surface proteins (Fig. S1).
To investigate IFITM3's impact on initial viral mRNA production, we infected canine kidney cells, either expressing IFITM3 (MDCK-IFITM3) or the empty vector (MDCK-Vector), with influenza A virus (A/Puerto Rico/8/34 H1N1 (PR8), moi 500). We used PR8 because of the purified high titer stocks available. Next, the viral supernatant was removed and warm media was added (0 min). At the indicated times, cells were processed and stained for the positive stranded NP mRNA of PR8 using a specific RNA probe set (red, Fig. 1B), then imaged on a confocal microscope. Based on NP mRNA staining, primary viral transcription begins by 60 min. p.i. in the vector control, with the NP mRNA signal increasing through to 180 min., when the export of viral mRNAs to the cytosol can be observed. A decrease in primary viral transcription can be seen when comparing the IFITM3 cells to the vector control line. Therefore, IFITM3 inhibits influenza A viral infection after viral-host binding but before primary viral mRNA transcription.
We next used confocal imaging to track the nuclear translocation of vRNPs (Fig. 2 [34], [35]). At the start of infection, the NP within infected cells is complexed with viral genomic RNA forming vRNPs. Therefore, immunostaining for NP permitted us to follow vRNP distribution intracellularly [16], [34], [36]. Normal diploid human lung fibroblasts (WI-38 cells) were stably transduced with empty vector (Vector), IFITM3 cDNA (IFITM3), or short hairpin RNAs (shRNA) either against IFITM3 (shIFITM3) or a scrambled non-targeting control (shScramble, Fig. 2, S2). WI-38s were chosen because of their normal karyotype and relatively larger and flatter morphology. Cells were first incubated on ice with PR8 (moi 500). Next, the viral supernatant was removed and warm media was added (0 min). At the indicated times after warming, cells were fixed, permeabilized, stained for NP and DNA, and imaged on a confocal microscope. Image analysis software was used to create an outline of each cell's periphery (white lines) and nucleus (blue lines). Based on NP staining, vRNPs arrive in the nuclei by 90 min in the vector control, shIFITM3, and in the shScramble cells, with the NP signal increasing through to 240 min (Fig. 2A, S2A–D).
In contrast, we observed decreased nuclear and increased cytosolic NP staining in the IFITM3 cells (Fig. 2, S2C). Moreover, in the IFITM3 cells greater than 60% of the cytosolic NP colocalized with Lysotracker Red (LTRed), a dye which marks acidic cellular compartments (late endosomes, lysosomes, pH≤5.5), and which was added to the warm media at time zero (Fig. S2A, D). The increased NP in the cytosol of the IFITM3 cells likely arises in part from an increase in the local concentration of viruses because α-NP Western blots (after trypsinizing the cells to remove adherent NP) did not show substantial differences in internalized NP levels between cell lines for up to 90 min post infection (p.i., data not shown). Because IFITM3 is required for the anti-viral actions of IFN in vitro [14], we performed a companion experiment with the WI-38 cells treated with IFN-α (Fig. 2B). IFN-α treatment also decreased NP nuclear staining in the WI-38-Vector cells, however this block was not as complete nor was it associated with similar levels of cytosolic NP staining as those seen with high levels of IFITM3. Consistent with the gain-of-function data, the depletion of IFITM3 decreased IFN's ability to block vRNP trafficking to the nucleus (Fig. 2A and B, compare top and bottom rows).
Similar results were obtained either using A549 cells (Fig. S3) or using MDCK cells, with the latter experiments employing additional influenza A viral strains (X:31, A/Aichi/68 (Aichi H3N2), Fig. S4A–C, WSN/33 and A/Victoria/3/75 H3N2, data not shown). It is important to note that the levels of IFITM3 protein in the A549-IFITM3 cells are higher than those seen after treatment with IFN-α or -γ (Fig. S3C). However, we have not observed that other overexpressed proteins have either protected against viral infection or expanded the lysosome/autolysosome compartment (data not shown), arguing that this is a specific effect. To better assess the expanded LTRed compartments observed with IFITM3 overexpression, we created MDCK cells stably expressing the lysosomal protein, LAMP1, fused to a red fluorescence protein (LAMP1-RFP) and IFITM3. As compared to control cells, the IFITM3 cells demonstrated extensive colocalization (>60%) between the NP and LAMP1-RFP signals, revealing that the entering viruses are trafficked to lysosomal compartments (Fig. S5).
We extended this analysis by directly tracking the location of the vRNA contained in the incoming vRNPs. MDCK cells stably expressing an empty vector or IFITM3, were used in time-course experiments as above (Fig. 3A–D). At the indicated times, cells were processed and stained for the negative stranded NP vRNA of PR8 using a specific RNA probe set (green). As seen with the WI-38 cells, we observed the nuclear translocation of vRNA by 80 min p.i. in the MDCK-vector cells (Fig. 3A). The nuclear vRNA signal was strongly decreased with IFITM3 overexpression based on the average number of vRNA particles present per nucleus (Fig. 3C). Consistent with the WI-38 results, the vRNAs accumulated in the cytosol of the IFITM3 cells, with >50% co-localizing with LTRed-staining acidic structures (Fig. 3D). Similar levels of retained cytosolic vRNPs were observed in experiments without LTRed (data not shown). Interestingly, we observed the loss of the vRNA signal in the acidic inclusions of the MDCK-IFITM3 cells between 80 and 240 min. p.i. (Fig. 3B). By comparison, the vRNAs in the control cells increased in number in both the nucleus and cytosol, as would be expected with the nuclear export of newly synthesized viral genomes [36].
We next evaluated vRNP translocation in murine embryonic fibroblasts (MEFs) derived from animals that have had all five Ifitm genes deleted (IfitmDel−/−, [14], [26]). Compared to wild-type (WT) matched litter mate controls, the IfitmDel−/− MEFs displayed 5–10 fold more nuclear NP staining, with or without IFN-γ treatment (Fig. 4, S6C). IFN-mediated viral restriction was restored when we transduced the null MEFs with a retrovirus expressing Ifitm3 (IfitmDel−/− Ifitm3, Fig. S6). Similar to what was observed with the IFITM3 overexpressing cell lines, the majority of the vRNP signal in the IFN-γ-treated WT and Ifitm3-rescued cells localized to acidic compartments (red, Fig. S6B). An increase in acidic compartments occurred after IFN-γ treatment with either the WT or the IfitmDel−/−Ifitm3 MEFs, but not in the IfitmDel−/− cells, suggesting that Ifitm3 is required for this event (Fig. 4, S6). Similar results were obtained with IFN-α (data not shown). We conclude from these experiments using orthologous reagents (cell lines and influenza A viruses) and methods, that IFN impedes vRNP nuclear entry, and IFITM3 is necessary and sufficient for this activity.
To further characterize the mechanism of IFITM3-mediated restriction, we used an established viral fusion assay [37], [38]. Lentiviral pseudoparticles containing the β-lactamase protein fused to the HIV-1 accessory protein Vpr (BLAM-Vpr) and expressing either HA and NA (H1N1, WSN/33), or VSV-G envelope proteins, were incubated for 2 h with cells, which were then loaded with the β-lactamase flourogenic substrate, CCF2. Upon viral pseudoparticle fusion, BLAM-Vpr enters the cytosol and cleaves CCF2, producing a wave length shift in emitted light (from green to blue) when analyzed by flow cytometry (Fig. 5A, [37]). In MDCK-IFITM3 cells we observed a decrease in both HA- and VSV-G-directed fusion, which was comparable to the block produced by poisoning of the host vacuolar ATPase (vATPases) with a low dose of bafilomycin A1 (Baf, Fig. 5B). The inhibition of vATPases prevents the low-pH activation required by these two viral envelope proteins to produce membrane fusion. A block to fusion of pseudoparticles expressing H1 (PR8), H3 (A/Udorn/72), H5 (A/Thai/74) or H7 (A/FPV/Rostock/34) subtypes of HA was also detected with MDCK cells or with chicken embryonic fibroblasts (ChEFs), in which IFITM3 strongly inhibited viral replication (Fig. S7A, B, C). In the case of the MDCK cells, the block to fusion closely paralleled the level of inhibition seen when the pseudoparticles were tested for productive infection using HIV-1 p24 expression as a readout (Fig. S7E). Consistent with earlier findings, pseudoparticles expressing an amphotropic MLV envelope protein were insensitive to IFITM3, showing the specificity of these results (Fig. S7D). Similarly to its effect on H5-expressing pseudoparticles, IFITM3 inhibited replication of infectious avian H5N1 influenza A virus, A/Vietnam/1203/04 (VN/04), isolated from a fatal human infection (Fig. S7F–H).
To enhance our analysis, we tested two additional cell lines, WI-38 and HeLa cells. A strong block to fusion in WI-38-IFITM3 cells, similar to that of the Baf and uninfected control samples, was seen at a range of serial dilutions of pseudoparticles, as well as an increase in fusion with IFITM3 depletion (shIFITM3, Fig. 5C, D). IFN treatment inhibited fusion of the H1N1 pseudoparticles, albeit to a lesser extent than IFITM3 overexpression (Fig. 5E), and this effect was largely absent when IFITM3 was stably depleted in HeLa cells (Fig. S8). Similar results were obtained with IFN-α (data not shown). Based on these experiments using multiple cell lines and HA, VSV-G, and MLV envelope-expressing pseudoparticles, we conclude that IFITM3 is required and sufficient for an IFN-mediated block of viral pseudoparticle fusion. Importantly, the increase in pseudoparticle fusion seen when endogenous IFITM3 was depleted in either the HeLa or WI-38 shIFITM3 cell lines argues that fusion inhibition underlies the first line defense provided by endogenous, as well as overexpressed, IFITM3.
MxA is an IFN-inducible large GTPase which interferes with secondary transcription during influenza A viral replication [39]. A549 cells express MxA and have been used extensively in influenza A viral replication studies [40]. Therefore to clarify the antiviral roles of IFITM3 and MxA, we tested the levels of viral replication in A549 cells stably expressing one of three shRNAs targeting IFITM3 (shIFITM3-1, -2, or -3). All three shIFITM3 cell lines showed increased infection (WSN/33 strain) and strong IFITM3 knockdown, when compared to the negative control cell line expressing a shRNA against firefly luciferase (shLuc), with or without IFN treatment (Fig. S9A, B). The majority of the protective effect of either IFN-α or γ was lost in the shIFITM3 cell lines. We next confirmed both the baseline levels, as well as the IFN-inducibility of MxA in the A549 cells (Fig. S9C). We also determined that MxA was both present and IFN-inducible in WI-38 normal fibroblasts, another cell line used in loss-of-function experiments in this work (Fig. S9D). Furthermore, IF studies of WI-38 cells showed that MxA is expressed in an IFN-inducible vesicular pattern and that these structures did not appreciably co-localize with vesicles containing IFITM3 (Fig. S9E, [39]). We conclude that MxA is expressed in the A549 and WI-38 cell lines, but cannot fully compensate for loss of the antiviral actions of IFITM3.
Our data demonstrate that IFN or IFITM3 inhibit viral fusion. Influenza A virus fuses with the host membrane in late endosomes when the pH decreases to 5 [6], [7], [41]. Rab7 is a late endosomal/lysosomal small GTPase that is required for the fusion of many pH-dependent viruses, including influenza A virus [6], [41]. Previous reports have shown that IFITM3 colocalizes with LAMP1 and CD63, components of lysosomes and multivesicular bodies, respectively [19]. However, the relationship of IFITM3 and Rab7 within the host cell infrastructure remains unknown. Therefore we investigated the location of IFITM3, by undertaking immunoflourescence (IF) studies using antibodies that recognize IFITM3, Rab7, or LAMP1 [42]. Although the baseline level of IFITM3 in the A549-Vector cells was low, there was partial colocalization observed with either Rab7 or LAMP1 (Fig. 6A–D, 7A,). IFITM3 also partially colocalized with LAMP1 and LTRed-containing structures seen with IFITM3 overexpression (Fig. 6A, B, 7A). Interestingly, either IFITM3 overexpression or IFN increased the staining intensity of Rab7 and LAMP1 (Fig. 7A, B, S10A). Partial colocalization of IFITM3 was also seen with either endogenous LAMP1, or an exogenously expressed Rab7-yellow fluorescence fusion protein (Rab7-YFP) in MDCK cells (Fig. 6E–I). However, in all cases, co-localization was not complete because cells contained areas that uniquely labeled for each of the proteins. Western blots indicated that IFITM3 over-expression led to modest increases in both LAMP1 and Rab7 proteins in the A549-IFITM3 cells (Fig. 7C). However, these blots also showed that while IFN treatment of the A549-Vector cells increased IFITM3 protein levels as expected, the amount of Rab7 and LAMP1 remained unchanged. We conclude that IFITM3 partially resides in the late endosomal and lysosomal compartments along with Rab7 and LAMP1, and that IFITM3 overexpression or IFN treatment expands these compartments through a mechanism that cannot be fully explained by increased protein expression alone.
Our assays showed that incoming influenza A viruses were retained in the expanded acidic compartments of both the IFITM3 overexpressing cell lines as well as the IFN-γ-treated MEFs, and that IFITM3 partially localized to these structures (Fig. 2–4, S2–4, S6). Therefore, we extended our investigation of these compartments. An increase in acidic structures was seen in MDCK and A549 cells overexpressing IFITM3 as compared to control cell lines, using either the vital acidophilic stain, acridine orange (AO), LTRed, or a cathepsin-L substrate that fluoresces only after it is proteolyzed, when compared to the corresponding vector control cells (Fig. 8A, B, a, b). Cathepsins are a family of lysosomal zymogens active in acidic environments (pH≤5.5) which are required for both the degradation of endocytic substrates and for the entry of several IFITM3-susceptible viruses [19]. Flow cytometry revealed an increase in the total LTRed fluorescent signal in both the MDCK and A549 IFITM3 cell lines when compared to controls (Fig. 8C). This expanded compartment represents a heterogeneous population of lysosomes and autolysosomes, based on confocal imaging showing the colocalization of the autophagosome marker, microtubule-associated protein 1 light chain 3 (LC3), with either LTRed or with CD63, with the latter being a resident of multivesicular bodies, amphisomes and autolysosomes (Fig. 8D, E). Furthermore, MDCK-IFITM3 cells stably transduced with an LC3 protein fused to both a red fluorescent protein (mCherry) and an enhanced green fluorescence protein (EGFP) showed a predominantly red signal, which occurs when the mCherry-EGFP-LC3 protein resides inside the acidified interior of an autolysosome (Fig. 8F, [43]). In keeping with previous reports that IFN-γ induces autophagy [44], [45], we detected enhanced LTRed staining in either IFN-γ treated MEFs or A549 cells (Fig. 4A, S10A). We conclude that increases in IFITM3 levels expand the lysosomal/autolysosomal compartment.
Here we report several novel findings regarding the antiviral actions of IFN and the transmembrane IEG, IFITM3. First, this study demonstrates that IFN inhibits the nuclear translocation of vRNPs, and that IFITM3 is required for this IFN-mediated block, with both endogenous and overexpressed IFITM3 inhibiting vRNP nuclear entry. Second, either endogenous or overexpressed IFITM3, as well as IFN treatment, block the fusion of viral pseudoparticles expressing various influenza A virus envelope proteins (H1, H3, H5 and H7 subtypes of HA), or the VSV-G envelope protein; this block is specific because the fusion of pseudoparticles expressing MLV envelope is not inhibited by IFITM3. Third, our work reveals that IFITM3 partially resides with Rab7 in late endosomes, thus placing it in position to block influenza A virus' cytosolic access. Fourth, IFITM3 overexpression or IFN induce the expansion of late endosomal and lysosomal compartments containing Rab7 and LAMP1. Fifth, we show that similar to IFN-γ treatment, IFITM3 overexpression expands the number and size of autolysosomes, and it is into these compartments that trapped viruses are trafficked and subsequently degraded. Consistent with previous reports, our data show that high levels of IFITM3 do not prevent viral access to acidified compartments and that IFITM3 colocalizes with CD63 and LAMP1 [19]. This is in contrast to a report noting the exclusion of overexpressed IFITM3 from LAMP1-containing structures [33]. Therefore, this work adds substantially to our interpretation of previous reports by demonstrating that key downstream events in the viral lifecycle, fusion and vRNP nuclear translocation, are prevented by either IFN or IFITM3. IFITM3 thus represents a previously unappreciated class of anti-viral effector that permits viral entry into the endosomal compartment, but prevents egress into the cytosol. These studies also raise new questions including i) how do IFN and IFITM3 prevent viral fusion? ii) how do IFN and IFITM3 alter the endosomal and autolysosomal compartments? and iii) is the latter action required for viral restriction, or alternatively does it arise as an outcome of IFITM3's potential cellular role?
Based on the substantial loss in IFN's potency observed when IFITM3 is depleted (50–80% loss of viral inhibition, Fig. S9A, B, [14]) we conclude that inhibition of viral emergence from the endosomal pathway is a prominent component of IFN's antagonism of influenza A virus replication in vitro. Our data also show that MxA cannot fully compensate for the loss of IFITM3 in IFN-treated cells challenged with influenza A virus. Recent work by Dittmann et al. [46] and Zimmermann et al. [47] reveal that human influenza A viral strains have evolved a means to evade MxA, suggesting a possible explanation for the cellular reliance on IFITM3 for protection in vitro. Similarly the IEG, IFIT1, prevents viral replication by targeting viral 5′ triphosphate-RNAs (PPP-RNA) for destruction [48], [49]. Given that IFITM3 is necessary for the majority of IFN-mediated restriction of influenza A virus in vitro, it may be that the virus has also evolved a means to at least partially nullify IFIT1, perhaps via the massive production of short “decoy” PPP-RNAs, as previously postulated [49], [50].
IFITM3 primarily resides in the endosomal compartment and partly colocalizes with Rab7 and LAMP1. IFITM3 overexpression or IFN stimulation caused the endocytosed viruses to accumulate in acidic compartments that contained both IFITM3 and LAMP1. Together with the BLAM-Vpr fusion assay data, these results reveal that IFITM3 prevents viral-host membrane fusion within late endosomes, and likely within lysosomes as well, in light of studies showing IFITM-mediated restriction of filoviruses and coronaviruses, which depend on cathepsin-mediated activation prior to fusion [19]. In doing so, IFITM3 traps the virus on a path which terminates in a degradative environment [51]. In support of this, our experiments show the eventual loss of a detectable vRNA signal in the LTRed-positive compartments of the IFITM3-transduced cells, thus revealing the fate of viral fitness under those conditions.
These studies also reveal that elevated levels of IFITM3 correlate with the expansion of host cell structures containing Rab7 and LAMP1, and that IFITM3 was also present in these structures. In the MEF and A549 experiments, IFN produced increased Rab7 and LAMP1 immunostaining, in addition to an increase in acidic structures. At present, we cannot explain the increased Rab7 and LAMP1 signals seen after IFN stimulation or IFITM3 overexpression solely on the slight elevations in the abundance of these proteins detected by immunoblotting. Two possible explanations for the increased immunostaining observed, are that IFN stimulation induced these proteins to cluster together or alternatively unmasked sequestered epitopes; we find the latter possibility less likely since LAMP1 and Rab7 flourescent fusion proteins also showed larger and more intense signals under similar conditions. We envision that IFITM3-mediated clustering of organelles and their protein cargoes might contribute to the host cell's antiviral state. Earlier work reported no correlation between the size of the IFITM3-induced acidified compartments and the level of viral restriction [19], however, we observe that increasing levels of IFITM3 result in both an expansion of lysosomes/autolysosomes and increased viral inhibition. These observations might be explained by a common mechanism underlying the increase in these structures and viral inhibition, in addition to raising the possibility that they play a role in IFITM-mediated viral restriction.
Is there a common characteristic shared by IFITM3-susceptible viruses? The late endosomal- and lysosomal-associated small GTPase, Rab7, is required for influenza A virus infection [7], [41]. The IFITM3-resistant viruses previously tested (MLV, the arena viruses and the hepacivirus, HCV) are all Rab7-independent, while the entry of the IFITM3-susceptible viruses (influenza A, dengue, Ebola, Marburg, and SARS) relies on Rab7 [14], [19], [41], [52], [53], [54]. Standing against this hypothesis, is the lack of effect on VSV-G-mediated entry with expression of a dominant negative Rab7 [41], [55], [56]). However, additional studies have shown that VSV-G-directed entry is dependent on transport to the late endosomes [57], [58]; these latter results, together with those of Huang et al. and Weidner et al. [19], [20], are consistent with the prediction that viruses that fuse in late endosomes or lysosomes are vulnerable to IFITM3's actions, while viruses whose genomes enter at the cell surface or in the early endosomes may avoid IFITM3's full effect. Of note, we have been unable to demonstrate that IFITM3 blocks HIV-1 replication using TZM-bl HeLa cells and are working to address these differences with a published study ([59], data not shown).
This study, together with previous work, demonstrates that IFITM3 permits endocytosis of viruses, but prevents viral fusion and the subsequent entry of viral contents into the cytosol [19], [20]. While the BLAM-Vpr fusion assay demonstrates inhibition of fusion by IFN or by IFITM3, we note that this assay uses an indirect readout to assess entry of viral contents. Therefore several possibilities could explain the containment and neutralization of viruses within the endosomal pathway, including alterations in endosomal trafficking, acidification, or the host membrane's fusion characteristics (bending modulus, elasticity). While additional work is required to further define the mechanism, the lack of toxicity seen with cells stably overexpressing high levels of IFITM3 suggests that gross alterations in endogenous trafficking or pH control are unlikely (data not shown). Therefore overexpressing or activating IFITM3 to produce an enhanced antiviral state may be an effective prevention strategy during high risk periods in vulnerable populations.
We propose that IFN causes the degradation of endocytosed viruses by preventing their contents from entering the host cytosol, and that IFITM3 is necessary and sufficient for this defense (Figure 8G). IFITM3's mode of defense could be envisioned as an effective means to neutralize pathogens during an organism-wide threat. Such actions might confer an advantage to the host because if IFITM3 simply decreased viral attachment and/or entry, the repulsed viruses would be free to attack neighboring cells. Of course while there are considerable differences between this simple scenario and the directed phagocytosis of pathogens by specialized immune cells, i.e. macrophages, the similarities none-the-less suggest an early prototype for a more evolved defense mechanism.
U2OS, A549, MDCK, HeLa cells (all from ATCC), and chicken embryonic fibroblasts (ChEFs, from Charles River Labs) were grown in complete media (DMEM, Invitrogen Cat#11965) with 10% FBS (Invitrogen). WI-38 cells (ATCC) were cultured in DMEM (Invitrogen Cat#10569), containing non-essential amino acids (Invitrogen Cat#11140) and 15% FBS. Wild type and matched IfitmDel−/− MEFs were from adult IfitmDel+/− mice [26] that were intercrossed and MEFs derived from embryos at day 13.5 of gestation, as described previously [14]. The MEFs were genotyped by PCR and Western blot, and the generation of the IfitmDel−/− Ifitm3 cells have been previously described [14].
The IFITM3 retroviral vector, pQCXIP-IFITM3 and empty vector control (Clontech) have been previously described [14]. The shRNA lentiviral vectors, pLK0.1-Scramble and pLK0.1-shIFITM3-3 (clone ID HsSH00196729) are available from the Dana Farber DNA core, Harvard Medical School, Boston, MA. pCAGGS-HA WSN/33 and pCAGGS-NA WSN/33 were kind gifts of Dr. Donna M. Tscherne and Dr. Adolpho Garcia-Sastre, Microbiology Dept., Mt. Sinai School of Medicine, NY, NY [38]. pBABE-mCherry-EGFP-LC3B was from Addgene (Plasmid #22418) and was kindly deposited by Jayanta Debnath. pLZS-Rab7-YFP and pLVX-RFP-LAMP1 were generously provided by Walther Mothes, Section of Microbial Pathogenesis, Yale University School of Medicine. The following shRNA sequences (sense strand sequence provided) were cloned into the pAPM shRNA-expression lentiviral vector [60], to create the viruses used to generate the A549 IFITM3 knockdown cell lines in Fig. S9:
IFITM3-1: 5′-TCCTCATGACCATTCTGCTCAT-3′
IFITM3-2: 5′-CCCACGTACTCCAACTTCCATT-3′
IFITM3-3: 5′-TTTCTACAATGGCATTCAATAA-3′
Influenza A virus A/Puerto Rico/8/1934 (H1N1) (PR8, Charles River Labs) and A/WSN/33 (H1N1) (kind gift of Dr. Peter Palese, Microbiology Dept., Mt. Sinai School of Medicine, NY, NY) were propagated and assessed for viral infectivity as previously described [14]. Influenza A virus A/Vietnam/1203/2004 (H5N1) was propagated and characterized as previously described [61].
Human interferon (IFN)-γ (Invitrogen) was used at 100–300 ng/ml, human IFN-αA2 (PBL Interferon Source) was used at 500–2500 U/ml. Cells were incubated with cytokines for 16–24 h prior to IF or viral infection experiments unless otherwise noted. Murine IFN-γ (PBL Interferon Source) was used at 100–300 ng/ml.
Whole-cell extracts were prepared by cell lysis, equivalent protein content boiled in SDS sample buffer, resolved by SDS/PAGE, transferred to Immobilon–P membrane (Millipore), and probed with the indicated antibodies.
Cells were seeded on glass coverslips for Influenza A virus infection experiments. Cells were incubated on ice with PR8 for 40 min. At time zero, the viral supernatant was removed and 37°C media was added with or without Lysotracker Red DND-99 (Invitrogen). At the indicated time points post-warming, cells were washed twice with D-PBS (Sigma) and incubated for 30 seconds with room temperature 0.25% trypsin (Invitrogen). The cells were then washed with complete media twice and fixed with 4% formalin (PFA, Sigma) in D-PBS. Image analysis for quantitation of vRNP nuclear translocation was done using Imaris 7.1 (bitplane scientific software). We generated a mask of the nucleus and applied this mask to the channel containing the viral signal (puncta) to determine vRNA puncta contained in each nucleus.
Cells were incubated at 37°C and 5% CO2 for 60 min. with either Lysotracker Red DND-99 or acridine orange (ImmunoChemistry Technologies). Hoechst 33342 (DNA stain, Invitrogen) was incubated (1∶10,000) with the cells for the final 15 min. The Cathepsin L flourogenic substrate assay was performed as per the manufacturer's instructions (Cathepsin L -Magic Red, ImmunoChemistry Technologies). Cells were visualized live by confocal microscopy.
Cells were fixed in 4% PFA in D-PBS, and then incubated sequentially in 0.25% Tween 20 (Sigma), then 1% BSA with 0.3 M glycine (Sigma), both in D-PBS. Primary and secondary antibodies are listed below. Slides were mounted in Vectashield with DAPI counterstain (Vector Labs). Slides were imaged using a Zeiss LSM 510, laser scanning inverted confocal microscope equipped with the following objectives: 40× Zeiss C-APOCHROMAT UV-Vis-IR water, 1.2NA, 63× Zeiss Plan-APOCHROMAT DIC oil, 1.4NA, and 100× Zeiss Plan-APOCHROMAT DIC oil, 1.46NA. Image analysis was performed using ZEN software (Zeiss). Laser intensity and detector sensitivity settings remained constant for all image acquisitions within a respective experiment. Nuclear outlines were generated using Metamorph software suite (Molecular Devices) using the Kirsch/Prewitt filter to define boundaries and then subtracting out the original binary images.
The following antibodies were used in this study for either Western blotting (WB) or immunoflourescence (IF), or both as indicated, along with their respective source and catalogue number: Primary antibodies: Actin (Sigma A5316, WB), CD63 (Developmental Studies Hybridoma Bank (DSHB) clone H5C6, IF), Fragilis (mouse Ifitm3) (Abcam ab15592, WB, IF), GAPDH (BD Biosciences 610340, WB), HA (Wistar collection, Coriell Institute, clone H18-S210, WC00029, IF), IFITM3 (Abgent AP1153a, WB, IF), IFITM3 (Abgent AP1153c, IF), LAMP1 ((DSHB) clone H4A3, WB, IF), LC3 (Nanotools Mab LC3-5F10, WB, IF), MX1 (Proteintech 13750-1-AP, WB, IF), NP (Millipore clone H16-L10-4R5 MAB8800, IF), RAB7 (Abcam 50533, WB, IF). Secondary antibodies for IF (all from Invitrogen): Alexa Fluor 488 and 647 (goat anti-rabbit and goat anti-mouse). The LAMP1 [H4A3] and CD63 [H5C6] antibodies were developed by J.T. August and J.E.K. Hildreth and were obtained from the DSHB and developed under the auspices of the NICHD and maintained by The University of Iowa, Department of Biology, Iowa City, IA.
These experiments employ the QuantiGene ViewRNA slide-based assay kit from Affymetrix (Cat #QV0096) with all components from that source unless noted. RNA was visualized following a modified manufacturer protocol; changes made include the omission of the ethanol dehydration step, and use of Vectashield mounting media. Post-fixation with 4% PFA, cells adherent on coverslips were incubated with 1× detergent solution or incubated in 0.25% PBS-Tween20. Cells were then incubated with Proteinase K. Next cells were incubated at 40°C in hybridization solution A containing a viewRNA probe set designed against either the negative stranded RNA NP genome (vRNA) of PR8 (Affymetrix VX1-99999-01 QG ViewRNA TYPE 1 Probe Set against NP Influenza A virus (A/PuertoRico/8/34(H1N1)) at 1∶100) or a probe set against the positive stranded NP mRNA. Cells were then incubated in hybridization preamplifiers (1∶100 in hybridization buffer B) at 40°C. Finally cells were incubated with labeled probes (1∶100 in hybridization buffer C), washed and imaged as above. All steps were followed by two D-PBS washes.
Pseudotyped lentiviral particles expressing the HA envelope were produced by plasmid transfection of HEK 293T cells with an HIV-1 genome plasmid derived from pBR43IeG-nef+ (NIH AIDS Research and Reference Reagent Program (Division of AIDS, NIAID, NIH, Cat#11349, from Dr. Frank Kirchhoff) modified with a deletion which abolishes expression of Env without disrupting the Rev-responsive element, pCAGGS-HA WSN/33, pCAGGS-NA WSN/33 and pMM310, which encodes a hybrid protein consisting of β-lactamase fused to the HIV accessory protein, Vpr (NIH AIDS Research and Reference Reagent Program, Division of AIDS, NIAID, NIH (Cat#11444) from Dr. Michael Miller). pCG-VSV-G together with pBR43IeG-nef+ and pMM310 were transfected to produce VSV-G pseudotyped lentiviral particles. For the H5N1, H3N1, and H7N1 pseudoparticles, pCAGGS-HA5 (A/Thailand2(SP-33)/2004) pCAGGS-HA3 (A/Udorn/72), and pCAGGS-HA7 (A/FPV/Rostock/34) expression plasmids were co-transfected with the pCAGGS-NA WSN/33, pMM310, and the pBR43IeG-nef+ lentiviral backbone. Cultures for pseudoparticle fusion assays, including stably transduced MDCK cells and WI-38 fibroblasts, were plated in 24-well dishes with 90,000 cells per well at the beginning of each assay. At the time of assay, 0.5 mL of virus stock was added to cells and incubated for 2–3 h (depending on cell type) at 37°C. In experiments using bafilomycin A1 (Sigma), the inhibitor was added at 0.1 nM final concentration (low dose) at 37°C for 1 h prior to incubation with virus. After infection, viral media was then aspirated and replaced with complete DMEM containing CCF2-AM (Invitrogen) along with 1.7 µg/mL probenecid (Sigma). Cells were incubated in the dark for 1 h, followed by dissociation from the dish using Enzyme Free PBS-based Dissociation Buffer, and fixation in 2% PFA. Flow cytometry was conducted on a Becton Dickinson LSRII using 405 nm excitation from the violet laser, and measuring 450 nm emission in the Pacific Blue channel and 520 nm emission in the Pacific Orange channel. Data was analyzed using FACSDiva and FlowJo8.8.7.
A549 cells stably transduced to overexpress IFITM3 or with empty expression vector (pQCXIP, Clontech) were grown to ∼50% confluency, dissociated with trypsin-free EDTA-based dissociation buffer (Invitrogen) for 10 min. at 37°C. Cells were incubated at 4°C with FITC-conjugated Sambucus nigra lectin (SNA, Vector Labs #FL-1301) to detect (α-2,6) sialic acid linkages, and biotinylated Maackia amurensis lectin II (MAL, Vector Labs #B-1265) to detect (α-2,3) sialic acid linkages, followed by streptavidin-PE-Cy7 (Invitrogen). Cells were incubated with lectins individually and in combination, and the results of staining were indistinguishable. All cells were stained with violet cell-impermeable dye (Invitrogen #L34955), and cells were included in the analysis if viable by FSC/SSC and viability dye.
A549 cells transduced with IFITM3 or the empty vector pQXCIP were detached using Enzyme Free PBS-based Dissociation Buffer, and then washed in cold PBS extensively. Cells and virus (WSN/33) were pre-chilled on ice for 30 min. and mixed at a moi of 50 and incubated at 4°C for 1 h with rotation. Cells were washed extensively with ice cold PBS and then fixed using 4% PFA. The cells were then probed with anti-HA mouse monoclonal antibody (Wistar collection, Coriell Institute, clone H18-S210, WC00029, IF) for 1 h at room temperature, followed by anti-mouse AlexaFlour-488 conjugated antibody (Invitrogen) for 1 h with PBS washes in between, then analyzed by flow cytometry.
|
10.1371/journal.ppat.1002736 | DNase Sda1 Allows Invasive M1T1 Group A Streptococcus to Prevent TLR9-Dependent Recognition | Group A Streptococcus (GAS) has developed a broad arsenal of virulence factors that serve to circumvent host defense mechanisms. The virulence factor DNase Sda1 of the hyperinvasive M1T1 GAS clone degrades DNA-based neutrophil extracellular traps allowing GAS to escape extracellular killing. TLR9 is activated by unmethylated CpG-rich bacterial DNA and enhances innate immune resistance. We hypothesized that Sda1 degradation of bacterial DNA could alter TLR9-mediated recognition of GAS by host innate immune cells. We tested this hypothesis using a dual approach: loss and gain of function of DNase in isogenic GAS strains and presence and absence of TLR9 in the host. Either DNA degradation by Sda1 or host deficiency of TLR9 prevented GAS induced IFN-α and TNF-α secretion from murine macrophages and contributed to bacterial survival. Similarly, in a murine necrotizing fasciitis model, IFN-α and TNF-α levels were significantly decreased in wild type mice infected with GAS expressing Sda1, whereas no such Sda1-dependent effect was seen in a TLR9-deficient background. Thus GAS Sda1 suppressed both the TLR9-mediated innate immune response and macrophage bactericidal activity. Our results demonstrate a novel mechanism of bacterial innate immune evasion based on autodegradation of CpG-rich DNA by a bacterial DNase.
| Group A Streptococcus (GAS) ranks among the top ten human pathogens causing fatal disease. GAS possesses an arsenal of virulence factors that circumvent the primary mammalian defence strategies, the innate immune system. Toll-like receptors (TLRs), allow the host to detect pathogens by recognizing structures or patterns abundant in pathogens but lacking in the mammalian host, including unmethylated CpG-rich bacterial DNA recognized by TLR9. Here we show that GAS DNA but not host DNA triggers TNF-α and interferon type 1 cytokine secretion by monocytic cells, and that this secretion is dependent on the presence of functional TLR9. The highly virulent M1T1 GAS clone expresses the virulence factor DNase Sda1. Sda1-mediated bacterial DNA degradation was shown to prevent TLR9-dependent cytokine release in monocytes, which then fail to effectively phagocytose and kill bacteria. In a mouse necrotizing fasciitis model, the streptococcal DNase Sda1 suppressed TLR9-dependent INF-α and TNF-α induction. Inhibition of TLR9 recognition by a bacterial DNase thus illustrates a novel mechanism of microbial innate immune evasion.
| The Gram-positive bacterium Group A Streptococcus (GAS) is a leading human pathogen, annually causing over 700 million cases of superficial infections such as pharyngitis or pyoderma, and more than 650,000 cases of invasive infections, including the potentially lethal conditions of necrotizing fasciitis (NF) and streptococcal toxic shock syndrome (STSS) [1]. Increased reports of severe GAS disease in recent decades have been in large part attributable to the emergence of a globally disseminated clone of the M1T1 serotype [2]. M1T1 strains are the most common cause of GAS pharyngitis and are strongly overrepresented in severe cases such as NF and STSS [3]. The ability of invasive GAS to produce life-threatening infections even in previously healthy individuals reflects a diverse array of virulence factors that together allow the bacterium to invade host cellular barriers and resist innate immune clearance [2], [4]. One important distinguishing feature of the globally-disseminated M1T1 GAS clone compared to less pathogenic GAS strains is the acquisition of a prophage encoding a potent secreted DNase, Sda1 [5]. Sda1 activity has been shown to promote GAS escape from phagocytic killing with DNA-based neutrophil extracellular traps (NETs) [6], [7], [8], frameworks of DNA containing antimicrobial peptides, histones and proteases that are generated by neutrophils to capture and eliminate bacteria at tissue foci of infection [9].
To control an infection quickly and to prevent disease progression, timely and accurate recognition of bacteria by the host innate immune system is crucial. Pattern recognition receptors (PRRs) such as Toll like receptors (TLRs) recognize conserved molecular patterns from pathogens. The Toll-like receptor 9 (TLR9) is located intracellularly and recognizes unmethylated CpG-rich DNA motifs commonly present in microorganisms but absent in the host genome [10]. Due to its intracellular localization, TLR9 was first appreciated to respond to intracellular pathogens such as Listeria monocytogenes and Legionella pneumophila [11], [12]. However, TLR9 has recently been shown to enhance resistance against the common Gram-positive bacteria Streptococcus pneumoniae [13] and GAS [14]. Following this lead, we hypothesized that the presence of the potent DNase Sda1 in the hyperinvasive M1T1 GAS clone could modify its own unmethylated extracellular CpG-rich DNA fragments and alter TLR9-mediated recognition by host innate immune cells. Combining studies with M1T1 GAS and an isogenic strain with loss of Sda1 expression with macrophages derived from WT and TLR-9 deficient mice, we demonstrate a novel mechanism of bacterial innate immune evasion based on autodegradation of a key pattern-recognition molecule.
DNA was purified from GAS strain 5448, representative of the globally disseminated hyperinvasive M1T1 clone, and incubated with murine bone marrow-derived macrophages (BMDMs). Cytokine release into the medium was used as readout for BMDM activation. GAS DNA induced time-dependent release of interferon type 1 (IFN-1), and specifically interferon-α (IFN-α), from the macrophages (Fig. 1A, B), peaking at 12 h of exposure. GAS DNA also induced BMDM TNF-α secretion, with maximal levels already detected after 6 h of incubation and remaining elevated for at least 24 h (Fig. 1C). Induction of IFN-α and TNF-α release by GAS DNA was also dose-dependent (Fig. 1D). In contrast, human DNA did not induce IFN-1 or TNF-α secretion from BMDMs (Fig. 1A–C). We did not detect specific induction of the cytokines IL-6, IL-1β, IL-10 or MIP-2 from BMDMs exposed to GAS DNA (data not shown).
Unmethylated CpG-rich DNA motifs have previously been reported to induce macrophage secretion of proinflammatory cytokines including TNF-α and IFN-1 [10], [15]. Further studies have shown that macrophages exposed to live GAS, DNA isolated from GAS or antibiotic-killed group B Streptococcus release IFN-β and TNF-α [16], [17]. However, bacterial DNA has not previously been shown to stimulate IFN-α secretion, a finding relevant to innate immune defense since IFN-α is known to provide protection against Gram-positive bacterial infections [18].
The experiments above showed that GAS DNA, containing unmethylated CpG motifs, but not human DNA, induced IFN-α and TNF-α release by BMDM. Since IFN type 1 secretion is partially mediated by TLR9 [19], we tested whether cytokine release in murine BMDMs expressing TLR9 [20], [21], [22], occurred in a TLR9-dependent manner. Chloroquine blocks endosomal acidification and is a known inhibitor of TLR9 [23], [24]. We observed a significant decrease of IFN-α and TNF-α secretion in response to GAS DNA and to the TLR9 agonist ODN2395 in BMDMs pretreated with chloroquine, whereas TLR4-mediated responses to LPS were unaffected (Fig. 2A). Similar results were obtained with the synthetic TLR9 antagonist G-ODN (Fig. 2B). To further corroborate the TLR9 dependency, experiments were repeated with BMDMs extracted from TLR9-deficient mice. Stimulation using the TLR9 agonist ODN2395 induced BMDM secretion of IFN-α and TNF-α only in the presence of a functional TLR9 pathway, whereas responses to LPS were not influenced (Fig. 2C). Similarly, after stimulation with GAS DNA, a significantly lower release of IFN-α and TNF-α was observed from TLR9-deficient compared to WT BMDMs (Fig. 2C). The stimulation of IFN-α secretion from TLR9-deficient BMDMs, albeit at a reduced level, is most likely explained by a ubiquitous interferon response to immunostimmulatory nucleic acids, mediated by cytosolic DNA sensors amongst others [25]. Similarly, recent work shows that IFN-β is secreted after challenge of TLR9-deficient macrophages with live GAS or GAS DNA complexed with RNA [16].
An important characteristic of the hypervirulent globally disseminated M1T1 clone of GAS is the presence of a prophage-encoded secreted DNase, sda1 [5]. Sda1 has been shown to promote M1T1 GAS virulence via degradation of NETs, allowing the bacteria to escape neutrophil killing and the tissue focus of infection, thus facilitating systemic spread of the pathogen [2], [6], [7]. Functional TLR9 is important in defense against GAS infection [14] and the DNA size required for optimal stimulation varies among host cells. Whereas B-cells are stimulated by small DNA fragments [26], macrophages show enhanced uptake and subsequent responses with increasing DNA length [26]. Having observed efficient BMDM activation by crude GAS DNA (above) we hypothesized that degradation by Sda1 could reduce stimulation of macrophage and thus be an additional immune evasion function of Sda1. To test this, we engineered recombinant GAS Sda1 (rSda1) in E. coli. Purification yielded a 45 kD recombinant protein which showed DNase activity at the expected size when analyzed by zymography (Fig. 3A). Recombinant Sda1 degraded GAS DNA in a time and concentration dependent manner (Fig. 3B–C). Recombinant Sda1 at around 4 µg/mL was similarly efficient in degrading DNA as the natively or overexpressed Sda1 in GAS supernatants (Fig. 3C). Degradation of GAS DNA by Sda1 abolished induction of TNF-α and IFN-α in BMDM's (Fig. 3D). DNase Sda1 on its own did not influence cytokine secretion (Fig. S2). Similarly DNase Sda1 treatment of GAS DNA did not affect the residual level of IFN-α and TNF-α induction when TLR9-deficient BMDMs were studied (Fig. 4A). We speculate that the decreased TLR9-dependent cytokine responses to Sda1-treated GAS DNA was mainly due to decreased average DNA size (Fig. 3B), which has also been shown by others to be crucial for cellular uptake of DNA and subsequent TLR9 stimulation [26]. In addition direct elimination of CpG motifs by the efficient enzymatic action of the bacterial DNase [10] could potentially further contribute to the differences observed. Since it has been reported that Sda1 can degrade RNA [27] and recent work shows that IFN-β is secreted by macrophages after challenge of GAS DNA complexed with RNA [16] we investigated the action of Sda1 against RNA, in addition to DNA, wondering if this could be a two-pronged approach to promote GAS infection. RNA co-incubated with our rSda1 showed no degradation when visualized by agarose gel electrophoresis (Fig. S1), suggesting that Sda1 possesses negligible or minor RNA-degrading activity under our assay conditions. To study the influence of Sda1 on TLR9-mediated macrophage responses to live GAS infection, BMDMs were challenged with the wild type M1TI GAS parent strain M1 5448 (M1WT), the isogenic GAS DNase sda1 knockout (M1Δsda1) and the sda1 complemented strain (M1Δsda1pDcsda1) using the pDcsda1 plasmid [6]. A significant increase in IFN-α and TNF-α secretion was observed from WT BMDMs challenged with the M1Δsda1 mutant strain compared to the parent and complemented strains (Fig. 4B). The observed Sda1-dependent reduction of cytokine responses to GAS was diminished in TLR9-deficient macrophages (Fig. 4B). Similarly, heterologous expression of sda1 in a less virulent M49 GAS strain diminished BMDM IFN-α and TNF-α secretion in a TLR9-dependent manner (Fig. 4C). Our paired loss- and gain-of-function analyses indicate that Sda1 is both necessary and sufficient to promote GAS avoidance of TLR9-dependent macrophage recognition [10].
TLR9 is activated by CpG-rich DNA motifs present in most bacteria. Evolution of reduced CpG content (CpG suppression) has been described in other microorganisms including nonpathogenic viruses, Plasmodium falciparum and Entamoeba histolytica [28]. In contrast GAS DNA possesses a high CpG content. By acquiring a potent secreted DNA-degrading enzyme, GAS has come across an alternative means to circumvent TLR9 activation in the host innate immune response.
To date, DNase Sda1 has been appreciated to promote M1T1 GAS resistance to neutrophil extracellular killing due to its capacity to digest NETs [6], [7], [8]. Since we here identified a capacity of Sda1 to diminish TLR9-mediated macrophage responses, we hypothesized that the DNase activity could blunt the innate immune killing capacity of macrophages to kill GAS. Mice depleted of macrophages or treated with inhibitors of macrophage phagocytosis cannot clear GAS infections even at relatively low challenge doses [29], demonstrating the essential first line defense function of these immune cells against the pathogen. WT and TLR9-deficient BMDMs were challenged with live M1 or M49 GAS either expressing or not DNase and total, intra and extracellular bacterial killing was quantified. GAS strains expressing Sda1 survived significantly better in both the total and intracellular killing assays compared to the strains in which Sda1 was not expressed (Fig. 5A, B and S3). The Sda1-mediated survival advantages for GAS were much more pronounced in WT compared to TLR9-deficient BMDMs.
Sda1-mediated resistance to total macrophage killing could be caused by interference with DNA-based extracellular traps, which, we have recently observed, are generated by macrophages [30] though to a much lesser extent than observed in neutrophils or mast cells exposed to GAS. To investigate if the bacterial DNase Sda1 interferes with extracellular killing by macrophages we pretreated the macrophages with cytochalasin D to inhibit phagocytosis. No difference between GAS WT and the isogenic GAS DNase sda1 knockout strains were observed for the extracellular killing indicating that the Sda1-dependent survival advantage seen in vitro is indeed mainly intracellular (Fig. S4). To further explore the Sda1-dependent survival advantage of GAS intracellularly following phagocytotic uptake into the macrophages, we measured oxidative burst activity, which has been reported to be a TLR9-induced mediator of intracellular killing in murine macrophages [14]. Oxidative burst activity was measured in WT and TLR9-deficient BMDMs after infection with the isogenic pairs of GAS strains either expressing or lacking DNase Sda1. BMDMs infected with GAS strains possessing Sda1 displayed a significantly reduced oxidative burst response compared to BMDMs infected with GAS strains lacking Sda1 (Fig. 5C). We propose that reduced oxidative burst is an additional mechanism by which Sda1 can contribute to M1T1 GAS resistance to macrophage killing. In order to test if blocking IFN-α and TNF-α can prevent phagocytic killing mediated by GASΔsda1 we repeated the BMDM killing assays with WT BMDM challenged with WT GAS M1 and GASΔsda1 bacteria after having pre-incubated the BMDM with either the neutralizing antibodies against TNF-α or IFN-α or their respective controls. Addition of neutralizing IFN-α antibodies increased the survival of GAS and GASΔsda1 when challenged with BMDMs from WT mice. The effect of neutralizing TNF-α antibodies was smaller and not statistically significant upon challenge with WT GAS and GASΔsda1 mutant bacteria (Fig. S7). These results suggest that certain cytokines may themselves contribute to enhance the phagocytic killing of bacteria.
Viability of the BMDMs was >90% after 4 and 12 h of stimulation with the bacteria at MOI of 1. No significant differences were observed in survival of WT or TLR9 deficient BMDMs stimulated with either GAS WT or the GASΔsda1 mutant (Fig. S6).
All GAS strains are known to express DNase activity, and some strains produce up to 4 different DNases. As first described by Wannemaker [27], [31] these proteins were designated DNase A, B, C and D. However the role of theses DNases remained unclear until 50 years later when it was shown that GAS DNase activity, particularly that of GAS DNase D (now known as Sda1) was important for virulence [6], [8]. By creating knockout mutants of the three DNases present in the GAS MGAS5005 strain [8], Sumby et al. determined that Sda1 was the most active. Sda1 very efficiently degraded DNA in vitro. In murine skin infection models, engineered GAS strains expressing Sda1 alone were found to be as virulent as wild-type GAS, supporting the conclusion that Sda1 but no other DNases mediate virulence in vivo. The strong activity of Sda1 compared to the weaker DNA-degrading activity of the other DNases found in GAS, may help to explain the pronounced phenotype we have observed in enhancing TLR9-mediated clearance when only Sda1, and no other DNases, is knocked out. However our data may underestimate the full collective potential of GAS DNases in TLR9-mediated innate immune evasion.
To provide in vivo corroboration of the ex vivo experiments carried out using BMDMs, we examined IFN-α and TNF-α levels in skin homogenates of mice infected subcutaneously with GAS. Despite the important contribution of Sda1 to GAS proliferation and necrotic ulcer development in this model [6], [7], WT mice infected with the GAS M1Δsda1 mutant showed higher levels of IFN-α and TNF-α in the skin samples than mice infected with the WT parent GAS expressing Sda1 (Fig. 6A, B). Parallel experiments performed in TLR9-deficient mice showed much lower cytokine levels in the infected skin compared to WT mice, again underlining the importance of TLR9 in mediating cytokine responses. However, in contrast to the WT mice, the presence or absence of Sda1 did not affect the level of cytokines produced in response to GAS in the TLR9-deficient mice (Fig. 6A, B). We had shown previously [14] that more bacteria are found in the skin of TLR9-deficient mice compared to WT mice, and that more surviving GASM1 WT bacteria compared to GASM1Δsda1 are present following experimental challenge of WT mice [6]. We also found more WT than ΔSda1 mutant bacteria present following injection into TLR9- deficient mice (Fig. S8). The observation that the TLR9-deficient mice injected with the GASM1Δsda1 mutant demonstrated similar bacterial counts compared to WT mice could be due to a large initial influx of neutrophils efficiently clearing the DNase-deficient mutant strain within extracellular traps. The increased cytokine levels detected in WT mice injected with GASM1Δsda1 mutant compared to WT bacteria are not explained by differences in bacterial counts, nor do bacterial levels account for increased cytokine levels in WT mice compared to TLR9- deficient mice. In sum, we documented that increased tissue expression of IFN-α and TNF-α in the mouse necrotizing skin infection model occurred in both a DNase- and TLR9-dependent manner.
In contrast to our ex vivo data and tissue culture experiments carried out by others [25], IFN-α and TNF-α secretion in our in vivo experiments was strictly TLR9-dependent. This finding suggests that TLR9 recognition of GAS is of true physiological relevance, and the described TLR9-independent pathways elicited ex vivo may be of diminished importance in the in vivo setting. Our experiments emphasize the critical nature of innate immune recognition at tissue foci of infection. It is important to note that while much evidence exists that IFN-α is beneficial to innate immune cells in combating bacterial infection [18], if IFN-α is produced systemically at high levels or in an uncontrolled fashion and deleterious effects on antibacterial clearance may be observed [32].
In our work we have focused on the cytokine responses and killing activities of isolated macrophages vs. GAS and the expression of cytokines at the site of GAS subcutaneous infection. We hypothesize that GAS Sda1 contributes to disease in multiple ways, including interfering with TLR9 recognition (blunting the initial innate immune response) and degrading extracellular traps (promoting phagocyte evasion). Together, these Sda1-dependent virulence phenotypes increase the risk of bacterial proliferation to produce severe necrotizing infections, septicemia or toxic shock syndrome, where ultimately high cytokine levels develop in response to the uncontrolled infection [33]. Additionally, variation in cytokine responses to GAS superantigens influences the severity of streptococcal toxic shock syndrome [34].
To summarize, we here describe a novel mechanism by which a bacterial pathogen can directly elude TLR9 recognition. The GAS virulence factor DNase Sda1 leads to decreased production of the proinflammatory cytokines IFN-α and TNF-α and to decreased killing efficiency of macrophages, which are key contributors for innate immunity to GAS infection [29]. DNase production is now recognized to be a virulence factor of a number of bacterial pathogens including Streptococcus pneumoniae [35], Staphylococcus aureus [36] and Pseudomonas aeruginosa [37] and it could be fruitful to determine whether evasion of TLR9-based detection complements NET degradation in the infectious pathogenesis of these species. Previously, inhibition of Sda1 activity by G-actin boosted neutrophil extracellular killing of the WT GAS bacteria and reduced lesion size in the necrotizing skin infection model, providing proof-of-principle that this DNase can represent a pharmacological target for virulence factor neutralization [6]. Our current data, demonstrating that loss of Sda1 enhances both TLR9-mediated innate immune responses and macrophage bacterial killing, provides additional rationale toward such a therapeutic strategy.
C57BL/6 wild-type (WT) and C57BL/6 TLR9-deficient mice were bred and handled 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 California, San Diego (Animal Welfare Assurance Number: A3033-01). All efforts were made to minimize suffering of animals employed in this study.
C57BL/6 TLR9-deficient mice were originally developed by Dr. Shizuo Akira (Osaka University, Japan); WT mice were purchased from Charles River Laboratories. The GAS strain 5448, a well-characterized M1T1 clinical isolate from a patient with necrotizing fasciitis and toxic shock syndrome [38] and the GAS M49 strain NZ131 [39] were used. In addition, to analyze gain and loss of function of Sda1, GAS strains expressing Sda1 (GASM1T1 WT, GASM1T1 Δsda1pDcsda1 and GASM49 pDcsda1) and lacking Sda1 (GASM1T1 Δsda1 and GASM49 WT pDcerm) were used in the assays as described previously [6], [40] Growth curves showed that all strains used in the experiments grew identically. GAS strains were propagated in Todd-Hewitt broth (THB) (Difco, BD Diagnostics) or Todd-Hewitt agar plates. For use in macrophages and mouse challenge studies, bacteria were grown to logarithmic phase in THB (OD600 = 0.4 corresponding to ∼2×108 cfu/ml), pelleted, washed and resuspended in PBS or tissue culture media at the desired concentration.
GAS genomic DNA was prepared using Bactozol kit (Molecular Research Center Inc.) with minor modifications. GAS strains were incubated overnight in THB medium, 2 ml of the overnight culture pelleted and resuspended in Bactozol buffer+50 U mutanolysin and 10 U Proteinase K, then incubated at 45°C for 90 min. Bactozol enzyme was added and the preparation incubated for 60 min at 45°C. From this step on, the standard Bactozol kit protocol was followed. Human genomic DNA was prepared from buffy coats of healthy volunteers from the blood donation center in Zurich, Switzerland. Cells were pelleted by centrifugation at 1000 g for 10 min. Genomic DNA was then extracted using blood and tissue DNA easy kit (Qiagen) following the manufacturer's protocol. Remaining RNA was digested by adding RNase during DNA purification. Purity of the isolated DNA, bacterial or human, was confirmed by lack of any cytokine stimulation after digesting the DNA with DNase I (Roche) (Fig. S5). Genomic DNA integrity was confirmed by agarose gel.
The sda1 gene was amplified from GAS M1T1 genomic DNA using primers forward 5′-TCGAGCTCTCTAAACATTGGAGACATCTAATTATTCACTCTG-3′ and reverse 5′-TGGTCGACTTATTCTATATTTTCTTGAGTTGAATGATG-3′. The PCR product was subcloned into vector pQE30 and the newly created plasmid transformed into the E. coli strain M15-pREP4 (Qiagen) for protein production. Bacteria were grown to OD600 = 0.5 at 30°C in the presence of 100 µg/ml ampicillin and 25 µg/ml kanamycin and the expression of Sda1 induced for 1 h via addition of 1 M IPTG. Bacteria were then harvested, resuspended in 30 ml lysis buffer (50 mM NaH2PO4, 300 mM NaCl, 20 mM imidazole pH 8) and lysed by sonication at full power (20 cycles of 15 seconds each). Cell debris were spun down at 12,000 g for 30 min, the supernatant filtered through a 0.45 µm PVDF filter, then run on a HiTrap nickel bead column (GE Healthcare) at a 1 ml/min speed. The column was then washed with 10 volumes lysis buffer and with 10 volumes wash buffer (50 mM NaH2PO4, 300 mM NaCl, 50 mM imidazole pH 8). Sda1 was eluted with 5 ml elution buffer (50 mM NaH2PO4, 300 mM NaCl, 250 mM imidazole pH 8), dialysed against storage buffer (50% glycerol in PBS) and stored at −20°C.
The quality and activity of the purified recombinant Sda1 were checked by performing a zymogram. Briefly, 1 µg of Sda1 and 1 µg of DNaseI, used as a control for activity, were loaded on a 12% polyacrylamide gel containing 10 µg/ml of calf thymus DNA. The gel was washed 2 times in ddH2O and then incubated overnight in 50 mM TRIS·HCl pH 7.4. The gel was then incubated (35 hours, 37°C) in Sda1 reaction buffer (50 mM TRIS·HCl pH 7.4, 5 mM CaCl2) containing 1 µg/ml EtBr. DNA degradation showed on gel as a dark band and was taken as a proof of nuclease activity.
Genomic DNA (2.5 µg) was incubated with DNase I from bovine pancreas (Roche) at 37°C for 6 hours. Furthermore, 2.5 µg of genomic DNA were incubated with purified recombinant Sda1 at 37°C for 0 to 6 hours. After 1, 2, 4 and 6 hours, 1 M EDTA was added to stop the reaction and the samples were loaded on an agarose gel or added to the BMDMs. DNA alone and DNase buffer without adding recombinant Sda1 served as controls. Degradation of genomic DNA was confirmed by agarose gel electrophoresis.
The GAS DNase Sda1 activity was tested as previously described with minor modifications [6]. 5 µl of a 1∶100 dilution of filtered supernatants taken at OD600 0.4 were mixed with 3 µl reaction buffer (50 mM Tris-HCl, 5 mM CaCl), 20 µl of water and 2 µl of GAS genomic DNA (125 ng/µl) and incubated for five minutes at 37°C. The reaction was stopped by addition of 1 M EDTA. The DNA was run on 1% agarose gel for visualisation.
Murine bone marrow-derived macrophages (BMDMs) were isolated as previously described [41] with some modifications. Bone marrow cells were collected from mice legs and cultured for 7 days in Dulbecco's modified Eagle's medium (high glucose) supplemented with 30% L-929 cell conditioned medium. The adherent cells (BMDM) were then collected, split to assay settings, and cultured in Dulbecco's modified Eagle's medium (high glucose) until being used for experiments on day 10.
BMDM (5×105) were seeded into each well of a 48 well plate in 500 µl medium. On day 10, 2 h before the inoculation of genomic DNA or bacteria, BMDM were washed twice with PBS, and 200 µl DMEM+10% FBS (70°C heat inactivated) were added to each well. GAS strains and genomic DNA, prepared as described above, were inoculated into wells at a multiplicity of infection (MOI) of 1 and 5 µg/ml respectively. In addition 80 µl of the DNA samples obtained after digestion with rSda1 for 0, 1, 2, 4 and 6 hours were used. Media without addition of DNA or rSda1 as well as media containing rSda1 alone served as controls. Plates inoculated with bacteria were centrifuged at 800 g for 10 min, incubated at 37°C in a CO2 incubator for 2 h, and penicillin G and gentamicin added to each well to a concentration of 10 and 100 µg/ml, respectively. As TLR9 specific agonist and antagonist, 5 µg/ml CpG-ODN 2395 (Microsynth, 5′-tcg tcg ttt tcg gcg gcg ccg-3′ with phosphorothioate on all bases) and G-ODN (Microsynth, 5′- ctc cta ttg ggg gtt tcc tat -3′ with phosphorothioate on all bases) were used. Challenged macrophages were incubated at 37°C with 5% CO2 for 12 h. The plates were centrifuged at 800 g for 10 min before the supernatants were taken and stored in −80°C until they were used in ELISA assays.
IFN-α and TNF-α ELISA: Levels of IFN-α in culture supernatants were analyzed by a standard sandwich ELISA using a monoclonal mouse IFN-α capture antibody (Hycult biotech) and polyclonal rabbit IFN-α antibody (PBL interferon source) together with a goat anti-rabbit antibody conjugated with HRP (Invitrogen). A serial dilution of recombinant mouse IFN-α (PBL interferon source) was used to calculate the absolute concentration in the supernatants. Levels of TNF-α were measured using mouse TNF-α ELISA kit (R&D) following the manufacturer's protocol.
Interferon type 1 cell luciferase assay: Luciferase cell reporter assay of IFN-1 was carried out using the LL171 luciferase reporter cell line as described [42].
Macrophages were harvested and seeded in 48 well plates as described above. Two hours before adding bacteria, macrophages were washed twice with PBS and 400 µl of DMEM+2% FBS were added to each well. Logarithmic phase bacteria were added to the wells at final MOI of 1 and plates were centrifuged for 5 minutes at 1500 rpm. For total killing, the plate was incubated for 4 hours. For intracellular killing assays, 100 µg/ml penicillin G and 100 µg/ml gentamicin were added to the wells and the plate was incubated for another 2 h at 37°C in 5% CO2 before macrophages were detached with trypsin and lysed with 0.025% Triton-X100. Serial dilutions of the lysates were plated on THA for enumeration of surviving bacterial colony forming units (cfu). Reactive oxygen species were measured following the protocol described before [14].
An established murine model of necrotizing skin infection was used [43]. Briefly, logarithmic phase GAS were resuspended in PBS, mixed 1∶1 with sterile Cytodex beads (Sigma) and an inoculum of 5×107 cfu of GAS was injected subcutaneously into one flank of 10–12 week old WT or TLR9-deficient mice. At day four the mice were euthanized and skin from the lesion was collected, homogenized and IFN-α and TNF-α measured by ELISA and bacterial counts assessed after serial dilutions and plating on THA plates.
Data were analysed and edited using the SPSS (SPSS 11.5 Inc., Chicago, Illinois, USA), the NCSS (Kaysville, Utah, USA) and Graphpad prism 5 software (Graphpad Software Inc, La Jolla, California, USA) packages. Two-sample two-tailed homoscedastic t-tests were used to calculate the indicated p-values except for the animal studies (Fig. 6) for which analysis of variance (ANOVA) followed by Bonferroni comparison and a factorial analysis (2-way ANOVA) were used to calculate indicated p-values.
|
10.1371/journal.pgen.1007781 | RNase H1 directs origin-specific initiation of DNA replication in human mitochondria | Human mitochondrial DNA (mtDNA) replication is first initiated at the origin of H-strand replication. The initiation depends on RNA primers generated by transcription from an upstream promoter (LSP). Here we reconstitute this process in vitro using purified transcription and replication factors. The majority of all transcription events from LSP are prematurely terminated after ~120 nucleotides, forming stable R-loops. These nascent R-loops cannot directly prime mtDNA synthesis, but must first be processed by RNase H1 to generate 3′-ends that can be used by DNA polymerase γ to initiate DNA synthesis. Our findings are consistent with recent studies of a knockout mouse model, which demonstrated that RNase H1 is required for R-loop processing and mtDNA maintenance in vivo. Both R-loop formation and DNA replication initiation are stimulated by the mitochondrial single-stranded DNA binding protein. In an RNase H1 deficient patient cell line, the precise initiation of mtDNA replication is lost and DNA synthesis is initiated from multiple sites throughout the mitochondrial control region. In combination with previously published in vivo data, the findings presented here suggest a model, in which R-loop processing by RNase H1 directs origin-specific initiation of DNA replication in human mitochondria.
| Human mitochondria contain a double-stranded DNA genome that codes for key components of the oxidative phosphorylation system. The mitochondrial DNA (mtDNA) is replicated by a replication machinery distinct from that operating in the nucleus and mutations affecting individual replication factors have been associated with an array of rare, human diseases. In the present work, we demonstrate that RNase H1 directs origin-specific initiation of DNA replication in human mitochondria and that disease-causing mutations may impair this process. A unique feature of mtDNA replication is that primers required for initiation of leading-strand DNA replication are produced by the mitochondrial transcription machinery. A substantial fraction of all transcription events is prematurely terminated about 120 nucleotides downstream of the promoter and the RNA remains firmly associated with the genome, forming R-loops. Interestingly, the free 3′-end of these R-loops cannot directly prime initiation of DNA synthesis, but must first be processed by RNase H1. The process is stimulated by the mitochondrial single-stranded DNA binding protein and faithfully reconstitutes replication events mapped in vivo. In combination with mapping of replication events in fibroblasts derived from patients with mutations in RNASEH1, our findings point to a possible model for replication initiation in human mitochondria similar to that previously described in the E. coli plasmid, ColE1.
| Mitochondrial DNA (mtDNA) is a 16.6 kb circular, double-stranded DNA (dsDNA) molecule that contains genes for 13 components of the oxidative phosphorylation (OXPHOS) system as well as the 22 tRNAs and 2 rRNAs required for their translation. The two strands can be separated by CsCl gradient density centrifugation and are accordingly referred to as the heavy (H) and light (L) strands. Polycistronic transcription of the two strands is initiated from the L-strand and H-strand promoters (LSP and HSP, respectively) and carried out by a transcription machinery consisting of a single subunit RNA polymerase (POLRMT), and transcription factors A (TFAM) and B2 (TFB2M) [1, 2]. Transcription is further stimulated by the mitochondrial transcription elongation factor (TEFM), which forms a stable ternary complex with the elongating POLRMT and template DNA [3–6].
Mitochondrial DNA is synthesized by the trimeric DNA polymerase γ (POLγ) that consists of one catalytic subunit (POLγA) and two identical accessory subunits (POLγB), which act to increase processivity [7–10]. The replicative DNA helicase TWINKLE is a hexameric protein and its activity is stimulated by the mitochondrial single-stranded DNA binding protein (mtSSB) [11, 12]. According to the strand displacement model for mtDNA replication, mtDNA synthesis is initiated from two separate origins of replication, OriH and OriL, one for each strand. DNA synthesis first commences at OriH and proceeds in one direction to produce a nascent H-strand [1, 13]. When the replication machinery has synthesized approximately two thirds of the H-strand, it passes OriL, which is exposed in its single-stranded conformation and activated. OriL adopts a stem-loop structure and POLRMT initiates synthesis of short RNA primers from a poly-dT stretch in the loop region. These primers are used by POLγ to initiate L-strand synthesis with the parental H-strand as template [14, 15]. H-strand and L-strand DNA replication subsequently proceed until two complete daughter molecules have been formed and separated in a Topoisomerase 3α-dependent process [16].
In the literature, there is some confusion with regard to the exact localization of OriH [1]. It is often defined as a single position, 191 in human mtDNA, since prominent free 5′-ends on DNA have been identified at this site [17]. It should however be noted that RNA to DNA transitions have never been identified at this position. How these 5′-ends are actually formed is not understood, but they are potentially the result of extensive primer processing far beyond the location of the actual RNA to DNA transition sites. The processing involves MGME1 and disease-causing mutations affecting this enzyme lead to accumulation of replication intermediates with incomplete processing of 5′-ends [18].
According to the current model for replication initiation, transcription initiated at LSP provides RNA primers for initiation at OriH [19–21]. In support of this notion, loss of POLRMT in a knockout mouse model abolishes primer synthesis in vivo [22]. The switch between primer formation and full-length transcription takes place in a region immediately downstream of LSP containing three evolutionary conserved sequence blocks, CSBI-III, and RNA to DNA transitions in newly synthesized H-strands have been mapped to multiple sites surrounding these elements [19–21]. Newly transcribed RNA remains associated with the CSB-region, forming R-loops that are resistant to RNase A and RNase T1 treatment [23–25]. The unique stability of these R-loops is explained by a G-quadruplex structure, which is formed co-transcriptionally between nascent RNA and the non-template DNA strand at CSBII. This G-quadruplex structure also stimulates premature transcription termination downstream of CSBII [26–28]. The 3′-ends of the terminated transcripts roughly overlap with RNA to DNA transitions mapped in the CSBII-region [26, 29], which has led to the hypothesis that sequence-dependent transcription termination may be responsible for primer formation. There is however no published experimental evidence demonstrating that transcripts prematurely terminated at CSBII can be directly used to prime DNA synthesis. In addition, a molecular understanding of the primer formation and replication initiation reaction is still missing.
Recently, a knockout mouse model demonstrated that RNase H1 is required for R-loop degradation in vivo. In addition, depletion of RNase H1 caused a reduction in mtDNA levels, suggesting that the enzyme is required for mtDNA replication [30]. These in vivo findings raise the possibility that RNase H1 may play a role in R-loop processing and primer formation. RNase H1 is an RNase H enzyme capable of cleaving RNA-DNA hybrids. It can cleave hybrids that are down to approximately 6 nucleotides in length [31]. The enzyme can also cleave Okazaki fragment-like structures, leaving approximately two ribonucleotides next to the RNA-DNA junction [32]. In addition to the potential role in R-loop processing, RNase H1 has also been proposed to be involved in mitochondrial pre-rRNA processing by interacting with the mitochondrial protein P32, which slightly enhances the RNase H1 enzymatic activity [33].
Initiation of mitochondrial DNA replication has been suggested to resemble replication of the E. coli plasmid ColE1. In ColE1 replication, a transcript denoted RNAII associates with the template strand, forming an R-loop that serves as a primer for DNA synthesis [34, 35]. The ColE1 origin of replication is situated downstream of a guanine-rich stretch that is essential for both replication initiation and R-loop formation [35, 36]. For proper initiation of replication, RNAII has to be cleaved by E. coli RNase H, thereby creating a primer 3′-end that can be used by the replication machinery. The similarity between OriH of mtDNA and the ColE1 origin [27, 28], together with the recent in vivo findings of RNase H1 involvement in mtDNA synthesis, suggests that RNase H1 could be involved in primer processing in human mitochondria. To address this intriguing possibility, we here set out to reconstitute initiation of mtDNA replication in vitro.
In vivo analysis has suggested that RNase H1 is required to process R-loops in vivo [30]. We decided to investigate this process in vitro and therefore set out to reconstitute R-loop formation. For our work, we used purified human mitochondrial transcription proteins (Fig 1A, lanes 1-4). As templates, we employed supercoiled or relaxed plasmids containing the LSP promoter and the downstream CSBI-III-region (pUC-LSP, S1 Table). On a relaxed template, a fraction of all transcription events initiated at LSP was prematurely terminated at CSBII (Fig 1B, lane 1) [26, 29]. This effect was stronger on a negatively supercoiled template, where nearly all transcription events were prematurely terminated at CSBII (Fig 1B, compare lanes 1 and 5). In agreement with previous reports, premature transcription termination was reduced by addition of the transcription elongation factor TEFM (Fig 1B, lane 7) [4, 5]. To detect if the pre-terminated transcripts remained as stable R-loops, we used an RNase A protection assay (Fig 1C). RNase A digests free RNA, whereas RNA associated with DNA is resistant to this nuclease at high salt concentrations. No R-loops were formed with relaxed LSP template (Fig 1B, lanes 2 and 4). In the reaction with the supercoiled template however, we observed RNase A-resistant products at sizes just below the transcript terminated at CSBII (Fig 1B, lane 6). This observation suggested to us that long R-loops were formed, possibly encompassing nearly the entire region from LSP to CSBII. Interestingly, R-loops were not detected when TEFM was added to the transcription reaction (Fig 1B, compare lanes 6 and 8).
In other systems, single-stranded DNA binding proteins can promote R-loop formation, probably by binding to the displaced DNA strand [37–39]. We therefore monitored the effects of mtSSB (Fig 1A, lane 8) on R-loop formation in vitro. Addition of mtSSB did not affect the overall transcription patterns on a negatively supercoiled LSP template (Fig 1D, compare lane 1 to lanes 3 and 5), but the R-loops formed in the presence of mtSSB appeared less processed and more uniform in size (Fig 1D, compare lane 2 to lanes 4 and 6). When 120 nM mtSSB was added nearly 70% of all pre-terminated transcripts end up as R-loops (Fig 1D, lanes 5-6), as compared to 50% in the absence of mtSSB (Fig 1D, lanes 1-2). To ensure that our results were not due to mtSSB binding directly to single-stranded RNA and protecting it from RNase A degradation, we also monitored the effects of mtSSB on transcription from a linearized HSP plasmid (pUC-HSP, S1 Table); a template that does not contain a downstream R-loop-forming region. The presence of mtSSB did not protect transcripts formed by HSP transcription from RNase A degradation (Fig 1D, lanes 7-10).
Finally, we investigated if RNase H1 (Fig 1A, lane 9) could process the R-loops. We found that RNase H1 gradually degraded R-loops in a concentration-dependent manner, generating shorter RNA species. (Fig 1E, compare lane 1 with lanes 2-7). The shorter RNA species may be involved in hybrid G-quadruplex formation, and therefore partially resistant to RNase H1 degradation [27]. The observed degradation of R-loops in vitro was in agreement with in vivo data indicating that RNase H1 can process mitochondrial R-loops [30].
It has been suggested that transcripts prematurely terminated at CSBII may be used to prime initiation of mtDNA replication [26]. We therefore decided to analyze if POLγ (Fig 1A lanes 5-7) can use the R-loops formed in vitro as primers for initiation of DNA synthesis. We first investigated if POLγ can use random RNA primers to initiate DNA synthesis on dsDNA. To this end, we utilized POLRMT’s ability to initiate transcription and produce short RNA molecules on negatively supercoiled dsDNA even in the absence of a promoter sequence [40]. In our experiments, we used an exonuclease deficient version of POLγ (exo-), since this protein has strand displacement activity and can use dsDNA as a template even in the absence of a DNA helicase [41]. To monitor DNA synthesis we performed the reactions in the presence of [32P]dTTP (Fig 2A). When both POLRMT and POLγ were added simultaneously to a negatively supercoiled dsDNA template (without LSP), we observed formation of [32P]dTTP-labeled DNA products ranging in size from 50 to 700 nts (Fig 2B, lane 3). We also did the experiment with a template containing LSP but did not find any major effect of its presence (Fig 2B, lane 6). When we repeated the experiment in the presence of increasing amounts of purified mtSSB, replication products disappeared (Fig 2C, lanes 2-4 and 6-8). POLγ can thus initiate DNA synthesis from random primers generated by POLRMT and mtSSB represses this activity. The repressive effect of mtSSB is probably due to its ability to prevent transcription initiation by POLRMT on ssDNA [15, 42].
We next monitored the effects of RNase H1 on replication. Using the LSP template, in the absence of mtSSB, we again observed abundant replication products ranging in size from 50 to 700 nts and we observed a mild stimulatory effect at high levels of RNase H1 (Fig 2D, compare lanes 1 and 4). Again, these non-specific products were reduced in the presence of low levels of mtSSB (20 nM, Fig 2D, lane 5) and abolished at high concentrations of mtSSB (120 nM, Fig 2D, lane 9). Interestingly, addition of increasing levels of RNase H1 in the presence of mtSSB caused the formation of new group of DNA products, ranging in size between 25 and 100 nts (Fig 2D, lanes 11-12). We hypothesized that these products could be DNA replication events initiated from processed R-loops formed by transcription from LSP. To address this possibility, we repeated the experiments with a template lacking LSP. On this template, RNase H1 had no apparent effects on DNA synthesis and no short DNA products were observed when RNase H1 was added together with mtSSB (Fig 2E, lanes 1-4). To further demonstrate that priming was dependent on R-loops, we performed our experiments with the LSP-containing template in the presence of TEFM, which at high concentrations prevents R-loop formation. Increasing amounts of TEFM reduced the products in the size range between 25 and 100 nts, thus suggesting that these shorter replication products were indeed dependent on R-loop formation downstream of LSP (Fig 2F, lanes 1-4). In the presence of higher concentrations of TEFM, a high molecular weight species started to accumulate (Fig 2F, lane 4), which may be due to some longer RNA molecules being used as non-specific primers for replication. Interestingly, even if TEFM reduces replication initiation, it does not completely abolish the reaction. Shorter replication products were observed even when TEFM was present in excess (Fig 2F, lane 4).
To further verify that the 25-100 nts replication products we had observed indeed originated from an LSP R-loop primer, we utilized the fact that the LSP R-loop region only contains a few guanines in the template strand. By adding ddCTP to our reactions we could therefore generate short specific length replication products. Before gel analysis, the replication products were treated with potassium hydroxide (KOH) to remove any RNA residues [43, 44] (Fig 3A). As expected, we could not observe any DNA synthesis in the absence of RNase H1 (Fig 3B, lane 1), again demonstrating that the unprocessed R-loops do not function as primers. Upon addition of RNase H1, we observed replication products with sizes between 12 and 21 nts (Fig 3B, lanes 2-8). The highest levels of DNA synthesis were observed at 2 nM of RNase H1 (Fig 3B, lane 5). Furthermore, we found that mtSSB had a strong stimulatory effect on the reactions (Fig 3C, compare lane 1 to lanes 2-7). By introducing mutations that changed the length of C-less stretches downstream of LSP, we could verify that replication products observed after incubation with ddCTP was due to initiation near CSBII and CSBIII (S1A–S1C Fig). Combined, our experiments therefore support the hypothesis that RNase H1 is required to process R-loops for primer formation and that mtSSB stimulates the process. For all subsequent experiments, we used an mtSSB concentration of 40 nM together with 2 nM of RNase H1 (Fig 3C, lane 6).
We next examined the effects of differing template topology and CSB region mutations on replication initiation. Relaxation of the DNA template, which impairs R-loop formation, almost completely abolished replication initiation (Fig 3D, compare lanes 1 and 2). When CSBIII was mutated, the levels of most replication products shorter than 20 nts were slightly decreased, whereas longer products were left unaffected (Fig 3D, compare lanes 1 and 3). When CSBII was mutated all replication initiation events disappeared (Fig 3D, compare lanes 1 and 4). These results confirm that the LSP R-loops are essential for replication initiation and demonstrate that whereas the CSBII element is crucial for successful primer formation, CSBIII only has a minor effect on initiation (Fig 3D, compare lanes 3 and 4).
Next, we examined the effects of TEFM on initiation of DNA synthesis (Fig 3E). At lower concentrations, TEFM did not affect replication initiation. Interestingly, robust levels of initiation were observed even at equimolar concentrations of POLRMT and TEFM (Fig 3E, lane 4). Not even at very high TEFM concentrations (molar ratio 4:1 relative POLRMT) could we observe a complete inhibition of replication initiation (Fig 3E, lane 6).
Our experiments so far had been performed with exonuclease deficient POLγ. We now repeated the experiments with WT POLγ (Fig 1A, lane 5). As previously demonstrated, initiation of DNA synthesis required the presence of POLγ, POLRMT, and RNase H1 (Fig 3F). Furthermore, the reaction was stimulated by the addition of mtSSB (Fig 3F, compare lanes 4 and 5). The POLγ WT enzyme produced the same products as seen with POLγ exo- (Fig 3F, compare lanes 5 and 6), although the levels of replication products were lower, probably due to the weaker strand displacement activity of the WT enzyme.
Based on our in vitro observations, we wanted to specifically analyze the effects of reduced RNase H1 activity on mtDNA replication initiation in mammalian cells. To this end, we decided to study the effects of disease causing mutations in RNASEH1, which are associated with adult-onset mitochondrial encephalomyopathy [45]. First, we expressed two disease-causing mutant forms of RNase H1, RNase H1:V142I and RNase H1:A185V (Fig 4A) as recombinant proteins and analyzed their effects on R-loop processing. Both RNase H1:V142I and RNase H1:A185V, alone or in combination (Fig 4B, lanes 5-13), displayed impaired R-loop processing activity compared to WT RNase H1 (Fig 4B, lanes 2-4). As a consequence, the mutant RNase H1 proteins could not support origin-specific initiation of DNA replication in vitro (Fig 4C and S2 Fig).
Second, we used fibroblasts isolated from a patient with two heterozygous RNASEH1 mutations, the V142I mutation (see Fig 4A), and a second mutation, (R157*), causing a truncated form of the enzyme lacking the active site. In our experiments, we used primer extension with DNA isolated from these RNase H1-deficient patient cells and a wild-type (WT) control to map initiation sites for mtDNA synthesis in vivo. We designed one primer to detect 5′-ends close to OriH and the CSB-region (primer complementary to H-strand positions 8-29, Fig 5A, Primer 1) and another to detect 5′-ends further downstream, in the D-loop region (primer complementary to H-strand positions 16,231-16,251, Fig 5A, Primer 2). The isolated DNA was analyzed before and after treatment with E. coli RNase H2, which will specifically degrade the RNA part of the covalently linked RNA-DNA molecule and thereby allow for exact mapping of RNA to DNA transitions. In control cells, 5′-ends were detected at mtDNA positions 191-194, 171-176, 148-153 and 110-113 (Fig 5B, lane 1) as well as around mtDNA position 60 (Fig 5C, lane 1). The observed 5′-ends were not altered by RNase H2 treatment, demonstrating that they were fully processed i.e. had no RNA residues attached (Fig 5B and 5C, compare lanes 1 and 2). Using mtDNA isolated from RNase H1 deficient cells, we observed a strong increase in 5′-ends. Some 5′-ends were similar to, but sometimes more abundant, than the products observed in WT cells at positions 191-194, 148-153, 171-176 and around position 60 (Fig 5B and 5C, lane 3). We also noted a range of new 5′-ends not present in WT cells at positions ~305-315, ~240, 209-217 and 119 (Fig 5B, lane 3) and throughout the D-loop region (Fig 5C, lane 3). Interestingly, at least three of these new 5′-ends were altered upon RNase H2 treatment suggesting that they represented new RNA to DNA transition sites not present in the WT control. These new sites were located in the CSB region at positions 305-315 and 209-217 (Fig 5B, compare lanes 3 and 4), and in the D-loop, at positions 16,371-16,383 (Fig 5C, compare lanes 3 and 4). Our data thus suggested that loss of RNase H1 activity caused initiation of mtDNA synthesis from multiple sites not used in WT cells.
To further support our findings, we also used an alternative method to identify 5′-ends and RNA to DNA transitions. We performed 5′-End-seq and HydEn-seq on the mtDNA from control and patient cells deficient in RNase H1. The 5′-End-seq method will detect all 5′-ends of DNA, including DNA molecules with ribonucleotides on the 5′-end. In contrast, HydEn-Seq will only detect DNA 5′-ends, since any RNA residues will be chemically removed [43]. By comparing results from 5′-End-seq and HydEn-seq, it is thus possible to identify RNA to DNA transition sites. The major DNA 5′-end in the control cells was mapped to position 111 (Fig 5D). There were no major differences between the 5′-End-seq and HydEn-seq samples indicating that no RNA-DNA transitions are found in this region in control cells (Fig 5D and 5E). We observed more reads in the control region of RNase H1-deficient patient cells (Fig 5F) compared to control cells, in agreement with the increased 5′-ends found by primer extension. Interestingly, 5′-ends of DNA were clearly shifted when compared to the control cells. New large peaks appeared around positions 235 and 56 and a region spanning 16,569/0 to 16,300 (Fig 5F). When comparing the 5′-End-seq to the HydEn-seq data for the peaks in the 16,569/0 to 16,300 region, it was clear that all the peaks decreased in the HydEn-seq data, with the exception of a peak at 16,374, which was slightly increased (Fig 5G). One region with new peaks also appeared in the HydEn-seq data, around position 200. In conclusion, the sequencing results agreed with the findings obtained by primer extension. The data from these two methods revealed an overall increase in 5′-ends in the control region of RNase H1-deficient patient cells as well as new RNA to DNA transition points. In conclusion, loss of RNase H1 activity leads to initiation of mtDNA synthesis at multiple sites not used in WT cells.
It has long been recognized that mammalian mtDNA replication is initiated at OriH [19–21]. Studies in mitochondrial extracts have also demonstrated that R-loops are formed in the region downstream of LSP and linked these structures to priming of mtDNA replication in the OriH-region [25]. The precise mechanisms of the proposed model have remained obscure, in part because of technical limitations. For instance, mitochondria cannot be transfected, a shortcoming that has prevented a detailed structural characterization of OriH. We here use an alternative approach to study OriH function, employing purified, recombinant proteins and defined DNA templates. Our work builds on recent in vivo observations, which demonstrated that RNase H1 is required for R-loop processing and mtDNA replication in a mouse knockout model system [30].
Early reports suggested that the LSP transcript involved in R-loop formation must be cleaved to form the 3′-OH termini required for initiation of replication at OriH. RNase MRP was identified as a candidate for this process [46–49], but the idea was later abandoned, since experimental evidence argued against the existence of RNase MRP in mitochondria [50–52]. As demonstrated here, the R-loops formed by LSP-transcription are instead processed by RNase H1 and the 3′-ends formed are used to prime mtDNA synthesis by POLγ. The process is stimulated by mtSSB, which acts to stabilize R-loops, most likely by binding to the displaced DNA strand. In addition, mtSSB prevents non-promoter-specific initiation of transcription from single-stranded stretches of DNA [15], thereby restricting replication initiation to the OriH-region. Our findings receive support from previous work, which has demonstrated that Rnaseh1 deletion in mice leads to mtDNA depletion [53] and that disease-causing mutations in RNASEH1 impair mtDNA replication [45, 53, 54].
In our work, we found that DNA replication in vitro was mainly initiated from the CSBII and CSBIII regions. The observed RNA to DNA transitions mapped in our recombinant system therefore correlate with the RNA to DNA transitions previously mapped in mtDNA by David Clayton and colleagues [19, 20, 47, 48]. In later papers, primer extension and ligation mediated PCR was used to map RNA to DNA transition points and only one of the mapped regions, that downstream of CSBII, was identified as a replication initiation site in vivo [26, 29]. What was not known at the time of these later studies was that the CSBII sequence forms strong G-quadruplex structures in both DNA and RNA [27, 28]. Many DNA polymerases have problems bypassing such G-quadruples and primer extension can in fact be used to detect these non-B-form DNA structures [55]. The way that these experiments were devised may therefore have correctly mapped replication initiation sites located downstream of CSBII, but failed to reach 5′-ends on nascent mtDNA located upstream of the G-quadruplex-forming CSBII region, thereby missing initiation events near CSBIII.
Interestingly, mutations that reduce RNase H1 activity in vivo lead to an increase of 7S DNA levels [45]. We believe that this effect is due to unregulated initiation of 7S DNA synthesis. In support of this notion, our primer extension and 5′-end sequencing results show an overall increase in free 5′-ends and identified multiple new RNA to DNA transition sites. Apparently, RNase H1 is required to restrict initiation of 7S DNA and mtDNA synthesis to OriH. The enzyme removes all RNA hybridized to the DNA template, with the exception of the RNA-loop molecules in the CSB region, which after RNase H1 processing can be used as primers. In the absence of RNase H1, 7S DNA synthesis is instead initiated from multiple locations, possibly from any RNA molecule with a 3′-end hybridized to the DNA template, leading to unregulated initiation of 7S DNA synthesis and an increase in overall 7S DNA levels. In contrast, the mtDNA levels remain largely unchanged in RNASEH1 mutant cells. This observation is in agreement with the notion that regulation of mtDNA levels is not correlated to 7S DNA levels, but takes place at the end of the D-loop [56]. At this place, there is a regulated switch between abortive (7S DNA) and genome length mtDNA replication, which explains why 7S DNA can be strongly up regulated, whereas mtDNA levels remain unaffected. The overall problem in the case of RNase H1 deficiency could instead be related to downstream events, such as DNA replication termination and mtDNA segregation, as the newly synthesized mtDNA will have shifted 5′-ends and thus also replication end points.
TEFM prevents premature transcription termination at CSBII, which led to the suggestion that this protein can function as a regulator of primer formation, controlling the switch between transcription and mtDNA replication [4, 5]. In support of this claim, addition of TEFM reduces R-loop formation in vitro. The effects of TEFM on replication initiation are less dramatic, as robust initiation is seen even when POLRMT and TEFM are present at equimolar levels. This observation does not necessarily argue against the idea that TEFM acts to regulate the switch between transcription and replication. It is still possible that variations in TEFM concentrations will affect the relative levels of primer formation and transcription elongation in vivo. However, TEFM does not cause an all or nothing effect, since the protein may be present at relatively high concentrations without abolishing replication initiation. Also arguing against an essential role for TEFM in regulating primer synthesis, depletion of this factor in cell lines does not lead to changed levels of mtDNA [3]. Clearly, more work is required to elucidate the precise role of TEFM in the regulation of mtDNA replication initiation.
RNase H1 processing of R-loops may be influenced by a multitude of factors. The G-quadruplex structure formed between RNA and the non-template DNA strand at CSBII can reduce RNase H1 cleavage [27]. RNase H1 activity can also be influenced by other proteins. There could be e.g. direct protein interactions between RNase H1 and the transcription or replication proteins. In addition, RNase H1 has been shown to interact with the mitochondrial protein P32 and that this interaction significantly enhances the cleavage activity of RNase H1 on heteroduplex templates. If P32 also affects primer formation is not known [33]. We will address these intriguing possibilities in future studies.
In conclusion, we here present a mechanism for primer formation and initiation of DNA replication in human mitochondria (as shown in Fig 6). As previously suggested [20, 23, 25], there are indeed striking similarities between replication initiation in mammalian mitochondria and the process described for the ColE1 plasmid. Both systems depend on promoter driven transcription, R-loop formation and primer processing. Important aspects of our model are supported by in vivo evidence, but more work is clearly needed to validate our ideas and refine the proposed mechanisms. Of special importance will be to clarify how the switch between transcription and primer synthesis is regulated.
RNase H1 was expressed using the baculovirus system (Sf9 cells). The protein coding sequence was PCR-amplified from human cDNA and cloned into the pBacPAK9 vector (Clontech). The construct lacked the N-terminal MTS (amino acids 1-26), and carried a C-terminal 6×His-tag. Recombination and cell infection was performed as described in the BacPAK manual (Clontech). The protein was expressed and purified as previously described for transcription proteins [5]. All other recombinant proteins were expressed and purified as described previously [5, 41].
All templates used were either empty pUC19 plasmids or mitochondrial DNA sequences with or without modifications cloned in pUC18. A list of all templates used in this study can be found in S1 Table. All supercoiled templates were carefully prepared with QIAprep Spin Miniprep Kit (QIAGEN) and kept at 4°C. To produce relaxed templates, supercoiled plasmids were treated with Topoisomerase I (New England Biolabs).
All transcription reaction volumes were 25 μL and contained 25 mM Tris-HCl pH 8.0, 10 mM MgCl2, 64 mM NaCl, 100 μg/mL BSA, 10 mM DTT, 400 μM ATP, 150 μM GTP, 150 μM CTP 10 μM UTP, 0.027 μM α-[32P]UTP (3000 Ci/mmol), 4 nM of indicated plasmid template, 20 nM POLRMT, 200 nM TFAM, 60 nM TFB2M, and 40 nM TEFM where indicated. The reactions were incubated at 32°C for 5 min and stopped by the addition of 200 μL stop buffer (10 mM Tris-HCl pH 8.0, 0.2 M NaCl, 1 mM EDTA, 100 μg/mL glycogen (Roche) and 100 μg/mL proteinase K (Ambion)) followed by incubation at 42°C for 45 min. The transcripts were recovered by ethanol precipitation and the pellets were dissolved in 20 μL gel loading buffer (98% formamide, 10 mM EDTA, 0.025% xylene cyanol FF, and 0.025% bromophenol blue) and heated at 95°C for 3 min. The samples were analyzed on 4% denaturing polyacrylamide gels (1 × TBE and 7 M urea) followed by exposure on photo film. Low Molecular Weight DNA Ladder (NEB) was used as a size marker. All experiments were performed multiple (>3) times with similar results and each figure shows a representative gel image for that experiment.
For R-loop detection, 1.5 μL of 5.0 M NaCl was added to each sample after the in vitro transcription reaction, followed by addition of 250 ng of RNase A (ThermoFisher Scientific) and incubation at 32°C for 5 min. The reactions were stopped and evaluated as the regular transcription reactions. Quantifications of R-loops were performed using ImageJ software (https://imagej.nih.gov/ij/). The intensity of R-loops larger than 100 nts was divided by the intensity of the CSBII transcript of the same reaction, to obtain a ratio indicating how efficient R-loop formation was. All experiments were performed multiple times with similar results and each figure shows a representative gel image for that experiment.
All reaction volumes were 25 μL and contained 25 mM Tris-HCl pH 8.0, 10 mM MgCl2, 50 mM NaCl, 100 μg/mL BSA, 10 mM DTT, 400 μM ATP, 150 μM GTP, 150 μM CTP 150 μM UTP, 100 μM dATP, 100 μM dGTP, 100 μM dCTP or ddCTP as indicated, 10 μM dTTP, 0.027 μM α-[32P]dTTP (3000 Ci/mmol), and 8 nM of indicated template. In the case of RNA labeling the concentration of dTTP were 100 μM and UTP concentrations were 10 μM UTP and 0.027 μM α-[32P]UTP (3000 Ci/mmol). All reactions (unless otherwise stated) contained 200 nM of TFAM, 60 nM of TFB2M, 20 nM of POLRMT, 20 nM of D274A POLγA exo- (or WT in Fig 3F lane 6) and 40 nM POLγB. RNase H1 was added at 2 nM and mtSSB was added at 40 nM unless otherwise stated. The reactions were incubated at 32°C for 30 min. Experiments with dCTP were stopped and evaluated as transcription reactions. Experiments with ddCTP were stopped by the addition of 5 μL stop buffer (to a final concentration of 10 mM Tris-HCl pH 8.0, 0.2 M NaCl, 1 mM EDTA, 660 μg/mL glycogen (Roche) and 100 μg/mL proteinase K (Ambion)) followed by incubation at 42°C for 45 min. The reactions were treated with KOH (300 mM) for 2 hrs at 55°C. The samples were recovered by ethanol precipitation in the presence of 0.5 volumes ammonium acetate (7.5 M), dissolved in 10 μL gel loading buffer (98% formamide, 10 mM EDTA, 0.025% xylene cyanol FF, and 0.025% bromophenol blue), heated at 95°C for 3 min. The products were analyzed on 6% denaturing polyacrylamide gels (1 × TBE and 7 M urea) for samples with dCTP or 12% denaturing polyacrylamide sequencing gels (1 × TBE and 7 M urea) for samples with ddCTP. The gels were exposed on photo film. All experiments were performed multiple times with similar results and each figure shows a representative gel image for that experiment.
Patient and control fibroblast cells were grown in DMEM GlutaMAX medium, supplemented with 10% FBS, PEST and 10 μg/ml uridine in 5% humidified atmosphere at 37°C. Approximately 5 × 106 cells were collected and lysed for 30 min at 42°C in lysis buffer (10 mM Tris-HCl pH 8.0, 5 mM EDTA, 10% SDS). An equal volume of phenol-chloroform was added, samples were mixed and centrifuged (15,000 g for 5 min, 4°C). The aqueous phase was saved and 100 mM NaCl and one volume isopropanol were added. After precipitation at -20°C for one hour, the samples were centrifuged (15,000 g, 20 min, 4°C) and the pellets were washed with 70% EtOH. The DNA was resuspended in 100 μl TE buffer. DNA concentrations were measured using the Qubit fluorometric instrument (ThermoFisher Scientific).
Isolated DNA (1.8 μg) was incubated for 1 hour at 37°C with or without RNase H2 (NEB). Primer extension was performed with 2 U Taq DNA polymerase (NEB) in 1X ThermoPol buffer, 200 μM dNTPs and 1.5 pmol labeled primer. The primers were 5´-end labeled with PNK (NEB) and γ-[32P]ATP and were corresponding to L-strand positions 8-29 (GGT CTA TCA CCC TAT TAA CCA C) and 16,331-16,351 (CAC ACA TCA ACT GCA ACT CCA). The primer extension reaction was performed with 5 minutes at 95°C, 30 seconds at 95°C, 30 seconds at 58°C, 45 seconds at 72°C and 5 minutes at 72°C with step 2-4 repeated in 20 cycles. The reactions were stopped and ethanol precipitated as for replication initiation reactions. The sequencing ladders were prepared with USB Sequenase Version 2.0 (Affymetrix) according to the manufacturers protocol. The primer extension experiment was performed multiple times with similar results and the figure shows a representative gel image for that experiment.
Free 5′-ends of mtDNA from fibroblast cells were mapped by 5′-End-seq or HydEn-seq essentially as previously described [43,57]. In brief, 1 μg DNA was treated with 0.3 M KCl or 0.3 M KOH at 55°C for 2 hours. Samples were then phosphorylated with 3′-phosphatase-minus T4 polynucleotide kinase (New England BioLabs) for ligation to oligonucleotide ARC140. After an adaptor was annealed (ARC76–ARC77), T7 DNA polymerase (New England BioLabs) was used to synthesize the second strand. Purified libraries were then sequenced using an Illumina NextSeq500 instrument. Reads were trimmed for quality and adapter sequence with cutadapt (-m 15 –q 10match-reas-wildcards). Pairs with one or both reads shorter than 15 nts were discarded. Mate 1 of the remaining pairs was aligned to an index containing the sequence of all oligos used in the preparation of these libraries with bowtie sing bowtie 0.12.8 (-m1 -v2), and all pairs with successful alignments were discarded. Pairs passing this filter were subsequently aligned to the hg38 H. sapiens reference mitochondrial genome where the mitochondrial genome was cleaved at position 4,000 and OriH region was religated (-m1 -v2 -X10000 --best). Single-end alignments were then performed using mate 1 of all unaligned pairs (-m1 -v2). The count of 5′-ends of all unique paired-end and single-end alignments were determined and these counts were converted to bedGraph format for visualization. Sequencing data have been deposited in the Gene Expression Omnibus under accession number GSE103612.
|
10.1371/journal.pgen.1007435 | A neuronal MAP kinase constrains growth of a Caenorhabditis elegans sensory dendrite throughout the life of the organism | Neurons develop elaborate morphologies that provide a model for understanding cellular architecture. By studying C. elegans sensory dendrites, we previously identified genes that act to promote the extension of ciliated sensory dendrites during embryogenesis. Interestingly, the nonciliated dendrite of the oxygen-sensing neuron URX is not affected by these genes, suggesting it develops through a distinct mechanism. Here, we use a visual forward genetic screen to identify mutants that affect URX dendrite morphogenesis. We find that disruption of the MAP kinase MAPK-15 or the βH-spectrin SMA-1 causes a phenotype opposite to what we had seen before: dendrites extend normally during embryogenesis but begin to overgrow as the animals reach adulthood, ultimately extending up to 150% of their normal length. SMA-1 is broadly expressed and acts non-cell-autonomously, while MAPK-15 is expressed in many sensory neurons including URX and acts cell-autonomously. MAPK-15 acts at the time of overgrowth, localizes at the dendrite ending, and requires its kinase activity, suggesting it acts locally in time and space to constrain dendrite growth. Finally, we find that the oxygen-sensing guanylate cyclase GCY-35, which normally localizes at the dendrite ending, is localized throughout the overgrown region, and that overgrowth can be suppressed by overexpressing GCY-35 or by genetically mimicking elevated cGMP signaling. These results suggest that overgrowth may correspond to expansion of a sensory compartment at the dendrite ending, reminiscent of the remodeling of sensory cilia or dendritic spines. Thus, in contrast to established pathways that promote dendrite growth during early development, our results reveal a distinct mechanism that constrains dendrite growth throughout the life of the animal, possibly by controlling the size of a sensory compartment at the dendrite ending.
| Lewis Carroll's Alice told the Caterpillar, "Being so many different sizes in a day is very confusing." Like Alice, the cells of our bodies face a problem in size control – they must become the right size and remain that way throughout the life of the organism. This problem is especially relevant for nerve cells (neurons), as the lengths of their elaborate dendrites determine the connections they can make. To learn how neurons control their size, we turned not to a Caterpillar but to a worm: the microscopic nematode C. elegans, in which single neurons can be easily visualized and the length of each dendrite is highly predictable across individuals. We focused on a single dendrite, that of the oxygen-sensing neuron URX, and we identified two genes that control its length. When these genes are disrupted, the dendrite develops correctly in embryos but then, like Alice, grows too much, eventually extending up to 1.5x its normal length. Thus, in contrast to known pathways that promote the initial growth of a dendrite early in development, our results help to explain how a neuron maintains its dendrite at a consistent length throughout the animal's life.
| Neurons embody the adage that structure determines function in biology. The geometries and lengths of axons and dendrites determine the path and timing of information flow. Dendrites, in particular, are sculpted in ways that control how information is integrated and computed. One well-studied aspect of dendrite morphogenesis is the shaping of subcellular compartments called dendritic spines, which are micron-scale protrusions that serve as sites of information transfer via synapses. Work from many groups has elucidated how the lengths and shapes of dendritic spines correlate with synaptic activity [1,2]. An analogous subcellular compartment, called a sensory cilium, serves as the site of information transfer on sensory receptor neurons (including photoreceptor rods and cones, olfactory neurons, and auditory hair cells) [3]. Similar to dendritic spines, sensory cilia exhibit highly regulated morphologies that can be altered by sensory input [4,5]. In addition to these very conspicuous structures, dendrites may also contain less obvious subcellular compartments that aid in signal processing; for example, excitatory and inhibitory synapses are often distributed non-uniformly along dendrites, suggesting de facto compartmentalization [6,7]. Further, studies of signal propagation within dendrites have suggested that dendritic branches may serve as discrete computational units [8]. Yet, we know relatively little about how subcellular regions of dendrites, other than spines and cilia, are shaped.
C. elegans provides a powerful model for dissecting the control of dendrite morphogenesis. It exhibits highly stereotyped neuronal anatomy, such that the position of each neuron and its pattern of neurite outgrowth, branching, and connectivity have been extensively catalogued and are highly reproducible across individuals [9]. Most C. elegans neurons extend one or two simple, unbranched processes that often have mixed pre- and post-synaptic identity and thus cannot be classified as strictly dendritic or axonal in character [9]. In contrast, most of its sensory neurons have a bipolar morphology with one neurite that solely receives information from the environment and is therefore a dendrite [10]. While C. elegans has been used extensively as a model for axon development [11], considerably less attention has been paid to its dendrites. Most of the work has focused on interesting sensory neurons with complex branched dendrites, and has revealed mechanisms of dendritic branch formation and spacing [12–23], dendritic self-avoidance [24], and even dendrite tiling [25]. By comparison, dendrites with simple unbranched morphologies have been used as a model for understanding axon-dendrite polarity [26,27]. They also offer a powerful reductionist system for identifying mechanisms that control dendrite length. Most sensory neurons in the head extend a simple unbranched dendrite to the nose and terminate in a sensory cilium that receives information from the environment [10,28,29]. Through studying one such class of neurons, the amphids, we showed that dendrite length is controlled by a process called retrograde extension, in which the neuron is born at the embryonic nose and attaches a short dendrite there, after which the cell body migrates away, stretching out its dendrite behind it [30]. The extracellular matrix protein DYF-7 is required to anchor the dendrite ending at the nose, and dyf-7 mutants exhibit severely shortened dendrites [30].
Another sensory neuron in the head, URX, offers an interesting contrast to the amphid. URX also extends a simple unbranched dendrite to the nose but it does not terminate in a sensory cilium and its dendrite length is unaffected in a dyf-7 mutant [10,28]. URX is an oxygen-sensing neuron that detects oxygen partly through guanylate cyclase proteins that localize to a subcellular region at the dendrite ending that is neither a sensory cilium nor a dendritic spine [31–35]. We reasoned that study of URX morphogenesis would reveal distinct mechanisms for the control of dendrite length, possibly including the organization of subcellular regions within dendrites. Therefore, we undertook a forward genetic screen to identify mutants that affect URX dendrite morphology. We identified a class of mutants that affect dendrite length in the direction opposite to what we had seen before: in these mutants, the URX dendrite dramatically overgrows up to 150% of its normal length. Analysis of one of these overgrowth mutants identified a novel cell-autonomous mechanism that constrains sensory dendrite growth in an age- and activity-dependent manner.
The two bilaterally symmetric URX neurons are bipolar, with cell bodies situated posterior to the nerve ring, and a simple unbranched dendrite projecting to the tip of the nose (Fig 1A) [9,10,28]. To identify genes involved in URX morphogenesis, we chemically mutagenized hermaphrodites bearing an integrated GFP reporter for URX (flp-8pro:GFP, expressed strongly in URX and more dimly in the AUA neuron; [36]). We visually screened non-clonal F2 progeny for individuals with defective URX dendrites. From this screen, we isolated 17 recessive mutants that affect URX morphogenesis. Eight mutants exhibit shortened dendrites with few other morphological defects and will be reported elsewhere. The remaining nine mutants were grouped into four phenotypic classes: (I) dendrite overgrowth (Fig 1B; hmn5, hmn6, hmn17); (II) branching or swelling near the dendrite ending (Fig 1C; hmn2); (III) pleiotropic morphological defects including misplaced cell bodies with shortened dendrites (Fig 1D; hmn16) or grossly disorganized or missing dendrites (Fig 1E; hmn11, hmn14, hmn15); and (IV) loss of flp-8pro:GFP expression in URX (Fig 1F; hmn13) suggesting a defect in specifying URX cell fate. The penetrance of these phenotypes is shown in Table 1.
The overgrowth (Class I) mutants will be described in detail below. The branching (Class II) mutant exhibited a bulge in most URX neurons (n = 13/16) at a stereotyped position along the length of the dendrite (80.4 ± 3.4% of the distance from the cell body to the dendrite ending). In 7/13 cases, this took the form of a short branch. We performed whole-genome sequencing of pooled recombinants and identified a premature termination codon in tni-3 (S1A Fig). A fosmid containing wild-type tni-3 completely rescued the mutant phenotype, suggesting this is the causative mutation (S1B Fig). tni-3 encodes one of four C. elegans homologs of troponin I, an inhibitory component of the troponin complex which regulates muscle contraction [37]. tni-3 is expressed in head muscle cells [38] and the location of the ectopic dendrite branch is consistent with the URX dendrite invading into the space between these cells, suggesting that the branching phenotype may be a secondary consequence of defects in head muscle. The Class III and IV mutants were not characterized further.
We chose to focus on the overgrowth mutants because the phenotype is visually striking and highly penetrant. Rather than ceasing growth at the nose tip, the URX dendrite makes a U-turn at the nose and loops back in the direction of the cell body in 74% or 65% of hmn5 and hmn6 adults, respectively, ultimately extending up to 150% of its normal length (Fig 1B, Table 1, and Fig 2C). Further, the URX phenotype appears specific to dendrite length, as we did not observe defects in the expression of cell-specific markers, cell body positioning, dendrite branching, or other aspects of URX development. Finally, this phenotype is the opposite of the shortened dendrite phenotypes we had previously studied, and thus promised to shed light on a different aspect of dendrite growth control.
To identify the causative mutations in these strains, we performed whole-genome sequencing of pooled recombinants for hmn5 and hmn6. hmn5 contains a premature termination codon in the predicted gene C05D10.2 (Fig 2A). A fosmid containing wild-type C05D10.2 completely rescues the phenotype (Fig 2D). In a subsequent screen, we isolated another allele hmn51 that also exhibits URX overgrowth (74% of animals, n = 50; compare to Table 1), fails to complement hmn5, and bears a point mutation in this gene (Fig 2A). Together, these data indicate that disruption of C05D10.2 causes URX overgrowth. C05D10.2 is homologous to the mammalian MAP kinase ERK8/MAPK15, so we named this gene mapk-15. Recently, Piasecki et al., Kazatskaya et al., and Bermingham et al. independently identified roles for mapk-15 in ciliated and dopaminergic sensory neurons of C. elegans [39–41]. As URX is neither ciliated nor dopaminergic, it is unclear how these roles relate to URX overgrowth.
hmn6 contains a premature termination codon in sma-1, is completely rescued by a wild-type sma-1 fosmid, and exhibits the characteristic small head and body size (Sma) phenotype associated with this gene (Fig 2B and 2C) [42]. hmn17 also exhibits the Sma phenotype, fails to complement hmn6, and contains a point mutation in sma-1 (Fig 2B). The previously described allele sma-1(e934) also exhibits URX overgrowth (Fig 2C). sma-1 encodes βH spectrin, which is expressed in epithelial cells, links the cytoskeleton to the apical plasma membrane, and is required for normal embryonic elongation [43,44]. However, the sma-1 dendrite defects are unlikely to be explained entirely by reduced head size, as sma-2 and sma-3 mutants also have reduced head size but do not exhibit comparable URX overgrowth (Fig 2C; mean distance from URX cell body to nose (μm) ± SD: wild type, 112 ± 8; sma-1(hmn6), 69 ± 6; sma-1(e934), 78 ± 8; sma-2(e502), 79 ± 5; sma-3(e491), 86 ± 5).
To understand when mapk-15 and sma-1 act, we examined the extent of URX overgrowth at different developmental stages. We very rarely observed dendrite overgrowth in first larval stage (L1) animals, suggesting overgrowth is not primarily due to defects in initially shaping the dendrite during embryonic development (Fig 3A and 3F). Rather, dendrite overgrowth increases with age, reaching moderate penetrance in L4 animals and becoming even more pronounced in two-day adults (penetrance in L1, L4, 2A: 4%, 44%, 74% in mapk-15(hmn5); 8%, 17%, 65% in sma-1(hmn6); Fig 3). The extent of the overgrowth also increases with age (Fig 3F). Moreover, rare adult mapk-15 animals on crowded, starved plates exhibit profusely overgrown dendrites that include ectopic branching, a phenotype we never observed in wild-type animals (Fig 3D). These results suggest that mapk-15 and sma-1 act post-embryonically to constrain dendrite growth throughout the life of the animal.
To test this hypothesis, we generated constructs in which a wild-type mapk-15 genomic fragment is placed under control of heat-shock-inducible promoters (hsp, see Methods). We did not perform this experiment with sma-1 due to its larger size (~12 kb coding sequence). We grew mapk-15(hmn5) animals bearing these transgenes, induced expression of wild-type mapk-15 in L4 animals, and then measured URX dendrite length in two-day adults. We observed a strong, albeit incomplete, rescue of the overgrowth defect (Fig 3G). These results indicate that expression of mapk-15 post-embryonically, at the time when the phenotype appears, is sufficient to restrict growth. Induction of expression in embryos or two-day adults showed a modest (p = 0.08) or not appreciable (p = 0.94) effect, respectively (S2 Fig). These results suggest that mapk-15 may temporally regulate URX dendrite growth.
Next, we wanted to know in which cells mapk-15 and sma-1 act to regulate dendrite growth. Previous studies had shown that sma-1 is expressed primarily in hypodermis, intestine, and pharynx [43,44]. To determine where mapk-15 is expressed, we generated a transcriptional reporter and found that it labels URX as well as many other head sensory neurons at all larval stages (Fig 4A; expression in ciliated sensory neurons is also reported in [39–41]), suggesting mapk-15 could act cell-autonomously in URX.
To determine if mapk-15 and sma-1 normally act in URX, we performed mosaic experiments in which mutant animals carried an extrachromosomal array bearing a fosmid encompassing the wild-type gene together with a fluorescent marker that allowed us to score the presence or absence of the transgene in URX. Such extrachromosomal arrays are stochastically lost during cell division, which creates genetic mosaics. We selected adult mosaic animals in which the array was present in one URX neuron but absent in the contralateral URX neuron, and assessed dendrite overgrowth. We considered four possible outcomes: (I) overgrowth in neither neuron; (II) overgrowth only in the neuron with the array; (III) overgrowth only in the neuron without the array; or (IV) overgrowth in both neurons (Fig 4B).
For mapk-15 mosaics, we found that 72% of animals were type III and 28% of animals were type I (Fig 4B). Thus, when the rescuing array is present in URX, the dendrite is always normal length, and when the array is absent from URX, the dendrite exhibits overgrowth at the previously observed penetrance of the mapk-15 phenotype (compare with 74%, Table 1). This result suggests that mapk-15 acts cell-autonomously in URX to constrain dendrite growth. In contrast, for sma-1 mosaics, we found that 98% of animals were type I, suggesting that the overgrowth phenotype was efficiently rescued regardless of the presence or absence of the array in URX (Fig 4B), and thus sma-1 likely acts in other cells to constrain URX dendrite growth.
An advantage of mosaic experiments is that the rescuing transgene is expressed under control of its native regulatory sequences; however, because these experiments rely on loss of the transgene during cell division, they can fail to resolve cells that are closely related by developmental lineage. Therefore, to further test whether mapk-15 can function cell-autonomously in URX, we expressed a transgene containing a wild-type mapk-15 genomic fragment under control of its own promoter or a URX-specific promoter (flp-8pro) in mutant animals and measured dendrite lengths in two-day adults. We found that expression of mapk-15 with either promoter rescued the overgrowth phenotype (Fig 4C). Together, these results suggest that sma-1 likely acts non-cell-autonomously from neighboring tissues, whereas mapk-15 is normally expressed in URX and its expression in URX is necessary and sufficient to constrain dendrite growth.
To gain insight into how MAPK-15 regulates URX dendrite growth, we asked where it localizes in the neuron and whether it requires its predicted kinase activity. We generated a construct encoding a superfolderGFP(sfGFP)-MAPK-15 fusion protein (see Methods) and expressed it in URX. This construct rescues the mapk-15 dendrite phenotype, indicating the fusion protein is functional (S3A Fig). We found that sfGFP-MAPK-15 localizes throughout the neuron with ~4-fold relative enrichment at the URX dendrite ending (relative fold enrichment ± SD, 3.7 ± 2.0), suggesting it may act locally to constrain growth (Fig 5A and 5B, S3C and S3D Fig). Notably, many neuronal receptive endings–including cilia of C. elegans sensory neurons and dendritic spines in mammals– are known to contain signaling factors that regulate their own growth, typically on the scale of 1–5 μm [3,45]. In these examples, the receptive ending is a discrete compartment that is biochemically isolated from the rest of the dendrite. In contrast, the URX dendrite is not ciliated and ultrastructural studies have not revealed any obvious physical compartmentalization at its ending. However, guanylate cyclase proteins that mediate URX sensory functions (e.g. GCY-35) have been shown to localize to the dendrite ending, raising the possibility that the dendrite contains a subcellular compartment at its ending that is specialized for sensory signaling [33].
To test whether MAPK-15 might participate in signaling, we asked whether it functions as a kinase. MAPK-15 has several conserved domains that are typical of MAP kinases, including an ATP-binding pocket (Fig 2B). Previous work on mammalian MAPK15 has shown that a conserved lysine residue in this domain is necessary for kinase activity [46,47]. We therefore generated a presumed kinase-dead allele (MAPK-15(K42A), see Methods) under the control of its own promoter, introduced it as a transgene into mutant animals, and asked whether it could rescue dendrite overgrowth. We observed no rescue, consistent with the idea that MAPK-15 requires its kinase activity (Fig 5C). We also generated an allele lacking the LC3-interacting region (MAPK-15(ΔLIR)), a motif that has been shown in mammalian MAPK15 to bind factors involved in autophagy [48,49]. This allele also failed to rescue (Fig 5C). While we cannot exclude kinase- or LIR-dependent roles for MAPK-15 elsewhere in the cell, taken together these results suggest that MAPK-15 may constrain dendrite growth through localized signaling or regulatory interactions at the dendrite ending.
Given that GCY-35 normally localizes to the URX dendrite ending near the nose, we wondered where it would localize in an overgrown dendrite. We considered three possibilities: it might localize to the overgrown dendrite ending; to the region of the dendrite closest to the nose; or, throughout the overgrown portion of the dendrite. The first two possibilities would suggest that mapk-15 regulates the growth of the dendrite but not of the sensory region, whereas the third possibility suggests an expansion of the sensory region itself.
Therefore, to distinguish these possibilities, we examined the localization of GCY-35-GFP in URX in wild-type and mapk-15 animals. We found that GCY-35-GFP was enriched at the dendrite ending in wild-type animals, consistent with previous reports, as well as in mapk-15 young animals prior to the appearance of the dendritic overgrowth (Fig 6A). However, in mapk-15 L4 animals in which the dendritic overgrowth was apparent, GCY-35-GFP localized throughout the overgrown region (Fig 6B). Its localization also expanded into more proximal regions of the dendrite where it is not normally enriched (Fig 6B). We obtained similar results with the globin GLB-5, which regulates oxygen sensing and also localizes to the wild-type URX dendrite ending (S4A Fig) [33]. These results are consistent with the notion that dendritic overgrowth corresponds to expansion of a subcellular sensory region at the dendrite ending.
While examining GCY-35-GFP localization, we were surprised to find that overexpression of this transgene suppressed the mapk-15 overgrowth phenotype (Fig 6C, penetrance with and without GCY-35-GFP overexpression, 62% and 24% respectively, p<0.0001). This result raises the possibility that elevated guanylate cyclase function might mitigate the effects of mapk-15 disruption.
We were concerned that overexpression of GCY-35 might cause phenotypic changes for reasons unrelated to its guanylate cyclase activity. Therefore, we performed two additional controls. First, we introduced a point mutation in GFP-GCY-35 that inactivates its catalytic site [31]. We found that overexpression of this transgene in URX failed to suppress the mapk-15 overgrowth (Fig 6C), indicating that suppression requires the guanylate cyclase activity. Second, we reasoned that overexpression of an active guanylate cyclase is likely to increase cellular cGMP levels, which could then act on cGMP-dependent protein kinases and cyclic nucleotide-gated (CNG) channels [50]. Therefore, we asked whether the CNG channel TAX-2 is important for GCY-35-mediated suppression of dendrite overgrowth. Indeed, we found that in mapk-15; tax-2 mutant animals, GCY-35-mediated suppression of overgrowth was partially relieved (Fig 6C).
Finally, we wanted to use an independent genetic manipulation to mimic elevated cGMP levels and ask whether that would also suppress dendrite overgrowth. For this purpose, we took advantage of a gain-of-function mutation in the cGMP-dependent protein kinase EGL-4 [egl-4(gf)] [51]. Consistent with the results above, egl-4(gf) efficiently suppressed the mapk-15 dendrite overgrowth (Fig 6C, penetrance with and without egl-4(gf), 62% and 22% respectively, p<0.00001). This effect is specific to egl-4 hyperactivity, as an egl-4 loss-of-function mutation did not alter URX dendrite length or enhance the mapk-15 overgrowth (S2 Fig). Together, these results suggest that genetically mimicking increased cGMP signaling suppresses dendrite overgrowth.
Sensory cilia can increase in length when deprived of sensory input, as if trying to compensate for reduced sensory signaling [4]. Although URX is not ciliated, we wondered whether mapk-15 mutants might have reduced sensory signaling relevant to dendrite overgrowth. Rising oxygen levels evoke a tonic increase in intracellular calcium levels in URX neurons [35,52]. We introduced a genetically-encoded calcium reporter, yellow cameleon 2.60, into mapk-15 animals and used 2-day adults exhibiting dendrite overgrowth to visualize calcium changes in the URX cell body in response to a 7%–21% oxygen shift. As a control, sensory-defective gcy-35 mutant animals completely lack an oxygen response (Fig 6D). In contrast, we found that oxygen-evoked calcium responses in the mapk-15 mutant were indistinguishable from those of wild-type animals (Fig 6D).
We also measured behavioral responses to changes in ambient oxygen levels. Because the standard laboratory reference strain (N2) carries a gain-of-function mutation in the neuropeptide receptor NPR-1 that suppresses behavioral responses to oxygen, we performed our behavioral assays in strains carrying an npr-1 loss-of-function mutation that mimics the behavior of true wild isolates [32,53–55]. These npr-1 animals increase their speed of locomotion and re-orient their direction when challenged with elevated oxygen concentrations [32,53]. This behavior depends on URX and the oxygen sensors GCY-35 and GCY-36 [32,53]. We found that mapk-15 mutants have overall slower speed in response to elevated oxygen, but these responses are equivalent among mutant animals whether or not they exhibit the URX dendrite overgrowth. Thus, this altered behavioral change is unlikely to be caused by URX dendrite length and probably reflects other disruptions to organismal sensory function or general physiology in the mutant.
While these experiments cannot exclude subtle alterations in URX sensory processing, or the possibility that larger sensory defects are masked by adaptation in the neuron, taken together these data do not support the idea that dendrite overgrowth is a response to URX sensory deficits. Conversely, other mutants that cause URX sensory deficits have not been reported to lead to overgrowth, and we did not observe overgrowth with egl-4(lf) (S4B Fig). Therefore, our results suggest that overgrowth is unlikely to be a secondary consequence of URX sensory deficits, and thus point to a more direct role for MAPK-15 in regulating dendrite length.
How cells measure their own size is a basic problem in cell biology [56]. This problem takes on extraordinary dimensions when neurons need to measure the lengths of their own axons or dendrites, and few mechanisms of neuronal length-sensing are known [57,58]. Here, we show that MAPK-15 acts in the URX neuron to constrain the length of its sensory dendrite. In the absence of MAPK-15, the dendrite exhibits age-dependent overgrowth. MAPK-15 acts temporally when the overgrowth first appears and spatially at the dendrite ending. These results suggest that MAPK-15 acts locally to constrain dendrite length.
Two lines of evidence suggest that dendrite overgrowth corresponds to expansion of a subcellular sensory compartment. First, the oxygen-sensing guanylate cyclase GCY-35, which normally marks a subcellular region at the dendrite ending, expands its localization to fill the overgrowth. Second, genetic manipulations that mimic hyperactive sensory signaling can suppress the overgrowth. Intriguingly, three recent studies described a role for MAPK-15 in other information-processing subcellular compartments in neurons. Piasecki et al., Kazastaya et al., and Bermingham et al. report localization of MAPK-15 in sensory cilia and at dopaminergic synapses of certain ciliated sensory neurons [39–41]. Loss of MAPK-15 leads to altered sensory cilia morphology, defects in mating behaviors mediated by ciliated sensory neurons and defects in swimming behaviors affected by dopamine signaling [39–41]. While URX is not ciliated, these results are consistent with the idea that MAPK-15 acts at subcellular sensory compartments.
How might MAPK-15 control dendrite length? We can imagine two general models. First, the dendrite ending might be a relatively static structure, and loss of MAPK-15 creates an aberrant positive signal that tells the neuron to "make more dendrite." Alternatively, the dendrite ending might be in a state of dynamic equilibrium, where its steady-state length reflects a balance between the addition and retrieval of material. In this model, loss of MAPK-15 might shift the balance towards addition, leading to overgrowth, whereas genetically mimicking hyperactive sensory signaling might reduce the rate of addition, shifting the balance back. Our current data cannot distinguish between these models. However, it is interesting to note some circumstantial evidence that may point to the idea that MAPK-15 normally promotes retrieval of dendrite material. First, MAPK15/ERK8 plays a role in autophagy in other systems [48,49]. Second, its interaction with autophagic components is mediated through its conserved LC3-interacting region [49], which we found to be required to restrict dendrite growth. Third, in C. elegans ciliated neurons, MAPK-15 localizes near the periciliary membrane compartment, a region that is enriched for proteins involved in membrane trafficking [40,41]. Fourth, the role of MAPK-15 in C. elegans dopaminergic neurons has been suggested to involve membrane trafficking of a dopamine transporter at synapses [39]. Perhaps an altered balance of membrane delivery and retrieval underlies the defects at the URX dendrite ending as well as the phenotypes observed in other cell types.
Our results define a genetic pathway that affects dendrite length throughout the life of the organism. Little is known about how dendrite arbors remodel with age in mammalian brain. Many studies have focused on changes in dendritic spines, while relatively few have examined changes in dendrite length or branching [59]. Classical studies using Golgi staining to examine dendrite arbors in mammalian cortex reported a reduction in dendrite length and branching with age, including in humans [60,61]. The cause of this change– and whether it is simply "deterioration"– is not known. Paradoxically, some studies have reported dendrite branching to increase rather than decrease with age [62,63]. An intriguing possibility is that age-dependent dendritic changes reflect the overall balance of dendrite growth and retraction in mature life, which may differ between brain regions or neuron types. Indeed, time-lapse imaging of cortical neurons showed that some dendrite arbors remain highly dynamic during adulthood [64]. The mechanisms that control dendrite length throughout life have remained unclear.
In C. elegans, a few examples of post-embryonic dendrite remodeling have been demonstrated. During the dauer larval stage, the unbranched dendrites of IL sensory neurons are converted into highly branched dendritic arbors [65]. The mechanosensory neurons PVD and FLP elaborate a branched dendritic arbor beginning in young larvae that becomes increasingly complex as development progresses [20]. Age-dependent changes in dendrite arborization have been carefully examined in PVD [66]. In older adults, the PVD arbor becomes increasingly hyper-branched and disorganized, reminiscent of some of the age-dependent changes in URX that we observe in mapk-15 mutants [66]. Throughout development, the PVD dendrite arbor is highly dynamic, and its branching pattern reflects the net result of many outgrowth and retraction events [17,20,66]. Interestingly, in both younger and older animals (L4 larvae and 5-day adults), PVD dendrite outgrowth is slightly favored over retraction [66]. The accumulated effects of this imbalance may explain its age-dependent hyper-branching. This example would be consistent with a dynamic equilibrium model for control of dendrite length in URX and, possibly, in other systems.
Dendrite morphogenesis has been studied in C. elegans in the context of branching (PVD, FLP, and IL neurons) and DYF-7-mediated retrograde extension (ciliated sensory neurons of the amphid) [12–23,30,65]. Here, we show that URX offers a complementary model of dendrite morphogenesis that provides new mechanistic insights. In particular, our results show that, in addition to established pathways that promote dendrite growth and branching, there are also yet-unidentified pathways that constrain excessive dendrite growth throughout the life of the animal. It will be important to determine whether these age-dependent changes reflect a shift in the rates of addition and removal of dendrite material, and how they relate to alterations in dendrite lengths observed in the aging mammalian brain.
The following DNA constructs and transgenes were used: ynIs78[flp-8pro:GFP] X and flp-8pro [36]; hsp-16.41pro and hsp-16.2pro (from pPD49.83 and pPD49.78, Andrew Fire); mapk-15pro (1.7 kb upstream of coding sequence); gcy-32pro [67]; mapk-15 (genomic fragment); MAPK-15(K42A) (codon change generated via overlap extension PCR from mapk-15 genomic fragment); MAPK-15(ΔLIR) (deletion of codons 342–345 (encoding YEMI) generated via PCR from mapk-15 genomic fragment); gcy-32pro, gcy-37pro, GCY-35-GFP, GCY-35-HA-GFP [33]; GCY-35(D473A)-GFP (codon change generated via overlap extension PCR from GCY-35-GFP fragment), superfolderGFP [68]; gcy-37pro and yellow cameleon YC2.60 [33,69]. SuperfolderGFP is a robustly folding GFP variant that was used for convenience [68]. The mapk-15 genomic region likely contains some transcriptional regulatory sequences, as the flp-8pro:superfolderGFP-mapk-15 construct shows weak expression in an unidentified head neuron not seen with flp-8pro:GFP alone; however, expression in this cell does not interfere with imaging and does not appear to be sufficient for rescue based on mosaic analysis and the phenotype observed in strains bearing hsp:mapk-15 in the absence of heat shock. Strains are listed in S1 Table.
L4 stage animals were mutagenized using 70 mM ethyl methanesulfonate (EMS, Sigma) at approximately 22°C for 4 hours [42]. Nonclonal F2 progeny were examined on a Nikon SMZ1500 stereomicroscope with an HR Plan Apo 1.6x objective, and animals with aberrant dendrite morphologies were recovered to individual plates. Mutant identification was performed by one-step pooled linkage analysis [70] and sequence variants were analyzed with Cloudmap [71]. Identified mutations are listed in S1 Table.
Animals bearing hsp:mapk-15 constructs (S1 Table) were subjected to heat shock by incubating them at 34°C for 30 min on standard growth medium in the presence of bacterial food. Even in the absence of heat shock, these constructs mildly rescued dendrite defects, possibly due to "leaky" expression (compare Fig 3F (2Ad) with Fig 3G (-h.s.)).
Animals were mounted on agar pads and immobilized with sodium azide. Image stacks were collected on a DeltaVision Core deconvolution imaging system (Applied Precision) with an InsightSSI light source; a UApo 40x/1.35 NA oil immersion objective, PlanApo 60x/1.42 NA oil immersion objective, or UPlanSApo 100x/1.40 NA oil immersion objective; the Live Cell Filter module; and a Photometrics CoolSnap HQ2 CCD camera (Roper Scientific). Image stacks were acquired and deconvolved with Softworx 5.5 (Applied Precision). Maximum projections were generated with contiguous optical sections in ImageJ (NIH), then linearly adjusted for brightness in Adobe Photoshop. Multicolored images were created by placing each channel in a separate false-colored screen layer in Photoshop. Figures were assembled in Adobe Illustrator.
Dendrite lengths were measured using the segmented line tool in ImageJ (NIH). The dendrite was traced from the point where it joins the cell body to the point where it ends at the nose, then normalized by the distance from the cell body to the nose to account for variance in the size of the animal. The data were initially recorded in Microsoft Excel or Apple Numbers, and statistical analysis was performed using R version 3.4.2 (https://www.r-project.org/). To test statistical significance, the Mann-Whitney U test was chosen to compare differences between independent samples that are not necessarily normally distributed.
A gcy-37pro:YC2.60 (yellow cameleon 2.60) transgene was used for ratiometric imaging of relative calcium concentration in URX cell bodies [33,69]. L4 animals expressing the sensor were picked 48h before imaging. Calcium imaging experiments were performed as described previously [33,69] using an inverted microscope (Axiovert, Zeiss) equipped with a 40x C-Apochromat lens (Zeiss), Optosplit II beam splitter (Optical Insights), Evolve Delta camera (Photometrics) and MetaMorph acquisition software (Molecular Devices). Photobleaching was minimized using a 2.0 optical density filter. Animals were glued to agarose pads (2% agarose in M9 buffer, 1 mM CaCl2) using Dermabond tissue adhesive (Ethicon) with the nose immersed in a mix of bacterial food (E. coli OP50) and M9 buffer. To deliver gas stimuli, glued animals were placed under a Y-shaped microfluidic chamber with inlets connected to a PHD 2000 Infusion syringe pump (Harvard Apparatus) running at a flow rate of 2.5 ml/min. An electronic valve system placed between the syringes and the microfluidic chamber allowed switching between two different gas mixtures in a controlled manner at pre-specified time intervals. URX calcium responses were recorded at 2 frames/s with an exposure time of 100 ms. Image and statistical analysis was performed using Neuron Analyser, a custom-written MATLAB program (code available at https://github.com/neuronanalyser/neuronanalyser). Statistical comparisons of calcium responses were done using a Mann–Whitney U test.
Locomotion was assayed as described previously [34], with slight alterations. L4 animals were picked 48 h before the assay. Two days later, 20–30 adult hermaphrodites were transferred to nematode growth medium plates containing low peptone (5% of standard bactopeptone concentration) that had been seeded 16–20 h earlier with 20 μl of E. coli OP50 grown in 2× TY medium (per liter, 16 g tryptone, 10 g yeast extract, 5 g NaCl, pH 7.4). To deliver gas stimuli, animals were placed under a 1 cm × 1 cm × 200 μm deep polydimethylsiloxane chamber with inlets connected to a PHD 2000 Infusion syringe pump (Harvard apparatus). Humidified gas mixtures were delivered at a flow rate of 3.0 ml/min. We recorded movies using FlyCapture (Point Gray) on a Leica M165FC dissecting microscope with a Point Gray Grasshopper camera running at 2 frames/s. Movies were analyzed using Zentracker, a custom-written MATLAB software (code available at https://github.com/wormtracker/zentracker). Speed was calculated as instantaneous centroid displacement between successive frames. Locomotion assays were done in triplicate on at least two independent days. For statistical comparisons, we chose time intervals where we expected the speed changes to have plateaued, that is, with a delay with respect to the timing of the switch in O2 concentration. For the intervals of interest, we determined independent per-subject means derived from individuals flagged as continuously valid for at least 10 s during the interval. We considered all individuals in the field of view as valid except those in contact with other animals and those that were off the food lawn or less than half a body-length from the border. Following these criteria, each individual was sampled at most once per interval; n indicates the minimum number of valid samples obtained per interval. Statistical comparisons for speed assays used a Mann-Whitney U test.
|
10.1371/journal.ppat.1002793 | The Lectin Pathway of Complement Activation Is a Critical Component of the Innate Immune Response to Pneumococcal Infection | The complement system plays a key role in host defense against pneumococcal infection. Three different pathways, the classical, alternative and lectin pathways, mediate complement activation. While there is limited information available on the roles of the classical and the alternative activation pathways of complement in fighting streptococcal infection, little is known about the role of the lectin pathway, mainly due to the lack of appropriate experimental models of lectin pathway deficiency. We have recently established a mouse strain deficient of the lectin pathway effector enzyme mannan-binding lectin associated serine protease-2 (MASP-2) and shown that this mouse strain is unable to form the lectin pathway specific C3 and C5 convertases. Here we report that MASP-2 deficient mice (which can still activate complement via the classical pathway and the alternative pathway) are highly susceptible to pneumococcal infection and fail to opsonize Streptococcus pneumoniae in the none-immune host. This defect in complement opsonisation severely compromises pathogen clearance in the lectin pathway deficient host. Using sera from mice and humans with defined complement deficiencies, we demonstrate that mouse ficolin A, human L-ficolin, and collectin 11 in both species, but not mannan-binding lectin (MBL), are the pattern recognition molecules that drive lectin pathway activation on the surface of S. pneumoniae. We further show that pneumococcal opsonisation via the lectin pathway can proceed in the absence of C4. This study corroborates the essential function of MASP-2 in the lectin pathway and highlights the importance of MBL-independent lectin pathway activation in the host defense against pneumococci.
| Streptococcus pneumoniae is a major human pathogen that causes pneumonia, septicemia and meningitis. The host defense against pneumococci is largely dependent on complement, a system of blood proteins which, when activated, attach to bacteria, targeting them for clearance by phagocytes. There are three routes of complement activation: The classical, lectin and alternative pathways. Limited information is available on the roles of the classical and alternative pathways in fighting pneumococci; the role of the lectin pathway has escaped the attention of previous research. This work demonstrates that the lectin pathway is critical in fighting pneumococcal infection. Of the five different lectin pathway recognition molecules in human serum, only L-ficolin and collectin 11 activate complement on pneumococci. Human mannose-binding lectin (MBL), the best-known lectin pathway pattern recognition molecule, has no role whatsoever in fighting pneumococci. Similarly, in mouse serum, only ficolin A and collectin 11 drive complement activation on S. pneumoniae. Hence, MBL deficient mice are not compromised in pneumococcal infection, while ficolin A deficient mice and mice deficient of the key lectin pathway enzyme MBL-associated serine protease-2 (MASP-2) are exquisitely susceptible to infection. This work explains why MBL deficiency, the most frequent hereditary immune deficiency, does not predispose to pneumococcal disease.
| Streptococcus pneumoniae infection is a major cause of pneumonia, otitis media, septicemia and meningitis [1], [2]. Complement–driven opsonophagocytosis is a prominent feature of the host response to pneumococcal infections, [3].
Complement provides protection against invading microorganisms through both antibody-dependent and -independent mechanisms. It mediates many cellular and humoral interactions within the immune response, including chemotaxis, phagocytosis, cell adhesion, and B-cell differentiation. Three different pathways initiate the complement cascade, which are known as the classical, alternative and lectin pathways.
In the classical pathway, the recognition subcomponent C1q binds to a variety of targets - most prominently immune complexes - to initiate the step-wise activation of associated serine proteases, C1r and C1s. Activated C1s cleaves C4 into C4a and C4b and then cleaves C4b-bound C2 to generate the C3 convertase, C4b2a, which converts the abundant plasma protein C3 into C3a and C3b; C3b is the major opsonin of the complement system. Accumulation of C3b in close proximity to the C4b2a complex leads to the formation of the C5 convertase, C4b2a(C3b)n, which initiates the terminal pathway of complement activation.
In the alternative pathway, spontaneous low-level hydrolysis of C3 leads to deposition of C3b on cell surfaces, triggering complement activation on foreign cells. Host cells are protected by surface regulatory proteins that suppress complement activation.
Like the alternative pathway, the lectin pathway may be activated in the absence of immune complexes. Activation is initiated by the binding of a multimolecular lectin pathway activation complex to pathogen-associated molecular patterns (PAMPs), mainly carbohydrate structures present on microorganisms or aberrant glycocalyx patterns on apoptotic, necrotic, malignant or oxygen-deprived cells [4], [5]. Rodents have at least four circulating lectin pathway recognition molecules, with differing, but overlapping, carbohydrate specificities; two mannan-binding lectins (MBL-A and MBL-C), collectin-11 (CL-11) and ficolin A (Fcna) [6]. A second murine ficolin, Fcnb, associated with monocyte and macrophage cell surfaces does not activate complement in mice, but may act as a lectin pathway recognition molecule in rats [7]. Humans have a single MBL (the product of MBL2; MBL1 is a pseudogene), CL-11 (collectin kidney 1, CL-K1) and three ficolins, FCN1 (M-ficolin), FCN2 (L-ficolin) and FCN3 (H-ficolin) [5], [8], [9].
These recognition molecules form complexes with three serine proteases, MASP-1, -2 and -3 (MBL-associated serine proteases 1, 2 and 3). The recognition molecules also interact with MAp19 and MAp44 (alias MBL/ficolin-associated protein 1), which are non-enzymatic, truncated alternative splice products of the MASP2 and MASP1/3 genes, respectively. Both truncated gene products lack the serine protease domain and may regulate lectin pathway activation by competing for the binding of MASPs to the carbohydrate recognition molecules [4], [10]–[15].
Of the three MASPs, only MASP-2 is required and essential to form the lectin pathway C3 and C5 convertases (C4b2a and C4b2a(C3b)n) [6], [10], [16], [17].
Like C1s, activated MASP-2 cleaves C4 and C4b-bound C2, generating C4b2a, the classical and lectin pathway C3 convertase. Neither MASP-1 nor MASP-3 can cleave C4 and therefore cannot compensate for the absence of MASP-2. Thus, formation of the lectin pathway C3 and C5 convertase complexes is impossible in absence of MASP-2 [6]. MASP-1 appears to facilitate lectin pathway activation by either direct cleavage of complex-bound MASP-2 or cleavage of C4b-bound C2, but MASP-1 cannot drive lectin pathway activation in the absence of MASP-2, as MASP-2 is required to initiate the formation of the lectin pathway convertases by the cleavage of C4 [6], [18], [19]. Interestingly, recent work demonstrated that MASP-1 (and possibly MASP-3) play a key role in the maturation and initiation of the alternative activation pathway [20], [21].
Infection studies using mice with targeted deficiencies in one or more complement components have provided evidence for the roles of the classical and alternative pathways in protection against S. pneumoniae. C1q deficient mice were found to be more susceptible to infection with S. pneumoniae than WT mice, indicating that the classical pathway has a protective role. The alternative pathway was also found to have a protective role against S. pneumoniae, but to a lesser extent than the classical pathway. Mice deficient in factor B had a significantly higher level of bacteria in lungs and in blood in comparison to their WT controls [22].
The contribution of the lectin activation pathway towards the defense against S. pneumoniae infection had not been assessed so far, mainly due to the lack of appropriate mouse models of total lectin pathway deficiency.
Using MASP-2 deficient mice, completely devoid of the ability to form lectin pathway C3 and C5 convertases, this report demonstrates that lectin pathway activation provides a critical degree of protection against S. pneumoniae. We identified mouse ficolin A and CL-11, but not MBL-A or MBL-C, to be the critical pattern recognition molecules that initiate complementactivation via the lectin pathway on the surface of this pathogen.
Complement deficient sera were used to determine which components contribute to C3b opsonisation of S. pneumoniae. C3b deposition was assayed by ELISA using formalin-fixed bacteria and by FACS analysis using live bacteria. MASP-2 deficiency leads to a total loss of C3b deposition (fig. 1a–e). Using a simple end-point ELISA, C3b deposition appears to be unaffected by factor B deficiency (fig. 1b). However, the conversion of C3 is slower in factor B deficient serum (t1/2≈28 min) than in WT serum (t1/2≈8 min), indicating that the alternative pathway amplification loop contributes to the C3 turnover (fig. 1c). Serum from MBL-null mice (deficient in both MBL-A and MBL-C) opsonised the bacteria as efficiently as WT serum, whereas ficolin A deficiency resulted in impaired C3b deposition. Ficolin B was not detectable in sera of WT and ficolin A deficient C57BL/6 mice (data not shown). Serum deficient in MBL-A, MBL-C and ficolin A produced similar results to those seen using ficolin A deficient serum, suggesting that the remaining lectin pathway recognition component CL-11 may drive residual lectin pathway activation on S. pneumoniae (fig. 1b). In addition to the recent publication by Hansen et al. (2010) [9] which demonstrated MASP-1 binding to CL-11, we have shown that recombinant human CL-11 binds recombinant MASP-2 to form a lectin pathway activation complex under physiological conditions (fig. S1).
Using C4 deficient mouse serum, the absolute amount of C3b deposited on S. pneumoniae was approximately half that observed using WT serum (fig. 1b, c, e & f) and the rate of conversion was lower (T1/2≈30 min in C4 deficient serum; 8 min in WT serum; fig. 1c). C4 is an integral part of the classical and lectin pathway C3 convertase, C4b2a, suggesting that the residual C3b deposition seen in C4 deficient serum is a result of either alternative pathway activation [23], [24] or the recently reported lectin pathway-specific C4-bypass [6]. Since (i) all experiments reported here were carried out at low serum concentrations (1.25%, fig. 1b; 2.5%, fig. 1c and; 5%, fig. 1e & f) where the alternative activation pathway is dysfunctional, and (ii) the activation of C3 on the surface of S. pneumoniae is absent in MASP-2 deficient mice (see fig. 1), we conclude that the C3 deposition on S. pneumoniae in C4 deficient serum is mediated by the MASP-2 dependent C4-bypass activation route. Likewise, inhibition of MASP-2 activity with the anti-MASP-2 mAb AbD04211 abolished C3b deposition on S. pneumoniae incubated with C4 deficient serum, supporting the hypothesis that the MASP-2 dependent C4-bypass plays a significant role in the opsonisation of S. pneumoniae (fig. 1f).
Similar results were obtained using human serum from a donor with a complete deficiency of both C4 genes, C4A and C4B: C4 deficiency halved C3b deposition, although the EC50 value was unaffected (≈0.05% for all sera assayed; fig. 1g). Human MBL deficiency had no impact on the deposition of C3b on the surface of S. pneumoniae (see fig. 1g). In contrast, analysis of 47 samples of normal human serum (NHS) revealed a significant correlation between serum L-ficolin concentration and the level of C3b deposition S. pneumoniae (fig. 2f; p<0.0001; Fisher transformation of Pearson's correlation coefficient). The test group included 8 MBL deficient sera (YO/YO or YO/XA genotypes, see [25]), none of which showed abnormally low C3b deposition, providing further evidence that that human MBL does not contribute to complement activation on S. pneumoniae.
Antibodies directed against the larger C4 activation fragments, C4b and C4c, failed to detect any C4 deposition on the surface of S. pneumoniae exposed to normal human serum (fig. 2a & c). However, when using an antibody directed against C4dg, the haemolytically inactive final degradation product of C4, an abundant deposition of covalently bound C4dg was detected (fig. 2b & d), supporting the hypothesis that S. pneumoniae sequester host complement control proteins to accelerate the degradation and inactivation of C4b [26], [27].
A series of solid-phase binding experiments were performed to identify lectin pathway recognition molecules that bind to S. pneumoniae. Microtitre plates coated with S. pneumoniae D39 and control substrates were used to capture lectin pathway recognition molecules from WT mouse serum and NHS (fig. 3). As anticipated from the C3b deposition assays (fig. 1b & g) and previous reports [28], MBL does not recognize the bacteria, whereas murine Fcna (fig. 3b) and CL-11 (fig. 3c), and human L-ficolin and CL-11 (alias CL-K1) (fig. 3f) do. Direct binding to the bacterial surface was confirmed using recombinant human CL-11 and mouse ficolin A (data not shown).
Nine other strains of S. pneumoniae representing three additional serotypes (18, 6B and 3) were tested for binding of human and murine recognition molecules. With human serum the results were remarkably consistent: Of the 5 different lectin pathway specific carbohydrate recognition subcomponents, only L-ficolin and CL11 (alias CL-K1) bound to the bacteria (table S1). Most of the strains showed little or no binding to murine MBL-C and none bound MBL-A. All of the strains tested bound murine CL-11 with similar affinities, and all strains tested bound Ficolin A. MBL-C and Ficolin A binding differed amongst strains of the same serotype, indicating that capsular polysaccharide is unlikely to be the main determinant of recognition by lectin pathway recognition subcomponents.
The inability of MASP-2 deficient mouse serum to opsonise S. pneumoniae with C3b led to defective phagocytosis in vitro (fig. 4). S. pneumoniae D39 were opsonised with WT or MASP-2 deficient serum and mixed with freshly isolated human peripheral blood polymorphonuclear leukocytes (PMN). Bacteria opsonised with WT serum were internalised (fig. 4a & c), whereas bacteria opsonised with MASP-2 deficient serum were excluded from the PMN, indicative of defective uptake and phagocytosis (fig. 4b). Samples taken from the mixture over a period of 2 hr were plated on blood agar and viable S. pneumoniae counted (fig. 4d). Pneumococci mixed with WT serum are efficiently killed by PMN, while those opsonised with MASP-2 deficient serum survive as well as non-opsonised controls. In addition, 20% MASP-2 sufficient (WT) serum from non-immunized mice has no bacteriolytic activity on S. pneumoniae. This observation was confirmed using 50% WT mouse and human serum (data not shown).
We used MASP-2 and Fcna deficient mice to determine to what extent the lectin pathway contributes to host defense against S. pneumoniae in vivo. In contrast to MBL-null [29] and Fcna deficient mice [30] (both of which having lectin pathway activation complexes formed with the remaining recognition molecules in their blood), MASP-2 deficient animals are unable to form the lectin pathway C3 and C5 convertases [6].
Ten-week old C57BL/6Masp2−/− mice and WT littermates were infected with S. pneumoniae D39 by intranasal inoculation and the course of the infection monitored for one week (fig. 5). 75% of the WT mice survived the infection, compared with only 20% of the MASP-2 deficient mice (p = 0.0006).
During the first 12 hours after infection, the WT mice showed signs of an initial clinical response (hunching), while the MASP-2−/− mice appeared unaffected. In both groups, the first animals reached the endpoint (severe lethargy) after 48 hours, and survival dropped further until 72 hrs after infection. All those animals alive after 72 hr survived (see fig. 5a). In another series of experiments, animals were sacrificed 12, 24 and 48 hr post infection to determine counts of viable S. pneumoniae in lung homogenate and blood. In MASP-2 deficient mice, CFUs in the lung (fig. 5c) and blood (fig. 5e) were significantly higher than in WT mice and rose progressively during the first 48 hr, at which point the experiment was stopped due to the high mortality in the MASP-2 deficient group. In WT mice that survived the first 72 hr, bacteria were progressively cleared from the both the lungs and blood. Similar results were obtained with C57BL/6Fcna−/− mice; 70% of the WT littermates survived, whereas 80% of the ficolin A deficient mice succumbed to the infection (fig. 5b). Lung infection and bacteraemia are shown in fig. 5d and f.
In contrast, intranasal infection of C57BL/6 MBL-null (MBL-A and MBL-C double deficient) mice resulted in neither significantly increased mortality nor compromised bacterial clearance mice compared to sex and age matched C57BL/6 WT controls (fig. S2).
Quantitative RT-PCR analysis of cytokine and chemokine expression in lungs from infected animals showed that the onset of the inflammatory response was broadly similar in WT, ficolin A and MASP-2 deficient mice. After 12 hr, however, the pro-inflammatory TNFα response increased more rapidly in the lectin pathway deficient mice, and levels were significantly greater at 24 hr and 48 hr than in the WT mice. In MASP-2 deficient mice, the IL6 response was also elevated. The INFγ response was significantly greater in the ficolin A deficient mice. In the ficolin A and MASP-2 deficient mice, MIP-2 (CLCX2) expression persists at 48 hr, indicating on-going macrophage activation at a time when the response is abating in the WT mice (fig. S3).
MASP-2 deficiency can be simulated in WT mice using a mAb that specifically inhibits MASP-2. A single i.p. dose of 0.6–1.0 mg/kg body weight leads to a loss of ≥90% of lectin pathway activity for up to 7 days [6]. WT mice treated with this mAb prior to infection with S. pneumoniae had significantly greater mortality and higher bacteraemia than untreated controls (fig. 6). Antibiotic treatment (Ceftriaxone, 20 mg per kg body weight i.p. 12 h before infection and every 12 h thereafter) resulted in complete protection against mortality in both Ab-treated and non-treated groups. These results suggest that increased susceptibility to S. pneumoniae in MASP-2 deficient mice is a direct result of the loss of MASP-2 driven lectin pathway activity, rather than an indirect result of MASP-2 deficiency on the development of the animal's immune response.
Complement-dependent opsonophagocytosis is a key feature of the host defense against S. pneumoniae. Thirty years ago, using a guinea pig model of pneumococcal bacteraemia, it was shown that the clearance of IgG and IgM opsonized pneumococci from the circulation is entirely dependent upon complement; animals deficient of C4, or depleted of complement using cobra venom factor (CVF), fail to clear the bacteria [31]. More recently, Brown and co-workers [22] used mice with engineered genetic deficiencies to demonstrate the importance of C1q, C4 and C3 in the defense against S. pneumoniae. Based on the observation that C1q deficient mice are as susceptible to infection as C4 deficient mice, the authors concluded that activation of the classical pathway is the predominant mechanism for complement-mediated opsonization and phagocytosis of S. pneumoniae and that the lectin pathway (which according to the present text book view requires C4 to work) plays a negligible role [22].
Here we analyzed experimental S. pneumoniae infection in the first available model of lectin pathway deficiency and reached the conclusion that the lectin pathway promotes innate resistance against pneumococcal infection in the non-immunized host. The results presented here show that MASP-2 deficient mice (which can still activate complement via the classical and alternative pathways [6]) are severely compromised in their ability to survive S. pneumoniae infection (fig. 5). Survival times and mortality rates were similar to those reported for C1q and C4 deficient mice, using the same strains of bacteria and mice, and the same dose and route of infection [22].
Neither the classical, nor the alternative activation pathway could compensate for the loss of lectin pathway mediated C3 opsonization of S. pneumoniae. In contrast, the absence of C1q had no effect on C3 opsonization of these bacteria in vitro (fig. 1) implying that C1q may contribute to bacterial clearance in a process independent of direct C3b or iC3b deposition on the bacterial surface.
Deficiency of factor B, a component of the alternative pathway C3 convertase led to a significantly slower C3 turnover (fig. 1C), This is probably due to the loss of the alternative pathway amplification loop, a positive feedback mechanism that amplifies C3 activation via of all three pathways, which may account for the reported susceptibility of factor B deficient mice to S. pneumoniae infection (Brown et al., 2002).
As previously reported by others [26], [32], we were unable to detect any C4b or C4c on the surface of S. pneumoniae opsonized with normal human serum (fig. 2). However, the bacterial surface was abundantly decorated with C4dg, the final product of C4 decay. This finding strongly supports the hypothesis that S. pneumoniae avoids the accumulation of active C4b by sequestrating complement regulatory proteins from host serum to accelerate the breakdown of C4b, thus preventing the formation of the C3 convertase C4b2a on the pathogen, rather than by preventing C4 binding. Recent work suggests that the pneumococcal virulence factors PspA and PspC are responsible for recruiting factor H and C4-binding protein from host plasma, both of which accelerate the factor I-mediated breakdown of C4b to C4dg [26], [27].
C3 deposition on pneumococci was impaired, but not completely blocked, by C4 deficiency in both mice (fig. 1b–f) and humans (fig. 1g). In the mouse, C4 deficiency led to a significantly slower C3 turnover and in both species the absolute amount of C3 deposited on the bacteria was approximately half of that observed using WT serum. The residual C3 deposition in C4 deficient murine serum could be inhibited using a monoclonal antibody directed against MASP-2 (fig. 1f), indicating that the MASP-2-dependent C4-bypass is active on S. pneumoniae [6]. Nevertheless, C4 deficient mice have an increased susceptibility to S. pneumoniae infection ([22]; our unpublished data), indicating that the MASP-2-dependent C4-bypass only partially compensates for C4-dependent lectin pathway activation in this setting. The C4-bypass activation of the lectin pathway was shown to play a significant physiological role in ischemia-reperfusion injury; MASP-2 deficient mice are protected from reperfusion injury following myocardial ischemia, whereas C4 deficient mice are not [6], [33].
We hypothesize that the rapid degradation of C4b on the bacterial surface seriously compromises the ability of the classical pathway to form a C3 convertase and thus opsonize S. pneumoniae with C3b, while the lectin pathway is still able to opsonize pneumococci with C3b via the C4-bypass, which provides a physiologically relevant degree of compensation for the impaired C4-dependent activation of C3 [6]. The loss of lectin pathway activity caused by MASP-2 deficiency or MASP-2 inhibition would remove a critical degree of C3 opsonization of S. pneumoniae in naive mice and hence explain the phenotype of compromised pneumococcal clearance in MASP-2 deficient or MASP-2 depleted mice. The rapid conversion of C4b to C4dg on the pathogen surface appears to be a feature of S. pneumoniae. This would explain why the absence of lectin pathway activity renders the host more susceptible to infections with this particular pathogen, whilst no increased predisposition to, or severity of, infection with other major pathogens, e.g. Pseudomonas aeruginosa and Neisseria meningitidis were observed in MASP-2 deficient mice ([34]; our unpublished data).
In murine serum, it is predominantly ficolin A and CL-11 that bind to S. pneumoniae (fig. 3 and table S1). The binding of both lectin pathway recognition molecules to the bacterial surface was confirmed using recombinant human CL-11 and recombinant murine ficolin A. Murine MBL-A does not bind to any of the S. pneumoniae strains studied here (covering 4 serotypes), and MBL-C bound weakly or not at all (table S1). It was therefore not surprising that the presence or absence of both MBL-A and MBL-C in C57BL/6 mice had no impact on overall survival in our model of S. pneumoniae D39 infection. In contrast, following pneumococcal infection, C57BL/6Fcna−/− mice were severely compromised with a significantly higher degree of mortality and higher bacterial loads in blood and lung tissue than WT controls. This phenotype underlines our in vitro results and indicates that ficolin A (but not MBL-A or MBL-C) is a key recognition component of the lectin activation pathway in the innate host defense against pneumococcal infection. Interestingly, serum from mice deficient in all lectin pathway recognition components except CL-11 still deposits C3b on the surface of S. pneumoniae (fig. 1b). Since mouse CL-11 binds strongly to the surface of S. pneumoniae (see fig. 3c) and since we show that CL-11 also forms complexes with the lectin pathway effector enzyme MASP-2 (fig. S1), we conclude that CL-11 acts synergistically with ficolin A as an initiator of lectin pathway activation following binding of specific PAMPs on the pneumococcal surface.
The situation is similar in humans; only L-ficolin and CL-11 recognize S. pneumoniae (fig. 3e & f). As previously reported by others [28], we found no binding of MBL to any of the pneumococcal strains (table S1), and there was no indication that MBL deficiency leads to defective C3b deposition on the bacteria (fig. 1g&h). As MBL deficiency is the most common hereditary complement deficiency in humans, affecting as many as 1 in 10 of the population [35], there has been much interest in the possibility of an association between MBL deficiency and infectious disease. In the case of S. pneumoniae, the results of association studies have been largely negative or inconclusive, with two of the largest studies finding no association between the risk of community-acquired pneumonia and MBL deficiency [36], [37]. Genetic variations of the L-ficolin gene are more subtle, with no complete functional deficiency in adults reported to date. Furthermore, there is no apparent association between polymorphisms in FCN2 that lead to low levels of L-ficolin and pneumococcal disease [38], indicating that even low levels of L-ficolin and/or the presence of CL-11 are sufficient to mount a robust immune response against S. pneumoniae.
We have previously shown that antibodies to MASP-2 can block lectin pathway driven inflammation and limit tissue loss in ischaemic pathologies [6], suggesting therapeutic utility for anti MASP-2 antibody therapy. Our experiments show that such treatment may increase susceptibility to S. pneumoniae infection in naive mice. Pneumococcal vaccination history or immune status may need to be considered prior to initiating anti MASP-2 treatment in patients. Alternatively, since the increased susceptibility was completely reversed by concurrent treatment with ceftriaxone, MASP-2 antagonists should be safe to use with appropriate prophylactic antibiotic treatment.
The results of this study call for a revision of the previously published conclusion by Brown et al. (2002) [22] that the lectin pathway of complement activation is not a major player in the host response to S. pneumoniae infection. This conclusion led the same research team to exclude any involvement of the lectin pathway in the clearance of pneumococci in subsequent publications. When studying the impact of human C2 deficiency on C3b opsonization and phagocytosis of S. pneumoniae [39], no consideration was given to the fact that C2 deficient individuals are not only deficient of the classical activation pathway, but also of C3 and C5 convertases formed by the lectin pathway. The results presented here, however, strongly suggest that in none-immune sera, it is the loss of lectin pathway functional activity that accounts for the loss of C3b/iC3b opsonization of pneumococci.
Finally, we describe clear evidence of lectin pathway activation in a physiological context in the absence of a discernable contribution by MBL. Thus, the prevailing view that MBL is the predominant initiator of lectin pathway activation may need to be revisited.
All animal experiments were authorized by the UK Home Office (Animals Scientific Procedures Act 1986; Home Office project licence 80/2111) and approved the University of Leicester animal welfare committee. Every effort was made to minimize suffering and mice were humanely culled if they became lethargic during infection experiments.
Unless otherwise stated, all reagents were obtained from Sigma-Aldrich. PSA, a polysaccharide produced by Aerococcus viridans that binds FCN3 was prepared as previously described [40]. AbD04211, a recombinant mAb that potently inhibits mouse MASP-2, has been described previously [6].
S. pneumoniae serotype 2 strain D39 was obtained from the National Collection of Type Cultures, London, United Kingdom (NCTC 7466). Bacteria were identified as pneumococci by Gram staining, catalase testing, alpha-hemolysis on blood agar plates, and determination of optochin sensitivity. Serotypes were confirmed by the Quellung reaction. To obtain pneumococci grown in vivo, bacteria were cultured and passaged through mice as described previously [41] and subsequently recovered and stored at −70°C. When required, suspensions were thawed at room temperature and bacteria were harvested by centrifugation before re-suspension in sterile PBS. Nine other clinical isolates of S. pneumoniae were kindly provided by Prof. Herminia de Lancastre, Instítuto de Tecnologia Química e Biológica, Oeiras, Portugal [42].
Mice deficient in MASP-2 and C4 have been described elsewhere [6], [43]. Ficolin A deficient mice were generated by targeting Fcna using a conventional replacement vector, as described elsewhere [44]. MBL-null mice were purchased from MMRRC, Bar Harbor, Maine and crossed with Fcna−/− mice to produce a strain deficient in all three components. Complement deficient mouse strains were backcrossed with C57/BL6 mice for at least ten generations before use. Blood was collected from these animals and from WT C57/BL6 mice by cardiac puncture, serum prepared, aliquoted and stored at −80°C. C1q deficient and factor B deficient murine plasma was kindly provided by Dr. Marina Botto, Imperial College London.
Human blood was obtained from healthy adult donors who had given written, informed consent, as required by the local ethics committee. L-ficolin concentration was determined as previously described [45]. Genomic DNA was prepared using a kit (Promega) and MBL2 A/O and X/Y genotypes were determined using fluorescent hybridization probes in a Roche LightCycler [46]. Serum from an individual with a complete deficiency of both C4 genes, C4A and C4B, has been described previously [47].
Nunc Maxisorb microtiter plates were coated with 100 µl of the following reagents: 10 µg/ml mannan (a control for MBL binding), 10 µg/ml zymosan (a control for CL-11 binding), 10 µg/ml N-acetylated BSA (Promega; a control for ficolin A binding), 5 µg/ml of the FCN2-specific mAb GN4, 10 µg/ml PSA, or formalin-fixed S. pneumoniae D39 (OD550 nm = 0.6) in coating buffer (15 mM Na2CO3, 35 mM NaHCO3, pH 9.6). Wells were blocked with 250 µl of 1% (w/v) BSA in TBS buffer (10 mM Tris-HCl, 140 mM NaCl, pH 7.4), then washed three times with 250 µl of TBS with 0.05% Tween 20 and 5 mM CaCl2 (wash buffer). Serial dilutions of serum in 100 µl of wash buffer were added and the plates incubated for 90 min at room temperature. Plates were washed as above and bound proteins detected using monoclonal rat anti-mouse MBL-A (Hycult), rat anti-mouse MBL-C (Hycult), rabbit anti-mouse ficolin-A, rabbit anti-human M-ficolin, rabbit anti-human L-ficolin, mouse anti-human H-ficolin, mouse anti-human CL-11 or rat anti-mouse CL-11 mAbs. Secondary antibodies were alkaline phosphatase-conjugates and bound antibody was detected using the colorimetric substrate p-nitrophenylphosphate (pNPP).
To measure C3 and C4 activation, Nunc MaxiSorb microtiter plates were coated with 100 µl of: 10 µg/ml mannan (Promega), or formalin-fixed S. pneumoniae D39 (OD550 nm = 0.6) in coating buffer. After overnight incubation, wells were blocked with 0.1% HSA in TBS then washed with wash buffer. Serum samples were diluted in BBS (4 mM barbital, 145 mM NaCl, 2 mM CaCl2, 1 mM MgCl2, pH 7.4), added to the plates and incubated for 1.5 h at 37°C. The plates were washed again, and bound C3b or C4b was detected using rabbit anti-human C3c (Dako) or rat anti mouse C4 (Hycult) followed by alkaline phosphatase-conjugated goat anti-rabbit IgG or alkaline phosphatase-conjugated rabbit anti rat IgG followed by the colorimetric substrate pNPP.
S. pneumoniae D39 were washed twice with TBS and re-suspended in BBS to a concentration of 106 cfu/ml. Two hundred µl of the bacterial suspension was mixed with 10 µl of NHS, WT mouse or complement deficient mouse serum and incubated for 1 h at 37°C. After opsonization, the bacteria were washed twice with wash buffer, re-suspended in wash buffer containing FITC conjugated rabbit anti-human C3c (Dako), mouse anti-human C4dg (Quidel) or mouse anti-human C4c (Santa Cruz) and incubated for 1 h on ice. Where non-conjugated primary antibodies were used, the pneumococci were washed twice and incubated for a further hour with FITC conjugated anti-mouse IgG (Dako). After two further washes, the bacteria were fixed using 1% w/v paraformaldehyde, and fluorescence intensity measured by FACS (Becton Dickinson FACS Calibur).
Polymorphonuclear leukocytes (PMN) were isolated from fresh human blood by discontinuous density gradient centrifugation using Histopaque-1119 and Histopaque-1077, according to the manufacturer's instruction. Leukocytes were washed twice with Hank's balanced salt solution (HBSS) containing 1.2 mM Ca2+ and 1.2 mM Mg2+, pH 7.4 (Invitrogen) and re-suspended in HBSS to a concentration of 107 cells/ml.
Killing of pneumococci by PMN was estimated by measuring the decrease in viable bacteria over time. Pneumococci were opsonized by incubation with 20% v/v WT or complement deficient murine serum at 37°C for 30 min. 1×106 PMNs were mixed with 105 pre-opsonized or non-opsonized S. pneumoniae D39 in a final volume of 250 µl in HBSS and incubated at 37°C on a rotary mixer. Samples were taken at 0, 30, 60, 120 and 240 min. To determine viable bacteria, samples were serially diluted in HBSS and plated onto blood agar plates.
For histological staining, 25 µl samples of the PMN/pneumococci mix were attached to glass slides by centrifugation at 1500×g for 3 min in a Cytospin 2 (Shanon). The slides were air dried for 15 min and stained using the RESTAIN Quick Diff. Kit (REAGENA). The slides were washed with water, air-dried and then mounted in DPX resin and photographed by bright-field microscopy.
For Transmission Electron Microscopy (TEM), a PMN/pneumococci mix was centrifuged for 5 min at 250× g. Cells were washed twice with 500 µl of 0.1M PBS (pH 7.2) and then fixed by re-suspension into 250 µl of 2.5% glutaraldehyde in 0.1 M PBS (pH 7.2). Fixed PMNs were then examined by TEM.
To assess whether serum complement alone can reduce the number of recoverable S. pneumoniae D39 through complement-mediated lysis, bacteria were incubated for 240 min. in 20% and 50% WT human and mouse sera at 37°C on a rotary mixer, samples taken at 0, 30, 60, 120 and 240 min and plated onto blood agar to determine viable bacteria.
Ten to twelve week old female MASP-2 and Fcna deficient mice, and their WT littermates were used. Mice were lightly anaesthetized with 2.5% (v/v) fluothane (AstraZeneca) over oxygen (1.5 to 2 litre/min), and 50 µl PBS containing 1×106 cfu of S. pneumoniae D39 was then administered into the nostrils of the mice. The inoculum dose was confirmed by viable count after plating on blood agar. For survival experiments, mice were monitored for clinical signs and culled when they became severely lethargic. This time was recorded as the survival time. To determine bacterial tissue counts, groups of mice were deeply anaesthetized at pre-chosen time intervals and blood was collected by cardiac puncture. Immediately afterwards, the mice were culled by cervical dislocation. Lungs were removed separately into 10 ml of sterile PBS, weighed, and then homogenized in a Stomacher-Lab blender (Seward Medical). Viable counts in lung homogenates and blood were determined by serial dilution in sterile PBS and plating onto agar plates supplemented with 5% (v/v) horse blood (Oxoid) and incubated for 18 h at 37°C in anaerobic conditions.
|
10.1371/journal.pntd.0001686 | Vasa-Like DEAD-Box RNA Helicases of Schistosoma mansoni | Genome sequences are available for the human blood flukes, Schistosoma japonicum, S. mansoni and S. haematobium. Functional genomic approaches could aid in identifying the role and importance of these newly described schistosome genes. Transgenesis is established for functional genomics in model species, which can lead to gain- or loss-of-functions, facilitate vector-based RNA interference, and represents an effective forward genetics tool for insertional mutagenesis screens. Progress toward routine transgenesis in schistosomes might be expedited if germ cells could be reliably localized in cultured schistosomes. Vasa, a member of the ATP-dependent DEAD-box RNA helicase family, is a prototypic marker of primordial germ cells and the germ line in the Metazoa. Using bioinformatics, 33 putative DEAD-box RNA helicases exhibiting conserved motifs that characterize helicases of this family were identified in the S. mansoni genome. Moreover, three of the helicases exhibited vasa-like sequences; phylogenetic analysis confirmed the three vasa-like genes—termed Smvlg1, Smvlg2, and Smvlg3—were members of the Vasa/PL10 DEAD-box subfamily. Transcripts encoding Smvlg1, Smvlg2, and Smvlg3 were cloned from cDNAs from mixed sex adult worms, and quantitative real time PCR revealed their presence in developmental stages of S. mansoni with elevated expression in sporocysts, adult females, eggs, and miracidia, with strikingly high expression in the undeveloped egg. Whole mount in situ hybridization (WISH) analysis revealed that Smvlg1, Smvlg2 and Smvlg3 were transcribed in the posterior ovary where the oocytes mature. Germ cell specific expression of schistosome vasa-like genes should provide an informative landmark for germ line transgenesis of schistosomes, etiologic agents of major neglected tropical diseases.
| Schistosomes, the blood flukes, are responsible for the major neglected tropical diseases termed schistosomiasis, which afflicts >200 million people in impoverished regions of the developing world. The genome sequence of these parasites has been decoded recently. The DEAD-box family is the largest of RNA helicase families. These enzymes play roles in RNA metabolic processes—transcription, pre-mRNA splicing, ribosome biogenesis, transport, initiation of translation, organelle gene expression, and RNA decay. Database searches indicated that S. mansoni has at least 33 DEAD-box helicases. A DEAD-box helicase known as Vasa is a determinant in germ line segregation and maintenance. Three schistosome DEAD-box RNA helicases exhibited Vasa-like sequences, and phylogenetic analysis confirmed they were members of the Vasa/PL10 DEAD-box subfamily. Quantitative real time PCR revealed that all three, termed Smvlg1, Smvlg2, and Smvlg3, were expressed in developmental stages of S. mansoni, with elevated expression in adult females, eggs, miracidia and sporocysts. Moreover, whole mount in situ hybridization analysis confirmed that these vasa-like genes were expressed in the schistosome ovary. These findings provide insights into the mechanisms underlying development of germ cell lines and gonads in schistosomes and have implications for understanding the role of RNA helicases, germ and stem cell biology.
| Schistosomiasis is considered the most important of the human helminthiases in terms of morbidity and mortality. It is endemic to 76 countries, affecting an estimated 200 million people with an additional 700 million people at risk of infection [1]–[3]. Draft genome sequences of all three of the major schistosome species are now available [4]–[7], underscoring a pressing need to develop functional genomic approaches to identify the role and importance of schistosome genes that might be targeted for development of novel anti-schistosomal interventions.
Retrovirus-mediated transgenesis is an established functional genomics approach for model species, e.g. [8]. It offers the means to establish gain- or loss-of-function phenotypes, can facilitate vector-based RNA interference, and represents a powerful forward genetics tool for insertional mutagenesis screens. Although murine leukemia virus (MLV) pseudotyped with vesicular stomatitis virus glycoprotein (VSVG) mediates somatic transgenesis in S. mansoni [9], vertical transmission of integrated transgenes has not been reported in schistosomes. A potential route to the germ line in schistosomes is to transduce the egg, and genetic manipulation of this developmental stage with MLV is feasible [10]. In addition, germ line transgenesis might be more readily monitored if schistosome germ cells could be localized and tracked in cultured forms of this pathogen.
Vasa, an adenosine 5′-triphosphate (ATP)-dependent DEAD-box RNA helicase, is an archetypal, metazoan germ cell specific marker [11], [12]. The highly conserved vasa gene has been studied widely as a germ cell marker in Caenorhabditis elegans, Drosophila melanogaster, Xenopus species, Danio rerio, Mus musculus and other species [13]–[22]. Reporter gene-vasa transgenic lines have been developed to isolate germ line cells [23]–[25]. Moreover, vasa-like homologues have been reported in Platyhelminthes including Macrostomum lignano, Schmidtea polychroa, several Dugesia species, Neobenedenia girellae, and Paragonimus westermani [11], [26]–[31].
Here, vasa-like genes in S. mansoni were identified and their developmental expression examined. Whole mount in situ hybridization and real-time quantitative PCR revealed specific expression in the schistosome ovary and strikingly elevated expression in the developing egg. Since Vasa is a reliable germ cell marker in other flatworms, we anticipate that reporter transgenes vectored by an integration competent vector such as MLV and driven by a vasa promoter might be of use to monitor transgene integration into schistosome germ cells.
Female Swiss-Webster mice infected with S. mansoni were obtained from the Biomedical Research Institute (BRI), Rockville, MD and housed at the Animal Research Facility of the George Washington University Medical Center, which is accredited by the American Association for Accreditation of Laboratory Animal Care (AAALAC no 000347) and has an Animal Welfare Assurance on file with the National Institutes of Health, Office of Laboratory Animal Welfare, OLAW assurance no. A3205-01. All procedures employed were consistent with the Guide for the Care and Use of Laboratory Animals. Maintenance of the mice and subsequent recovery of schistosomes were approved by the Institutional Animal Care and Use Committee (protocol approval no. A137) of the George Washington University.
Mice and Biomphalaria glabrata snails infected with the NMRI (Puerto Rican) strain of S. mansoni supplied by Dr. Fred Lewis, Biomedical Research Institute, Rockville, MD.
Eggs were recovered from livers of infected mice [32]. Aliquots of eggs were harvested while the remaining eggs were washed three times with 1×PBS supplemented with 2% penicillin, streptomycin, fungizone and transferred to sterile water under a bright light to induce egg hatching [33]. Newly hatched miracidia were snap frozen or transferred to sporocyst media and cultured as primary sporocysts for up to 15 days, as described [33]. In vitro laid eggs (IVLE) were obtained as described [34]. In brief, adult S. mansoni recovered from mice by portal perfusion were transferred to six-well culture plates with mesh inserts of 74 µm pore diameter, containing Basch's medium at 37°C under 5% CO2 (see [33]). Media were replaced twice a day. IVLE fell through the pores whereas the adult worms remained on the mesh. At 48 hours, the IVLE were harvested by filtering media containing the eggs through a six-well culture plate insert of 0.8 µm pore size (Greiner Bio-One, Art. No. 657638). After concentrating the IVLE, aliquots of IVLE were harvested at several intervals - 12 hours, 2, 4, 6, and 7 days, during which time IVLEs matured, as described [35]. After perfusion from mice, other male and female adult S. mansoni worms were separated, washed three times with 1×PBS, pH 7.4, and stored at −80°C. Cercariae released from infected B. glabrata snails were transformed mechanically into schistosomula [33], and cultured for 10 days in Basch's medium [36] at 37°C under 5% CO2 in air.
To search for S. mansoni vasa-like genes, Pfam searches of the draft genome of S. mansoni at Wellcome Trust Sanger Institute GeneDB, http://www.genedb.org/Query/pfam?taxonNodeName=Smansoni were carried out using DEAD/DEAH helicase (Pfam accession PF00270) and eukaryotic RNA helicase (Pfam accession PF00271) domains that included DEAD/DEAH helicases along with other RNA helicases as queries [37]. The Pfam database uses the profile hidden Markov model software, HMMER3. Separately, the genome was screened using the keyword ‘DEAD’ as a query in a motif search at GeneDB, http://www.genedb.org/Query/motif?taxonNodeName=Smansoni. Putative orthologous proteins for each S. mansoni DEAD-box gene were predicted by protein BLAST at NCBI with the current default settings, including, analysis against all databases using Blosum 62 scoring matrix with the following criteria: at least 20% identity to the query sequence and e-value lower than 0.001. Pairwise comparisons were performed using the pairwise BLAST program at NCBI, http://blast.ncbi.nlm.nih.gov/ to determine identity/similarity percentages.
The deduced amino acid sequences of three putative S. mansoni DEAD-box helicases were aligned with reference sequences from informative species by ClustalW (Bioedit) software [38]. Phylogenetic analysis was conducted in MEGA5 [39]; 30 amino acid sequences were included in the analysis. The sequences were aligned in Bioedit and gaps were excluded from the alignment. A bootstrapped Neighbor-joining tree was generated based on the whole deduced amino acid sequences, with the eukaryotic initiation factor-4A (eIF4A) as the out-group [40]. Bootstrap analysis was performed with 1,000 data sets. Bootstrap values shown on the tree represent the percentage of replicate trees in which the associated taxa clustered together. 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 p-distance method, with distances in units of the number of amino acid substitutions per site. Sub-cellular sites of locations of proteins were predicted using PSORT II, http://psort.ims.u-tokyo.ac.jp/ [41]. In addition, gene structures including exon-intron boundaries were examined in order to predict evolutionary relationships among the schistosome vasa orthologues. Using GeneDB accessions Smp_033710, Smp_154320 and Smp_068440,and cDNA sequences JQ619869– JQ619871 (below), along with exon structures displayed by Artemis, we predicted exon-intron boundaries including positions of motifs characteristic of Vasa encoded by exons. Exon-intron boundaries were displayed using FancyGene [42].
Total RNA was isolated from S. mansoni developmental stages using the RNAqueous®-4PCR kit (Ambion, Austin TX). Residual DNA contaminating the RNA was removed by digestion with RNase-free DNase I (TurboDNase, Ambion) at 37°C for 1 h. cDNAs were synthesized using the iScript™ cDNA synthesis kit (Bio-Rad, Hercules CA) using 75 ng of total RNA as the substrate. Relative expression of genes of interest was analyzed by quantitative RT-PCR (qRT-PCR). TaqMan probes and qPCR primers were designed with the assistance of Beacon Designer (Premier Biosoft International, Palo Alto, CA): Smvlg1 forward primer, 5′-ACG ACT ATA ATG AGA ATA ATC TTG-3′; Smvlg1 reverse primer, 5′-CCA AAC TTT ATG TGC CTC-3′; Smvlg1 probe, 5′-/56-FAM/GTT CAG ATG GTG GTG GT/3IABlk_FQ/-3′; Smvlg2 forward primer, 5′-TGA GGC TAT AAC ACT TAT TC-3′; Smvlg2 reverse primer, 5′-CCT TCT TGA ATT TCC ATA TTG-3′; Smvlg2 probe, 5′-/56-FAM/AAG CAA CAA CCA TCA AGC AGA ACA/3IABlk_FQ/-3′; Smvlg3 forward primer, 5′-TAC CGT TCC AAC ATT AAG-3′; Smvlg3 reverse primer, 5′-GTT CCA GAC TCA CTT TAC-3′; Smvlg3 probe 5′-56-FAM/CTG CTC ACC ATA TCA ACA AGA CGC/3IABlk_FQ/-3′. TaqMan probe and primers targeting the schistosome alpha tubulin (GenBank XM_002581074; SCMSAT1A), as a reference gene were: SmAtubulin forward primer, 5′-GTG CTG TAT GTA TGT TAA GTA-3′; SmAtubulin reverse primer, 5′-CGT GCT TCA GTA AAT TCA-3′; and SmAtubulin probe, 5′-56-FAM/TCT TCC ATA CCT TCA CCA ACA TAC CAA/3IABlk_FQ/-3′. Quantitative PCRs were performed, with duplicate biological replicates each with triplicate technical replicates. Reactions were carried out in 20 µl volumes with primer-probe sets and Perfecta qPCR FastMix, UNG (Quanta Bioscience, Gaithersburg, MD). The qPCR conditions included an initial denaturation at 95°C for 3 min followed by 40 cycles of 30 s at 95°C and 30 s at 55°C, performed in a thermal cycler (iCycler, Bio-Rad) and a Bio-Rad iQ5 detector to scan the plate in real time. Relative quantification was calculated following the 2−ΔΔCt method [43]. Alpha tubulin was employed as the reference because it maintains steady expression levels among developmental stages [44] and the cercarial stage was used as a calibrator.
Whole mount in situ hybridization (WISH) to adult S. mansoni worms was carried out using the method of Cogswell and colleagues [45], adapted from methods for planarians [46]. Each experimental group included ∼20 female and ∼20 male adult schistosomes; the experiment was repeated three times. In brief, schistosomes were fixed in 4% paraformaldehyde (PFA), adult males were reduced with 50 mM dithiothreitol, 1% NP-40, and 0.1% SDS, and both males and females were dehydrated in a methanol series, then stored at −20°C. Prior to WISH, worms were bleached with 6% H2O2, rehydrated through a methanol series to 1×PBS with 0.3% Triton-X 100, permeabilized using proteinase K, and re-fixed in 4% PFA. Hybridization was performed at 56°C with digoxigenin [DIG]-labeled riboprobes (below) for 18–20 h. Incubation in anti-DIG antibody at 1∶2000 diluted in blocking solution was performed overnight at 4°C, and signals developed with nitro blue tetrazolium/5-bromo-4-chloro-3-indolylphosphate. WISH specimens, mounted in glycerol, were examined using a Zeiss Axio Observer A.1 inverted microscope fitted with a camera AxioCam ICc3 camera (Zeiss). Manipulation of digital images was undertaken with assistance of AxioVision release 4.6.3 (Zeiss) and ImageJ 1.45 software [47]. Manipulations were limited to insertion of scale bars, adjustments of brightness and contrast, cropping and the like. Image enhancement algorithms were applied in linear fashion across the entire image and not to selected aspects.
Transcripts encoding Smvlg1, Smvlg2, and Smvlg3 (below) were amplified from mixed-sex, adult worm cDNA with the following primers: Smvlg1 cDNA forward primer, 5′- ATG TCT TAC GAC TAT AAT GAG AAT A-3′ and Smvlg1 cDNA reverse primer, 5′-CTA ATT GCC CCA CCA GTC TGG AGA A-3′, PCR product size 1,914 bp; Smvlg2 cDNA forward primer, 5′-ATG AAT AGA GTA ATT CTG AGT CAG C-3′ and Smvlg2 cDNA reverse primer, 5′-TCA CAA ATA TCG CAT CAA ACT ATC A-3′, PCR product size 1,839; Smvlg3 cDNA forward primer, 5′-ATG GAG AGC CTA GAA AAC AAT TTT GGC-3′ and Smvlg3 cDNA reverse primer, 5′- AAA CAT CAA AGT CTG TCT TTT TC-3′, PCR product size 1,065 bp of the 2,835 bp complete cDNA. PCR products were cloned into vector pCR4-TOPO (Invitrogen) and the nucleotide sequences of the inserts determined. These sequences have been assigned GenBank accessions JQ619869–JQ619871 for Smvlg1, Smvlg2, and Smvlg3, respectively. Previously, we have described Argonaut 2 (Ago2) cloned into pJC53.2 [45].
Antisense and sense digoxigenin-labeled riboprobes were synthesized by in vitro transcription employing PCR products as templates using gene specific primers tailed with the T7 promoter sequence as follows, (T7 promoter sequence italicized): Smvlg1 sense forward primer, 5′- TAA TAC GAC TCA CTA TAG GGC AAA CGG TTC AGA TGG TGG TGG TGC C-3′ and Smvlg1 sense reverse primer, 5′- GCG CTT ATA GAA TCA CCA GGA CCT TGC -3′, spanning coding DNA positions 62–737; Smvlg1 antisense forward primer, 5′-CAA ACG GTT CAG ATG GTG GTG GTG CC-3′ and Smvlg1 antisense reverse primer, 5′- TAA TAC GAC TCA CTA TAG GGG CGC TTA TAG AAT CAC CAG GAC CTT GC -3′, spanning coding DNA positions 62–737; Smvlg2 sense forward primer, 5′-TAA TAC GAC TCA CTA TAG GGT TAC TTT ACA TTC ATC ACG TCA TA-3′ and Smvlg2 sense reverse primer, 5′- CTA TAA AAT GAC ATC CCT TAT TCA AC -3′, spanning coding DNA positions 32–769; Smvlg2 antisense forward primer, 5′-TTA CTT TAC ATT CAT CAC GTC ATA-3′ and Smvlg2 antisense reverse primer, 5′- TAA TAC GAC TCA CTA TAG GGC TAT AAA ATG ACA TCC CTT ATT CAA C -3′, spanning coding DNA positions 32–769; Smvlg3 sense forward primer, 5′- TAA TAC GAC TCA CTA TAG GGT ATC TGA GCT TAA GAG ATG CGT CC-3′ and Smvlg3 sense reverse primer, 5′-ACA TGC CAT CAA ATC ACG TTT-3′, spanning coding DNA positions 28–606; Smvlg3 antisense forward primer, 5′-TAT CTG AGC TTA AGA GAT GCG TCC-3′, and Smvlg3 antisense reverse primer, 5′-TAA TAC GAC TCA CTA TAG GGA CAT GCC ATC AAA TCA CGT TT-3′, spanning coding DNA positions 28–606; Ago2 antisense forward primer, 5′-CCA GTG AAA GTC GTT GCA GA-3′ and Ago2 antisense reverse primer, 5′-TAA TAC GAC TCA CTA TAG GGA CTT GCG GAC TTG CTG AGT T-3′, spanning coding DNA positions 955–2,149. The cycling protocol included an initial denaturation at 95°C for 2 min followed by 35 cycles of 30 s at 95°C, 30 s at 55°C, 60 s at 72°C, and a final extension at 72°C for 10 min. Riboprobes labeled with digoxigenin-11-UTP (Roche) were synthesized using the T7 polymerase Riboprobe System (Promega).
The presence of genes encoding the conserved DEAD/DEAH helicase (PF00270) or eukaryotic RNA helicase (PF00271) domains in the genome of S. mansoni was investigated by sequence query and by keyword searches. Positive hits were analyzed for presence of the DEAD-box motif, Asp-Glu-Ala-Asp, with degeneracy permitted in the third residue of the motif, and for nine other motifs diagnostic of DEAD-box RNA helicases [48], [49]. However, positive hits bearing a DExH-box motif (Asp-Glu-x-His) were excluded from subsequent analyses [48], [50]. This search identified 33 putative RNA helicases that by BLAST exhibited e-values≤2e−69 and percent identities of 39–99%, thus suggesting they were DEAD-box proteins (Table S1). A multiple sequence alignment of these 33 orthologues revealed that all ten diagnostic motifs of DEAD-box RNA helicases were mostly conserved although, for example, motifs V and V1 were absent from Smp_034190.2, Smp_095920, Smp_096530, and Smp_166400, (Figure S1; Table S1).
Among the 33 DEAD-box helicases of S. mansoni, BLAST revealed that three of the deduced schistosome enzymes, Smvlg1, Smvlg2 and Smvlg3 (GeneDB accessions Smp_033710, Smp_154320 and Smp_068440) (GenBank JQ619869, JQ619870, JQ619871) showed high identity to Vasa proteins from other flukes: Smvlg1 was 89% identical and 95% similar to the DDX3X ATP-dependent RNA helicase of the blood fluke S. japonicum (GenBank CAX73517), and was 74% identical and 84% similar to the ATP-dependent RNA helicase of the human liver fluke Clonorchis sinensis (GenBank GAA28330); Smvlg2 was 51% identical and 68% similar to the DDX3/DED1 ATP-dependent RNA helicase of C. sinensis (GenBank GAA56795), and 44% identical and 63% similar to the N. girellae vasa-like 2 (Ngvlg2) protein (GenBank BAF44660); Smvlg3 was 90% identical and 94% similar to the Vasa of S. japonicum (GenBank AFC17964), and 60% identical and 72% similar to the partial ATP-dependent RNA helicase sequence, DDX3X, of C. sinensis (GAA39366). Smvlg1 spanned 9,569 bp (CABG01000013; chromosome 2 unplaced supercontig 0108) where the gene included nine exons that encoded 637 deduced amino acid residues. Smvlg2 spanned 12,752 bp (CABG01000090; supercontig 0203); the gene included eight exons that encoded 625 deduced amino acid residues. Smvlg3 spanned 15,318 bp (HE601625; chromosome 2; supercontig Smp_scaff000223); the nine exons encoded 944 deduced amino acid residues.
Multiple sequence alignment of the three S. mansoni Vasa-like proteins revealed that, with the exception of Smvlg2, they exhibited the ten motifs characteristic of DEAD-box proteins [51], [52] (Figure 1). The Q motif (GaccPoPIQ; see key in Figure 1) and the conserved Phe upstream of Gln, were absent from the Smvlg2 protein, in like fashion to Ngvlg2 of the monogenean N. girellae [28], [53]. In addition to the ten conserved DEAD-box motifs, an EARKF motif found only in Vasa, PL10, and An3 [30], [54], [55] was present in Smvlg1 as DARKF (residues 275–279) and in Smvlg3 as EARKF (residues 259–263). This motif was absent from Smvlg2 (Figure 1B).
Beyond conserved motifs of the core region, the alignment revealed similarities in the amino and carboxy termini of the schistosome Vasa proteins. The amino-terminus of Smvlg1 is glycine-rich (13% Gly residues 23–111), including five Arg-Gly repeats and one RGG motif. In Smvlg3, asparagine-rich stretches occurred at the COOH-terminus. Moreover, in both Smvlg1 and Smvlg3 the characteristic Trp and Asp residues occurred proximal to the stop codon [56]. PSORT II analysis predicted that Smvlg1 and Smvlg3 were nuclear proteins (PSORT II scores, 60.9–69.6%) whereas Smvlg2 Vasa-like protein appeared to be located in the mitochondrion (score, 65.2%).
Phylogenetic analysis confirmed that Smvlg1 belonged to the PL10 family. However, it grouped within a separate cluster with orthologues from other flatworms classified as vasa-like genes (Figure 2). Smvlg2 grouped with Ngvlg2 of the monogenean N. girellae which expression and RNAi analyses had confirmed to be a vasa-like gene, although phylogenetic analysis had indicated that it was closely related to the Vasa subfamily or p68 family [28], and the DDX3/DED1 ATP-dependent RNA helicase of C. sinensis. Smvlg3, like DjvlgB of the planarian D. japonica, appeared to be a member of the PL10 subfamily. Smvlg3 most closely clustered with the Vasa of S. japonicum and DDX3X of C. sinensis.
Also, we examined gene structure and exon-intron boundaries for the three schistosome vasa-like genes. This revealed apparently closer evolutionary relationship between Smvlg1 and Smvlg3, and more distant relation to Smvlg2 (Figure 2B), which in turn supports the relationships established in the phylogenetic analysis (Fig. 2A). Smvlg1 and Smvlg3 shared identical numbers of exons and clear synteny of gene fragments encoding the characteristic DEAD-box motifs (Fig. 2B).
Developmental expression of the vasa-like genes was investigated using qRT-PCR. Transcripts encoding Smvlg1, Smvlg2, and Smvlg3 were detected in all of developmental stage of S. mansoni examined (Figure 3). However, much higher levels of expression were detected in 12 hour IVLE (often >10 times higher than in other stages) and to a lesser extent in adult females. Generally, patterns of expression of each of the three genes were similar among the developmental stages (Figure 3, panels A, B and C).
To determine sites of expression of Smvlg1, Smvlg2, and Smvlg3 in adult schistosomes, adult S. mansoni worms were examined in whole mount in situ hybridization (WISH). WISH probed with digoxigenin-labeled riboprobes revealed that expression of each of Smvlg1, Smvlg2, and Smvlg3 was confined to the ovary of the adult females (Figure 4; and Smvlg3 not shown), this pattern of expression was observed in the antisense treated worms, but not in the sense treated worms. All or most of the ∼20 female worms examined in each of the three replicates of the WISH analysis were positive for antisense probes, for each of the three vasa-like genes. In females, transcripts were localized to the posterior region of the ovary where mature oocytes develop. Cognate sense strand probes were hybridized in parallel; signals in the ovary were not detected using any of the sense probes (Figure 4; and Smvlg3 not shown). In addition, anti-sense probe signals were not evident after hybridization to the testes or other sites in male worms (not shown). An antisense probe to Argonaut2 (Ago2) was used as a positive control for WISH (not shown) (see [45]). The three Smvlg probes targeted the 5′-region of the transcripts, which do not encode diagnostic DEAD-box motifs.
The DEAD-box family is the largest of RNA helicase families, belonging to the helicase superfamily II. These enzymes are involved in RNA metabolic processes - transcription, pre-mRNA splicing, ribosome biogenesis, RNA transport, translation initiation, organelle gene expression, and RNA decay [48], [57]. DEAD-box helicases are over-expressed in cancer cells, further indicating roles in diverse cellular processes [58], [59]. Vasa, an adenosine 5′-triphosphate (ATP)-dependent DEAD-box RNA helicase, is an archetypal, perhaps indispensable, metazoan germ cell specific marker [11], [12]. As befits a founding member of the DEAD-box RNA helicase family, Vasa exhibits nucleic acid unwinding activity especially as a translational regulator in determination and maintenance of germ cells [12], [60]. For example, in D. melanogaster, Vasa plays a key role in the localization and translational regulation of germ line specific mRNAs such as nanos, gurken, and the Oskar protein [61]–[64]. Recent interest in Vasa including in non-model species also indicates that Vasa plays roles in mitosis as well as germ line maintenance and meiosis. The expanding catalogue of Vasa roles and locations beyond the germ line includes multipotent stem cells, embryonic cells, tumors, primordial germ cell specification, stem cell maintenance, cell cycle progression, and piRNA biogenesis (see [65]).
Database searches indicated that S. mansoni has at least 33 DEAD-box helicases, consistent with repertoires in other species - the human genome codes for 36 DEAD-box helicases [66], Saccharomyces cerevisiae has 26 [50], and 22 are known from Plasmodium falciparum [53]. Among these, three vasa-like genes were identified and designated Smvlg1 to 3. A complement of three vasa-like genes for S. mansoni conforms with other species where from one to four genes have been identified: humans [67], D. melanogaster [19], and Xenopus [15] all possess one, the anemone Nematostella vectensis possesses two [68], N. girellae has three [28], and C. elegans has four [17].
Phylogenetic analysis confirmed that Smvlg1 was a member of the PL10 subfamily of the DEAD-box RNA helicases, although it formed a separate cluster with orthologues from other flatworms that have been classified as vasa-like genes. Smvlg1 was most closely related to ATP-dependent RNA helicases of S. japonicum (DDX3X) and C. sinensis, and Smvlg3 appeared closely related to Vasa of S. japonicum and another ATP-dependent RNA helicase of C. sinensis (DDX3X). Smvlg2 showed high identity to the DDX3/DED1 helicase of C. sinensis, to N. girellae vasa-like gene 2 (Ngvlg2), for which earlier expression and RNAi analyses had confirmed as a vasa-like gene [28]. Smvlg1 and 3 share substantial identity with PL10 (PL10 is termed DDX3 in humans); DDX3 is a DEAD box RNA helicase that promotes mitotic chromosome segregation in somatic human cells [69]. Differences in function have been described between PL10 and Vasa, although there is uncertainty about phylogenetic relationships among flatworm vasa-like genes and their homologues. PL10 is not restricted to the germ line and is expressed in numerous other tissues [70], [71]. Notably, PL10-related DED1 of yeast may be required for translational initiation of all mRNAs [72]. The phylogenetic relationship between Vasa and PL10 suggests that Vasa is derived from an ancestral PL10 related gene that subsequently acquired specificity for the germ line [27], although phylogenetic analysis has not been able to reliably assign flatworm DEAD-box helicases into PL10 or Vasa subfamilies (see [28], [30]). Nonetheless, mapping the exon-intron boundaries corroborated the evolutionary relatedness of Smvlg1 and Smvlg3 compared to Smvlg2.
Sequence analysis of Smvlg1 and Smvlg3 indicated that they shared the 10 conserved motifs characteristics of DEAD-box proteins plus the EARKF motif that is diagnostic of PL10 and Vasa-like sub-groups of the DEAD box helicases [51], [52]. In Smvlg2, the Q and EARKF motifs were absent in like fashion to Ngvlg2 of the N. girellae [28], [53]. Furthermore, Smvlg1 and Smvlg3 retained conserved Trp and Asp residues proximal to the stop codon. Smvlg2 has evolved two Asp residues near the stop codon. The presence of Trp, Glu, and Asp residues in close proximity to start and stop codons is common within the Vasa family [56]. Moreover, at the amino-terminus of Smvlg1 there was a short glycine-rich region that contains five Arg-Gly repeats and one RGG motif. Repetitive RGG motifs and G-rich regions near the amino-terminus are known from Vasa orthologues but not from PL10 enzymes [73]. The RGG repeats are thought to function in RNA binding as well as a site for Arg methylation [74]–[76]. In both Danio rerio and D. melanogaster, the RGG repeats are required for subcellular localization of Vasa to the nuage structure [21], [77]. Vasa localizes at the subcellular level to polar granules (also variously termed or co-localized with P-granules, P-bodies, nuage, mitochondrial cloud and/or chromatoid bodies), structures that are electron-dense, perinuclear, and rich with ribonucleoproteins [11], [77]–[79]. In Smvlg3, asparagine-rich stretches were present in the C-terminal region, similar to other Vasa proteins including the D. japonica vasa-like gene B (DjvlgB) [30], [74]. PSORT II analysis predicted that Smvlg1 and Smvlg3 were nuclear proteins whereas Smvlg2 was mitochondrial, which is consistent with other species [79].
Developmental expression profiles of Smvlg1, Smvlg2 and Smvlg3 were similar and all three were expressed in all developmental stages of S. mansoni examined. Notably, however, 12 hour old IVLE displayed expression levels strikingly higher than other stages, and expression was elevated in adult females compared to males. When released from the female, the schistosome egg is undeveloped; accordingly, the elevated levels of these vasa-like transcripts in IVLE likely reflected higher ratio of germ to soma cells in these undeveloped/underdeveloped eggs compared to the other developmental stages. If so, targeting IVLE with transgenes in order to establish lines of transgenic schistosomes may be a judicious strategy [34]. (Alternatively or in addition, these elevated levels of vasa-like transcripts in the developing egg may be of maternal origin [80].) WISH was employed to examine the spatial expression of Smvlg1, Smvlg2, and Smvlg3 in the adult schistosomes. Elevated expression was observed in the posterior ovary where mature oocytes develop [81]. Tissue-specific, transcriptomic analysis of S. mansoni also indicates up-regulation of Smvlg1 and Smvlg2 in ovary and Smvlg2 in the testis [82] (not shown). Up-regulation of vasa-like genes in mature oocytes has been shown in other species [67], [80].
Expression of Vasa in late oogenesis indicates that Vasa has a role as a maternal factor for the determination of germ lineages [80]. In D. rerio, Xenopus species, C. elegans, and D. melanogaster, primordial germ cells are specified by maternally inherited cytoplasmic determinants that involve Vasa-related proteins. By contrast, in the mouse, primordial germ cells are induced during gastrulation [83]–[85]. The Vasa homolog MVH is found in primary oocytes but not in mature oocytes of the mouse [86]. The timing of developmental expression of vasa like genes in some Lophotrochozoans is known to take place continuously - during embryogenesis, after embryogenesis and continuously in adults [87]. Accordingly, it is also feasible that the schistosome vasa-like genes may not be expressed only in germ cells but also in somatic cells since all three are expressed in numerous developmental stages. As noted, the schistosome vasa-like genes may be valuable markers in functional genomics of schistosome to monitor introduction of transgenes into the schistosome germ line, but this pattern of expression may impact on the potential use of vasa-like genes as markers for germ cell identification and transformation [34], [88]. This potential impediment notwithstanding, WISH corroborated expression of Smvlg genes in germ cell rich tissues and organs of the adult female schistosome.
Deeper investigation of the cell and molecular biology of Smvlg1-3 should enhance our understanding of oogenesis, reproduction and the germ line in schistosomes. Gender dissimilarities in essentiality of Vasa, e.g. in mice null for Vasa, males are sterile whereas female are not, while in the hermaphroditic monogenean fluke N. girellae silencing of vasa-like genes results in loss of germ cells [12], [28], will be relevant to this investigation. Finally, and following up these reports of these critical roles of vasa in diverse species, DEAD-box RNA helicases have been identified as essential for schistosome survival [7], and hence the potential of the vasa-like gene products as intervention targets in schistosomes is also worthy of consideration.
|
10.1371/journal.pcbi.1000575 | Computational Model of Membrane Fission Catalyzed by ESCRT-III | ESCRT-III proteins catalyze membrane fission during multi vesicular body biogenesis, budding of some enveloped viruses and cell division. We suggest and analyze a novel mechanism of membrane fission by the mammalian ESCRT-III subunits CHMP2 and CHMP3. We propose that the CHMP2-CHMP3 complexes self-assemble into hemi-spherical dome-like structures within the necks of the initial membrane buds generated by CHMP4 filaments. The dome formation is accompanied by the membrane attachment to the dome surface, which drives narrowing of the membrane neck and accumulation of the elastic stresses leading, ultimately, to the neck fission. Based on the bending elastic model of lipid bilayers, we determine the degree of the membrane attachment to the dome enabling the neck fission and compute the required values of the protein-membrane binding energy. We estimate the feasible values of this energy and predict a high efficiency for the CHMP2-CHMP3 complexes in mediating membrane fission. We support the computational model by electron tomography imaging of CHMP2-CHMP3 assemblies in vitro. We predict a high efficiency for the CHMP2-CHMP3 complexes in mediating membrane fission.
| Membrane fission is a key step of fundamental intracellular processes such as endocytosis, membrane trafficking, cytokinesis and virus budding. The fission reaction requires substantial energy inputs provided by specialized proteins. Recently, the ESCRT-III proteins have been implicated in membrane budding and fission involved in multivesicular body formation, cytokinesis and virus budding. The ESCRT-III proteins self-assemble into circular filaments and flat spirals in the membrane plane and generate tubular structures with dome-like end caps. We suggest and elaborate computationally on a mechanism by which the ESCRT-III complexes can drive membrane fission. The essence of the mechanism is in generation in the course of membrane attachment to the dome-like surface of an ESCRT-III assembly of a thin membrane neck accumulating large elastic stresses. Relaxation of these stresses can drive the neck fission and formation of separate vesicles of biologically relevant sizes. Estimations of the membrane affinity to the protein surface required for the neck fission to occur and comparison of these values with the experimentally expected values justify quantitatively the proposed mechanism and demonstrate that ESCRT-III assemblies must be highly effective in promoting membrane fission.
| Membrane fission leading to division of one continuous membrane into two separate ones is ubiquitous in cell physiology. It is one of the crucial events in generation of transport intermediates from plasma membranes and intracellular organelles; steady-state dynamics of the endoplasmic reticulum, mitochondria and Golgi complex; virus budding, cytokinesis and other fundamental phenomena (see for review e.g. [1]–[3]).
In the process of fission, a membrane changes its shape and undergoes a topological transformation which includes transient perturbations of the membrane continuity. To overcome the membrane resistance to shaping and remodeling, a substantial energy has to be invested into the system, which requires action of specialized proteins (see for review [2],[3]). Identification of proteins which shape and remodel membranes in the course of diverse intracellular processes has become a hot topic of cell biology [1],[3],[4]. The major advance has been achieved in discovering proteins generating and/or sensing the membrane curvature. The list of such proteins is constantly expanding and the mechanisms of their action are being elaborated [1],[4],[5]. Less progress has been made in understanding how proteins drive the membrane fission per se. While several protein types such as the dynamin-family proteins (see e.g. [6]–[10]), CtBP1/BARS [11] and PKD [12] have been implicated in fission of cell membranes, until recently, the ability to split membranes was unambiguously demonstrated for, perhaps, only one protein, dynamin-1 [9], [13]–[15]. Whereas different versions of the mechanism of membrane fission by dynamin-1 were suggested (see for review [10]), the idea unifying the majority of these proposals is that dynamin self-assembles on the membrane surface into helical oligomers constricting the membrane underneath into thin tubes. Strong mechanical stresses induced by dynamin in the tubulated membrane upon GTP hydrolysis can relax as a result of membrane division and, therefore, drive membrane fission.
Accumulating evidence suggests that the ESCRT (Endosmal Sorting Complexes Required for Transport) complexes [16] – are able to catalyze the membrane budding and fission processes. The ESCRT machinery consists of five different complexes - theVps27complex (ESCRT-0), ESCRT-I, -II, and -III, and the Vps4 complex - whose coordinated action sorts trans-membrane proteins into intralumenal vesicles (ILV), which bud off from the limiting membranes of endosomes and transform endosomes into multivesicular bodies (MVB) [16]–[19].
In addition to the MVB generation, the combined action of ESCRT-III and VPS4 complexes are required for the budding of some enveloped viruses including HIV-1 [20]and during late steps in cytokinesis [21]–[24]. It is thus most likely that ESCRT-III and VPS4 catalyze membrane fission reactions, common to all three biological processes [21]–[25].
The ESCRT-III complex in yeast consists of four core subunits Vps20, Snf7, Vps24, and Vps2 [26] whose mammalian analogues are the charged multivesicular body proteins CHMP6, CHMP4, CHMP3 and CHMP2, respectively. The subunits are consecutively recruited to the membrane in the order of Vps20/CHMP6, Snf7/CHMP4, Vps24/CHMP3 and Vps2/CHMP2 [27]–[29] and their assembly into higher order complexes was suggested to drive the inward membrane budding in vitro [28]. Moreover, these four proteins are able to act as minimal budding machinery as was confirmed by demonstration that their sequential addition to giant unilamellar vesicles (GUV) generated membrane invagination and abscission of the inward vesicles [29]. Specifically, formation of membrane buds connected by open necks to the initial membrane was shown to depend, critically, on the Snf7(CHMP4) and Vps20(CHMP6) subunits, while the neck fission proved to require the Vps24(CHMP3) subunits [29].
Three different albeit similar models for ESCRT-III catalyzed budding have been suggested [30]. First, Snf7 (CHMP4) circular filaments or flat spirals lying in the membrane plane [31] start at the center of a newly formed membrane bud and catalyze membrane bending as the bud grows [31]. A second model suggests that a circular ESCRT-III filament with asymmetric ends delineates a membrane patch containing cargo molecules and constricts the neck of an evolving membrane bud via the disassembly action of Vps4 [27]. A third model, similar to the second one, proposes that an ESCRT-III spiral surrounds and constricts a cargo containing membrane domain leading to membrane budding and fission [29]. However, spiral polymers of ESCRT-III have only been observed for hSnf7(CHMP4) in vivo [31] and in vitro [32], whereas the detachment of the forming vesicle including fission of a membrane neck was shown to be crucially dependent on Vps24(CHMP3) [29]. Therefore, in addition to the Snf7(CHMP4) filaments, the structures formed by self-assembly of Vps24(CHMP3) must play an indispensable role in the ESCRT-III mediated membrane budding and fission.
CHMP3 (Vps24) and CHMP2A (Vps2) form heterodimers [26],[33] that assemble into tubular nano-structures which display a variety of end-cap shapes including nearly hemispherical dome-like end-caps ([34] and the section “Experimental support for the model” below). The external and internal radii of these structures are approximately 52 and 43nm, respectively [34]. In vitro, the AAA ATPase VPS4 binds to the inside of the CHMP2-CHMP3 polymers and leads to their disassembly in the presence of ATP [34]. The external surface of a CHMP2-CHMP3 nano-structure has a considerable affinity to membranes containing acidic lipids [34]. Therefore, in the process of self-assembly, the CHMP2-CHMP3 complex must be able to attract a lipid bilayer, hence, scaffolding the bilayer into a strongly curved shape, a process that might drive membrane fission reactions [34].
In spite of the apparent similarities between the dynamin-I and CHMP2-CHMP3 assemblies such as (i) the ability to scaffold membranes into cylindrical shapes, and (ii) the energy input by nucleotide hydrolysis, CHMPs cannot employ any of the mechanisms of membrane fission suggested for the dynamin action. Indeed, topologically, the fission reactions mediated by dynamin and ESCRT-III are directed differently: dynamin and its partners drive membrane budding and abscission towards the cytosol, while ESCRT-III mediates membrane abscission away from the cytosol and towards the lumen of an endosome. Structurally, a membrane portion tubulated by a dynamin oligomer is situated within the protein scaffold and, hence, could undergo further thinning upon detachment from dynamin and divide by self-fusion within the protein framework [14]. In contrast, the membrane wrapped around a CHMP2-CHMP3 structure is attached to the outside surface of the protein scaffold and, hence, the scaffold hinders the membrane sterically from direct thinning and self-fusion. Thus, the character of membrane deformation leading to fission driven by CHMP2-CHMP3 structure must differ essentially from that generated by dynamin and the mechanics of the fission reaction must be dissimilar in the two cases.
Here, we suggest and integrate the current structural knowledge on ESCRT-III complexes to elaborate on a novel mechanism of membrane fission by dome-like assemblies formed by the CHMP2-CHMP3 subunits of ESCRT-III. The essence of our proposal is that, in contrast to the fission mechanisms suggested for the dynamin action (see for review [10]), the site of membrane fission driven by ESCRT-III is not co-localized with the protein scaffold but rather emerges aside of it within a membrane neck which forms in the course of membrane wrapping around the ESCRT-III dome. The major energy for the fission reaction comes from the energy of membrane attachment to the surface of the ESCRT-III complex. We discuss a possibility for a reinforcement of the ESCRT-III based mechanism by the Vps4 binding.
Our calculations predict that ESCRT-III domes can serve as effective mediators of membrane fission resulting in generation of vesicles of biologically relevant dimensions.
We propose the following scenario for the membrane budding and fission by ESCRT-III complexes. At the first stage, the CHMP4 subunits are recruited to the membrane via CHMP6 [28] and self-assemble on the membrane surface into a circular filament or flat spiral [31], which leads to sequestering of a membrane patch and its bending into an initial bud, as proposed in [29] and illustrated in (Fig. 1a). We assume that the area of the initial bud sequestered by the CHMP4 spiral remains constant in the course of all downstream processes. In fact, attachment of the CHMP4 oligomers to the membrane surface [31] evidences a considerable attractive interaction between the CHMP4 and the lipid polar head groups. The lipid molecules whose head groups are bound to the protein spiral along the periphery of the bud (Fig. 1a) must build an effective “fence” preventing, within the time scale of membrane fission, the lipid exchange between the bud and the surrounding membrane, and, hence, restricting the changes of the bud area.
Next, the CHMP2A and CHMP3 subunits start self-assembling within the neck of this initial membrane bud which is accompanied by a concomitant attachment of the membrane to the emerging protein complex (Fig. 1b). The attachment is mediated by the attractive membrane-protein interaction. The total area of the initial bud is assumed to exceed considerably the area of the CHMP2-CHMP3 complex even after completion of its assembly. As a result, only a portion of the initial bud membrane can be directly attached to the protein structure (Fig. 1b). The rest of the membrane remains free and is connected by a neck to the attached membrane (Fig. 1b).
In the course of self-assembly, the CHMP2A-CHMP3 polymer builds up a tube whose end-cap gradually closes into a nearly hemi-spherical dome-like shape (Fig. 1b). The larger the fraction of the protein dome is assembled and covered by the membrane the thinner the neck. The neck tightening is accompanied by an increasing bending of its membrane and the related accumulation of the membrane elastic energy [35].
At a certain stage, the membrane elastic energy accumulated within the neck becomes so large that its relaxation can drive the neck scission, which results in formation of a spherical vesicle and a membrane cap covering the CHMP2A-CHMP3 dome (Fig. 1c). Two requirements have to be satisfied for fission to occur. First, the membrane scission event has to be overall energetically favorable meaning that the total energy of the system before fission must exceed the energy of the post-fission vesicle and membrane cap attached to the ESCRT-III dome. Fulfillment of this condition ensures the general feasibility of the fission reaction but does not guarantee that the reaction will be sufficiently fast to make it biologically relevant. The second requirement concerns the fission rate which can be limited by the energy barriers. According to this requirement, the energy barriers produced by the intermediate structures formed in the course of membrane splitting have to vanish or remain small. Based on electroporation experiments, feasible energy barriers which can be overcome within a time scale of few seconds by a membrane of large area is about , (where is the product of the Boltzmann constant and the absolute temperature) [2]). For small membrane fragment making up a membrane neck, the feasible energy barrier must be a few times lower and constitute less than . A major energy barrier is related to the strongly deformed intermediate structures forming transiently in the course of the process. In analogy to the well understood process of membrane fusion (see for review [36]–[38]), we assume that this energy barrier is associated with the hemi-fission intermediate in which the internal monolayer of the membrane neck is already split, while the second monolayer is still intact [35]. According to the analysis of fission of a membrane neck emerging during membrane budding by a spherical coat, the fission reaction is energetically favorable and the hemi-fission intermediate does not represent a kinetic barrier if the membrane neck in its thinnest cross-section narrows down to the threshold radius of about [35].
The attractive interaction between the subunits of the CHMP2-CHMP3 structure must be much stronger than all other relevant interactions characterizing the system. According to the results below for a characteristic energy needed to bend the membrane around the protein dome (∼0.25 mN/m) and a characteristic area of about 22.5 nm2 exposed by one CHMP protomer to interaction with the membrane [33], the low limit for the energy of the subunit interaction needed for the protein structure to remain stable upon bending of the attaching membrane, can be estimated as . In reality, the interaction energy of the CHMP protomer must exceed considerably this estimate since their self assembly is, practically, irreversible [34]. Based on this assumption, we propose that the protein self-assembly proceeds irrespectively of the membrane attachment, while the latter follows the dome formation and its extent is determined by the interplay between the membrane bending energy and the membrane affinity to the protein surface.
In the following, we will analyze quantitatively the above scenario of the membrane neck fission by the CHMP2-CHMP3 dome. Since the thinning of the membrane neck is driven by the progressing membrane attachment to the protein dome, we will consider only the dome part of the protein complex. We will compute the extent of the ESCRT-III dome coverage by the membrane and the corresponding shapes of the membrane bud for different values of the membrane affinity to the ESCRT-III complex. We will find the affinity values at which the membrane neck becomes sufficiently narrow to favor energetically the fission reaction. We will also determine the affinity required to reach the threshold neck radii at which the energy barrier associated with the hemi-fission intermediate becomes negligible and does not limit the fission rate.
We consider a hemi-spherical protein dome of radius serving as a scaffold for attachment of a membrane fragment of a total area (Fig. 2a). While, in reality, the membrane attachment to the dome proceeds concomitantly with the dome assembly, for the calculation purposes we will regard the dome to be completed. This is based on a plausible assumption that the attractive interaction between the subunits of the CHMP2-CHMP3 structure must be much stronger than all other relevant interactions characterizing the system. Therefore, the protein self-assembly proceeds irrespectively of the membrane attachment, while the latter follows the dome building and its extent is determined by the interplay between the membrane bending energy and the membrane affinity to the protein surface.
The absolute value of the energy of the membrane interaction with the dome surface per unit area of the membrane-protein interface will be referred to as the membrane affinity and denoted by . Since the membrane-protein interaction is attractive its energy is negative and its value per unit area is . Note that, according to our definition, the affinity accounts only for the direct (probably, electrostatic) interaction between the protein and the lipid polar groups and does not include the energy of membrane bending, which accompanies the membrane binding to the protein dome and contributes to the total energy of this process. Therefore, the value of is not supposed to depend on curvature of the protein surface. In this respect, the notion of the affinity we are using differs from the total energy of the membrane attachment to the protein complex, which includes the bending contribution and is commonly used to characterize interaction of proteins with bent membranes (see e.g. [1],[4],[39],[40]). In our approach the curvature effects are considered separately from the direct membrane-protein interaction.
The membrane adopts a curved shape of a bud characterized at each point by the total curvature and the Gaussian curvature [41]. The radius of the narrowest cross-section of the bud neck will be referred to as the neck radius, (Fig. 2a). The membrane bending energy per unit area of the membrane mid plane, , is given by [42],[43],(1)where is the bilayer bending modulus (see e.g. [44]), and is the bilayer modulus of Gaussian curvature whose values were not directly measured but estimated to be negative (see e.g. [45],[46]).
We analyze two alternative states of the system: the fore-fission state where the membrane bud is connected by a membrane neck to the membrane portion attached to the protein dome (Fig. 2a), and the post-fission state represented by a separate spherical vesicle and the protein dome completely covered by the membrane (Fig. 2b). Our goals are (i) to compute the energies of the two states and to find, by their comparison, the affinity values at which the membrane fission event is energetically favorable, and (ii) to determine at which the membrane neck in the fore-fission state becomes as small as guaranteeing fast fission [35].
In the fore-fission state, the extent of the membrane attachment to the protein dome will be characterized by the angle referred below to as the attachment angle which indicates the position of the upper border of the attached area (Fig. 2a). The total energy of the system in the fore-fission state, , is the sum of two contributions. First, the total attachment energy found by integration of the attachment energy density, , over the attached area . Second, the total bending energy of the membrane, , determined by integration of over the whole area of the membrane including and the area of the bud . Taking into account Eq.1 and the system geometry (Fig. 2a), the total energy of the fore-fission state can be expressed as(2)
The first contribution to the Eq. 2 represents the sum of the attachment energy and the bending energy of the attached membrane portion whose total curvature, , is related to the dome radius, , by . The second contribution is the bending energy of the bud, which depends on the curvature distribution along the bud surface. The third contribution is the energy of the Gaussian curvature, which does not depend on the system configuration. The energy (Eq. 2) has to be minimized with respect to the attachment angle and the distribution of the total curvature along the surface of the bud for any given value of the affinity . This will give the equilibrium values for and the corresponding attached area , determine the equilibrium shape of the membrane bud including its neck radius , and provide the equilibrium total energy of the fore-fission state. Because of a complex shape of the membrane bud, minimization of Eq.2 will be performed numerically by the standard method of finite elements using the COMSOL Multiphysics software.
In the post-fission state, consisting of a spherical vesicle and the hemi-spherical dome covered completely by the membrane (Fig. 2b) the total energy is(3)
In the following, we can skip the Gaussian curvature contribution to the fore-fission energy , and account for the addition of to the energy of the post-fission state .
CHMP2A/CHMP3 polymers were assembled and analyzed by negative staining electron microscopy as described [34]. CHMP2A/CHMP3 polymers were applied to a holey carbon grid and plunge frozen in liquid ethane. The samples were examined in an FEI F30 Polara microscope, equipped with a Gatan GIF post-column energy filter [47]. Tilt series were acquired over an angular range of 120 degrees, at a nominal magnification of 27,500 times, which corresponded to a pixel size of 0.49nm, and at a defocus of 5 to 7 microns. Tomograms were generated from these tilt series using the IMOD software package [48] and visualized in Amira (Visage Imaging).
We consider the membrane affinity, , as the major parameter determining the system configurations and the conditions for membrane fission. Other parameters whose values may vary for different membranes are the membrane area and the membrane modulus of the Gaussian curvature, . For we consider the range [45],[46]. The range of the membrane area is chosen to be , where is the external radius of the dome surface. This corresponds to variation of the vesicle diameters in the post-fission state in the biologically relevant range between 20 nm and 100 nm.
A typical computed shape of the membrane bud corresponding to a certain attachment angle , is presented in Fig. 2a and can be described as a sphere-like cap connected to the attached membrane by a funnel-like neck. The larger the angle , the smaller the neck radius (Fig. 3). At the attachment angle the neck radius becomes smaller than the threshold value, , which fulfills the condition of the fast fission [35]. Therefore, we limited the considered range of the attachment angles by . Generally, the computation could be stretched to higher attachment angles corresponding to even narrower necks. This would require, however, including in the elastic energy model additional terms of higher order in the curvature of the internal monolayer of the neck, and taking into account the energy of the short range hydration repulsion through the neck lumen between the elements of the internal surface of the neck. Such sophistication of the model would complicate considerably the computation without significant changes of the model predictions on the neck fission.
The character of the dependence of the system energy on the attachment angle is determined by the affinity (Fig. 4). According to the first term in Eq.2, the membrane binding to the protein dome will occur only if the affinity exceeds a certain value, , which is the least affinity needed for compensation of the energy penalty of membrane bending accompanying the attachment to the dome surface.
At each particular affinity value larger than , the system can reside in a stable or quasi-stable configuration described by the values of corresponding to the energy minima (Fig. 4). There are four different ranges of the affinity determining different regimes of the possible system configurations. Transitions between these regimes are determined by the three characteristic values of the affinity denoted by , and and presented in Fig. 5.
The first regime corresponds to the affinities smaller than the first characteristic value, . Here, the energy has one minimum at small values, , of the attachment angle (Fig. 4), meaning that the stable configuration of the system is a bud with a neck whose radius is somewhat smaller than but comparable with the radius of the protein dome . We will refer to this configuration as the broad neck configuration.
In the second regime, the affinity varies between the first and the second characteristic values, . In this range, a second energy minimum emerges at the largest possible attachment angle within the considered range, (Fig. 4), corresponding to a bud with a neck of radius (Fig. 3). This configuration will be called the narrow neck configuration. The total energy in the second minimum is higher than in the first one, , which means that the narrow neck is a quasi-stable while the broad neck is a stable configuration. It has to be noted that, in contrast to the first energy minimum, the second one is not characterized by a vanishing first derivative of the energy function and represents the minimal energy value found in the considered range of the attachment angle. This feature of the second minimum does not influence, however, the conclusions of the analysis of the membrane fission conditions.
In the third regime, the affinity is in the range between the second and third characteristic values, . Under these conditions, the narrow neck is energetically more favorable (Fig. 4) and, hence, becomes stable whereas the broad neck turns quasi-stable.
Finally, in the fourth regime the affinity is larger than the third characteristic value, . Here, the energy minimum corresponding to the broad neck vanishes and the only stable state of the system is that of the narrow neck.
The three characteristic affinity values, , and , and the geometrical characteristics of the membrane bud in the four regimes of configurations are illustrated in the phase diagrams (Fig. 5a,b,c). The first two phase diagrams represents the total energies (Fig. 5a) and the corresponding attachment angels (Fig. 5b) of the broad and narrow neck configurations for a specific value of the membrane area . The third phase diagram (Fig. 5c) shows how , and depend on the membrane area and, hence, on the area of a vesicle which would form if fission occurs. All the three characteristic affinities decrease with the membrane area which means that the larger the membrane, the lower affinities are needed for generation of buds with narrow necks.
Recall that we analyze two requirements for membrane fission. According to the first requirement, the fission reaction has to be energetically favorable meaning that the total system energy in the post-fission state must be lower than in the fore-fission state, . Upon this condition, the fission reaction may be slow because of the existence of kinetic barriers.
According to the second requirement, the energy barriers of the fission reaction must, practically, vanish, which guarantees fast rates of the membrane splitting. Particularly, the membrane neck has to narrow up to the threshold value , which guarantees that not just the overall fission reaction but also the intermediate hemi-fission stage is energetically favorable and does not limit the fission rate [35].
The computed system energies in the fore- and post- fission states for different values of the affinity and different moduli of the Gaussian curvature are presented in Fig. 6. According to these results the first requirement is always satisfied in the narrow neck configuration confirming the previous works. Also for the broad neck configurations the fission reaction may be energetically favorable. To this end the affinity has to be larger than a certain value varying in the range between 0.27mN/m and 0.37mN/m for feasible values of the Gaussian curvature modulus (Fig. 7). The more negative is , the looser are the fission conditions, i.e. the lower affinity is needed for fission to be energetically favorable. However, to undergo fission from the broad neck configuration, the system has to overcome a substantial energy barrier and, in practical terms, the membrane splitting will not occur.
The requirement of fast fission can be fulfilled if the system reaches the narrow neck configuration. However, to achieve this state in the course of the membrane attachment to the protein dome, the system has to proceed through the whole range of the attachment angles beginning from and up to . This means that the system has to move along one of the energy profiles represented in Fig. 4. According to Fig. 4, if the affinity value is smaller than , there is an energy barrier and the system has to overcome to reach the narrow neck configurations. This means that for the membrane fission will be restricted kinetically. At the larger affinity values, , evolution of the membrane bud up to the narrow neck configuration is accompanied by a monotonous decrease of the energy and, hence, proceeds without kinetic restrictions. Summarizing, the condition for the fast fission is .
To support the model, we studied the structures resulting from the CHMP2-CHMP3 self-assembly by negative staining [34] and cryo electron tomography (see Materials and Methods). We observed assembly of open tubes, tubes with flat closures, tubes with hemispherical almost closed ends (defects in closure) and closed tubular structures with hemi-spherical end-caps (Fig. 8). The presence of closure defects observed in the structures assembled in vitro might be due to fact that they have been assembled in the absence of membranes. In the current model we propose that these structures assemble directly on membranes. Formation of the closed hemi-spherically capped tubes substantiates the existence of the protein domes which play the central role in the model. These structures should represent the final stage of CHMP2-CHMP3 polymerization and our model suggests that they are physiologically relevant.
We suggested and analyzed a mechanism by which a minimal ESCRT-III complex composed of the mammalian ESCRT-III proteins CHMP2A and CHMP3 can drive fission of membrane necks. The mechanism is based on the experimental results which demonstrate that CHMP2A and CHMP3 heterodimers self-organize into tubular assemblies some of which reveal closed hemispherical dome-like end-caps. The external surfaces of these assemblies have a considerable affinity to lipid bilayers containing acidic lipids.
The essence of the model is that a CHMP2-CHMP3 tube with a dome-like end-cap self-assembles in the neck of an initial membrane bud generated by a circular filament of a CHMP4 (the latter suggested in [27],[29],[30]) (Fig. 1a,b). The CHMP2-CHMP3 self-assembly is accompanied by membrane attachment to the dome surface which results in narrowing of the membrane neck as illustrated in (Fig. 1b and Fig. 2a). Because of the hemi-spherical shape of the dome, progression of the dome assembly and the concomitant membrane binding to its surface leads to thinning of the neck and accumulation of the elastic stresses within its strongly curved membrane. If a certain degree of the neck thinning is achieved, fission of the neck membrane accompanied by the stress relaxation becomes energetically favorable. The proposed mechanism entails containment of the ESCRT-III proteins towards the cytosolic side after fission, which is consistent with the observation that the ESCRT-III proteins have not been detected within intra-luminal vesicles of the MVBs.
Since both CHMP2A and CHMP3 interact with Vps4 [34],[49],[50], it is important to understand a possible role this protein can play in action of CHMP2A-CHMP3 complexes on membranes. Although the results by Hanson and colleagues [31] indicated that Vps4 might play an active role during the ESCRT-III driven membrane remodeling process, in vitro budding experiments with GUVs suggested that vesicle formation and fission occurred in the absence of Vps4, albeit it seems to accelerate the process [29]. We suggest that Vps4 could still play an important role other than disassembly of ESCRTs from membranes [51]. The hemispherical shape of the protein end-cap can be maintained only if the bending rigidity of the end-cap wall greatly exceeds that of the lipid membrane. In case the end-cap bending rigidity is similar to or smaller than that of the membrane, the top segment of the end-cap, which is not covered by the membrane, will flatten. This would result in a decrease of the membrane attachment angle and, hence, hinder, to some extent, the membrane neck narrowing necessary for the neck fission. While this effect is small for the large degree of the membrane coverage corresponding to the narrow neck configuration, it can be considerable for the broad neck configuration, and may influence the probability of transition from the broad to the narrow neck. Given that the 4.5 nm thickness of the ESCRT-III shell [34] is, practically, equal to that of a lipid membrane (see e.g. [52]), the rigidity of the purely ESCRT-III complex might be not large enough to prevent flattening of the end-cap top. Strengthening of the ESCRT-III end-cap by binding of a Vps4 dodecamer, that exposes 12 CHMP binding sites on the inside of the ESCRT-III polymer, may provide the protein structure with an additional rigidity required for a more effective fission.
The neck fission results in formation of a separate vesicle and a hemi-spherical membrane cap covering the protein dome (Fig. 1c and Fig. 2b). Based on the model of membrane bending elasticity [42], we computed how large the membrane affinity to the protein dome has to be in order to enable fast fission of the membrane neck leading to formation of a separate vesicle. Below we discuss the feasibility of the obtained results for the affinity and show that the CHMP dome must be an efficient mediator of membrane fission.
According to our computations, the affinity required to drive fission of the membrane neck depends considerably on the area of the membrane fragment undergoing budding and, hence, on the dimension of the vesicle generated in the result of fission (Fig. 5c). The ESCRT-III proteins have been implicated in generation of multivesicular bodies (MVBs) consisting of vesicles with characteristic diameters between 20 and 100 nm [19],[53] and in budding of enveloped viruses with diameters varying up to about 100 nm. Therefore, we performed calculations for the areas of the membrane bud between and corresponding to the relevant range of the vesicle diameters.
The largest affinity denoted as is needed to drive a kinetically unconstructed formation of a bud with a narrow neck of radius less which enables fast fission. The affinity (as well as two other characteristic affinities, and , determining conditions for slower fission processes), decreases with increasing membrane area. The maximum value of is needed for generation of the small 20 nm vesicles of MVBs. According to our results (Fig. 5c), the required affinity is .
The feasible values of the membrane affinity to the protein dome can be estimated based on a thermodynamic analysis of the kinetic measurements of the CHMP2A and CMHP3 monomer binding to the DOPS-SOPC bilayers [34]. According to these measurements, the CHMP2A and CHMP3 monomers dissociate from lipid with a dissociation rate constant (koff) of 0.08 s−1 and 0.3 s−1 respectively [34]. The association to lipid for both, CHMP2A and CHMP3, was found to be diffusion controlled thereby putting a lower limit on the association rate constant (kon) of 1×106 M−1 s−1. The condition of equilibrium between the lipid-bound and free protein monomers resulting from the equality of the rates of their association to and dissociation from the lipid can be expressed by the equation(4)where is the number of the lipid-bound protein monomers, is the number of the lipid molecules and is the volume concentration of the free protein monomers. On the other hand, thermodynamically, the same equilibrium condition can be expressed through the equality of chemical potentials of the lipid-bound and free protein monomers,(5)where and are the so called standard chemical potentials of the free and lipid-bound protein monomers accounting for the free energy of the direct monomer interaction with the surrounding, and are the contributions of the free and lipid-bound protein monomers from the translational entropy in the solution and on the membrane surface, respectively, is the molar concentration of water molecules. Eq.5 takes into account that the whole lipid is organized into one or few extended membranes whose translational entropy has a vanishing effect on the chemical potentials.
The protein-membrane binding energy per protein monomer is related to the standard chemical potentials by , so that the affinity which represents, according to the definition above, an absolute value of the binding energy related to the unit area of the protein-membrane interface, is given by(6)where is the area of a CHMP monomer exposed to interaction with the membrane. Combining Eqs.4–6 we obtain for the affinity . Given the kinetic constants above, and the estimation for the monomer contact area [33] we determine the membrane affinities of CHMP2A and CHMP3 to be and . Taking into account that the protein dome consists of the CHMP2A-CHMP3 heterodimers, the average affinity should be about , which exceeds almost by a factor of six the above estimation of for the affinity required for fast fission of the vesicles. Fission of larger vesicles requires even lesser affinities. Hence, the binding energy provided by the CHMP-membrane interaction must be excessively large and guarantees fast membrane budding and fission under all biologically relevant conditions.
The suggested mechanism of membrane fission by the ESCRT-III proteins CHMP2A-CHMP3 and the related calculations demonstrate that dome-like assemblies of these proteins could scaffold membrane necks into strongly curved shapes and favor membrane fission. Since, in contrast to the proteins of the dynamin family, the ESCRT protein complexes attach the membrane to their external surfaces, the fission site emerges within a free membrane fragment aside of the zone of protein-lipid interaction. The task of the CHMP4 and CHMP6 subunits, which are recruited to the membrane upstream of the CHMP2 and CHMP3 recruitment, is to generate an initial membrane bud with a fixed membrane area whose neck has to undergo fission to complete the vesicle formation. A role for Vps4, in addition to its recycling function, can be in reinforcing the wall of the ESCRT-dome which facilitates membrane bending and fission. It is conceivable that the suggested mechanism is not limited by the action of ESCRT-III proteins but rather has a more general character.
|
10.1371/journal.pgen.1004823 | Mutation of Npr2 Leads to Blurred Tonotopic Organization of Central Auditory Circuits in Mice | Tonotopy is a fundamental organizational feature of the auditory system. Sounds are encoded by the spatial and temporal patterns of electrical activity in spiral ganglion neurons (SGNs) and are transmitted via tonotopically ordered processes from the cochlea through the eighth nerve to the cochlear nuclei. Upon reaching the brainstem, SGN axons bifurcate in a stereotyped pattern, innervating target neurons in the anteroventral cochlear nucleus (aVCN) with one branch and in the posteroventral and dorsal cochlear nuclei (pVCN and DCN) with the other. Each branch is tonotopically organized, thereby distributing acoustic information systematically along multiple parallel pathways for processing in the brainstem. In mice with a mutation in the receptor guanylyl cyclase Npr2, this spatial organization is disrupted. Peripheral SGN processes appear normal, but central SGN processes fail to bifurcate and are disorganized as they exit the auditory nerve. Within the cochlear nuclei, the tonotopic organization of the SGN terminal arbors is blurred and the aVCN is underinnervated with a reduced convergence of SGN inputs onto target neurons. The tonotopy of circuitry within the cochlear nuclei is also degraded, as revealed by changes in the topographic mapping of tuberculoventral cell projections from DCN to VCN. Nonetheless, Npr2 mutant SGN axons are able to transmit acoustic information with normal sensitivity and timing, as revealed by auditory brainstem responses and electrophysiological recordings from VCN neurons. Although most features of signal transmission are normal, intermittent failures were observed in responses to trains of shocks, likely due to a failure in action potential conduction at branch points in Npr2 mutant afferent fibers. Our results show that Npr2 is necessary for the precise spatial organization typical of central auditory circuits, but that signals are still transmitted with normal timing, and that mutant mice can hear even with these deficits.
| Millions of people suffer from debilitating hearing defects, ranging from a complete inability to detect sound to more subtle changes in how sounds are encoded by the nervous system. Many forms of deafness are due to mutations in genes that impair the development or function of hair cells, which are responsible for changing sound into electrical signals that can be processed by the brain. Both mice and humans carrying these mutations fail standard hearing tests. In contrast, very little is known about the genetic basis of central auditory processing disorders, which are poorly defined and difficult to diagnose, since these patients can still detect sounds. By finding genes that are required for the normal wiring of central auditory circuits in mice, we can investigate how changes at the circuit level affect circuit function and therefore improve our understanding of central auditory processing disorders. Here, we show that the natriuretic peptide receptor Npr2 is required to establish frequency maps in the mouse central auditory system. Surprisingly, despite a dramatic change in circuit organization, Npr2 mutant mice are still able to respond to sounds with normal sensitivity and timing, underscoring the need for better hearing diagnostic methods in mice as in humans.
| The sense of hearing is mediated by precisely organized neural circuits that encode the frequency content, timing, and intensity of sounds. Frequency information is encoded in the spatial organization of hair cells in the cochlea, with high frequencies detected in the base and low frequencies in the apex. SGNs transmit this information to the cochlear nuclei, where their axons bifurcate into an ascending branch that innervates the aVCN and a descending branch that targets the pVCN and DCN. In each of these regions, the systematic innervation by SGN fibers forms frequency maps that maintain the tonotopic order that is established in the cochlea and that is preserved along the auditory pathway. Tonotopy also governs intrinsic connections between neurons in the cochlear nuclei, including tuberculoventral cell projections from the DCN to the VCN [1], [2].
SGN axons are responsible for delivering all acoustic information from the cochlea to the cochlear nuclei. By contacting a variety of target neurons with distinct projection patterns, each SGN feeds information to parallel pathways in the brainstem [3]. Through their ascending branches, SGNs convey auditory signals to bushy cells that are involved in comparing interaural time and intensity for localizing sounds in azimuth [4], [5], [6], [7], as well as to some T stellate cells. Through their descending branches, SGNs innervate T stellate cells that encode the spectrum of sounds [8], [9], [10], octopus cells that mark the onset of sounds [11], [12], [13], and fusiform and giant cells of the DCN that use spectral cues to localize sounds monaurally in the vertical plane [14]. Together, the activation of these diverse populations of cochlear nuclear neurons by SGN axons enables animals to detect, recognize, and locate sounds in their environment.
In order to make the precise pattern of diverse connections that enable the interpretation of sound, developing SGN axons must elaborate a variety of synapses that are tonotopically organized but that show distinct signaling properties depending on the nature of the target neuron. For instance, within one isofrequency lamina of the VCN, SGN axons contact bushy cells, T stellate, and octopus cells and form functionally distinct synapses with each cell type. The branches that innervate bushy cells terminate in unusually large and complex endbulbs of Held that mediate large post-synaptic responses that depress, yet still signal with high temporal precision [15], [16]. In contrast, on T stellate cells, SGN axons form typical bouton endings that induce smaller post-synaptic responses with less depression. Traditionally, the mechanisms that control axon guidance and synaptic function have been studied independently. However, recent evidence indicates that these two events can in fact be linked, as maturation of the calyx of Held does not progress normally in bushy cell axons that fail to cross the midline [17]. Whether the spatial organization of SGN axons is similarly coordinated with subsequent synaptogenesis remains unclear.
One of the primary obstacles towards understanding how functional auditory circuits are assembled is the lack of genetic mutations that disrupt SGN central wiring. As a result, there are no clear predictions for how changes in the pattern of central innervation might impact hearing either in mice or humans. Indeed, although a growing number of genes affecting cochlear function have been implicated in sensorineural deafness in humans [18], almost nothing is known about the genetic basis of central auditory processing disorders, which disrupt central auditory circuit function without obvious loss of hearing sensitivity [19], [20]. Identifying and characterizing genetic mutations that affect the formation of central auditory circuits in mice is an important step towards understanding how these disorders may arise.
During the course of a screen to identify genes required for auditory circuit assembly, we discovered that the natriuretic peptide receptor Npr2 is required for central axon bifurcation in SGNs [21]. Npr2 is a receptor guanylyl cyclase that activates a cGMP-dependent protein kinase signaling cascade upon binding to the C-type natriuretic peptide (CNP), which is expressed dorsally along the length of the embryonic neural tube [22], [23]. In Npr2 mutant mice, the axons of both dorsal root ganglion (DRG) and cranial ganglion neurons fail to bifurcate [24], [25]. However, interstitial branches can still form from the unbifurcated axons. These observations suggest that Npr2 ensures that bifurcations form in an orderly manner as sensory axons enter the spinal cord and encounter CNP, but that other mechanisms determine how and when axons arborize within their targets [26]. Although Npr2 signaling leads to changes in cytoskeletal-associated proteins [27], [28], [29], how Npr2 affects growth cone behavior is unclear, as CNP can also act as a chemoattractant [28]. In addition, the full extent of Npr2's effects on the organization and function of the mature nervous system has not yet been defined. Since NPR2 is mutated in human patients with achondroplasia [30], a deeper understanding of the Npr2 phenotype in an animal model is needed. Here, we seek to gain insight into the mechanisms that govern SGN central wiring by characterizing the long term effects of Npr2 mutations on both the spatial organization and functional maturation of synaptic connections between SGN axons and their cochlear nuclear targets.
Previous studies established that Npr2 is absolutely required for bifurcation of sensory neuron axons with no obvious effects on the peripheral processes [24], [25], highlighting the utility of the Npr2 mutant mouse as a model for central auditory wiring defects. However, the peripheral auditory system has not yet been examined. To investigate whether there are any obvious changes in the innervation of the mature cochlea, we evaluated the gross organization of SGN peripheral processes after the onset of hearing, which is at about postnatal day 12 (P12) in mice [31]. Neurofilament immunostaining revealed no obvious differences in the overall pattern of cochlear innervation at P18, with orderly arrays of radial bundles present in both Npr2 mutants (n = 2) and controls (n = 2) (Fig. 1A,B). Moreover, individual SGN fibers, as visualized at P14 by crossing Neurog1-CreERT2 to AI14-tdTomato reporter mice, exhibited normal morphologies, with single, unbranched peripheral processes extending directly towards the organ of Corti both in controls (n = 3) and mutants (n = 5) (Fig. 1C,D). Similarly, the SGN central processes in the auditory nerve peripheral to the bifurcation in the nerve root seemed normal in Npr2 mutant mice at P21, as assessed by electron microscopy (Fig. 1E, F). The g-ratio, the ratio of the axon diameter to the total myelinated fiber diameter, did not differ significantly between control (n = 3) and Npr2 mutant (n = 3) animals (P = 0.87) (Fig. 1G). In both groups, the observed ratio was near the optimal for conduction [32]. Hence, development of the peripheral auditory system appears normal in the Npr2 mutant strain, which therefore offers a useful model for examining the anatomical and functional consequences of central auditory wiring defects.
Previously, we showed that Npr2 is expressed in SGNs and is required for the bifurcation of central SGN axons at E12.5 [21], and independent studies have shown a complete absence of axon bifurcation in these and other sensory neuron populations in Npr2 mutants [24], [25]. Although DRG bifurcation defects have been shown to persist and ultimately disrupt the functional connectivity of the mature spinal cord [24], the long term effects of the Npr2 mutation for central auditory wiring remain poorly characterized. To determine whether SGN central axons acquire additional defects, we characterized Npr2 mutants at E16.5, when both branches have formed and projected tonotopically within the developing cochlear nuclei (Fig. 2A) [33]. Lipophilic dye labeling revealed that in Npr2 mutants (n = 4), the cochlear nerve root lacked the Y-shaped morphology typical of control animals (n = 4) (Fig. 2B,C), consistent with a persistent bifurcation defect. In addition, whereas SGN axons were neatly bundled both proximal and distal to the bifurcation of the nerve in control animals, Npr2 mutant axons were disorganized in the region where they would normally branch (Fig. 2B, arrow), resembling the first exploratory SGN axons that reach the hindbrain at E11.5 in wild-type embryos [21].
We next examined SGN central projections in Npr2 mutant animals between P14–P18, after the auditory circuit is fully formed and functional hearing has begun. To label SGN axons, Neurog1-creERT2 mice were crossed to AI14 tdTomato reporter mice, resulting in offspring with tdTomato expression in a random subset of SGNs due to leaky Cre activity in the absence of tamoxifen. In Neurog1-CreERT2;AI14 mice, tdTomato expression in SGNs is sparse enough that neuronal morphology can be examined, but is distributed uniformly along the length of the cochlea so that the overall pattern of SGN innervation throughout the intact cochlear nuclear complex can be visualized and qualitatively assessed in cleared tissue. In control animals (n = 5), auditory nerve fibers projected in a highly stereotyped and ordered manner to the aVCN, pVCN, and DCN (Fig. 2D). In contrast, SGN afferent innervation was consistently disrupted in all Npr2 mutants (n = 6) (Fig. 2E). SGN axons were able to reach all three divisions, but exhibited several signs of disorganization that were not observed in controls. First, whereas control axons formed distinct bifurcations that aligned with each other within the nerve root, we could not recognize an obvious zone of bifurcation in Npr2 mutants. Moreover, a closer look at sections through the cochlear nuclei revealed a severe disorganization of projections in the aVCN and pVCN in Npr2 mutants (Fig. 2G) compared to controls (Fig. 2F). Control axons exhibited clear bifurcations in the VCN (Fig. 2H), resulting in neat bundles of axonal branches in the aVCN (Fig. 2H′). In contrast, Npr2 mutant SGN axons generally turned instead of bifurcating (Fig. 2I) and often followed aberrant paths in the aVCN (Fig. 2I′). Additionally, some patches of the aVCN appeared to be underinnervated (Fig. 2E, arrowheads). Thus, loss of axon bifurcation and abnormal trajectories persist even after the onset of hearing in Npr2 mutant mice.
Although the overall pattern of innervation was clearly abnormal, SGNs projections were nevertheless present in all divisions of the Npr2 mutant cochlear nuclei (Fig. 2E). Given that Npr2 is required for axon bifurcation but not for development of collaterals [24], [25], we hypothesized that unbifurcated SGN axons might eventually form interstitial branches that are able to grow into other regions of the cochlear nuclei. To determine whether SGNs innervating the aVCN can still form branches that project to the pVCN, we labeled fibers with biocytin injections into the aVCN and searched for labeled fibers in the pVCN. In control animals (n = 49), such injections labeled the ascending branch retrogradely to the nerve root and they labeled the descending branch anterogradely through the pVCN and into the DCN (Fig. 3A). The labeled descending branches formed a tight bundle in the octopus cell area, where the SGN fibers converge on their way to the DCN (Fig. 3A′). In Npr2 mutant animals (n = 28), obvious bundles were never seen. However, a few widely spread fibers in the octopus cell area were consistently labeled (Fig. 3B,B′), indicating that some individual SGN fibers managed to innervate both regions in spite of the axon bifurcation defect. Since there are no molecular markers to distinguish bifurcations from interstitial branches, we instead relied on morphological criteria to recognize bifurcations. Bifurcations are usually the first branch points within the cochlear nuclei, are found in a predictable location, and exhibit a characteristic Y-shaped morphology. No branch point in any of the 34 mutant cochlear nuclei in which biocytin was injected into the aVCN or into the cut end of the nerve displayed the morphology of normal bifurcations. Whereas in controls, the parent axon gave rise to two equally thick branches at roughly 120° angles (Fig. 3C), branches in the vicinity of the nerve in Npr2 mutants exhibited more varied angles and one branch was often abnormally thin (Fig. 3D, arrows). Additionally, Npr2 mutant SGN axons extended branches (Fig. 3G, arrowheads) that resembled the interstitial branches found in controls (Fig. 3E, arrowheads). Thus, it seems likely that the branches that SGN axons form in Npr2 mutant mice are interstitial branches rather than true bifurcations. Nonetheless, axonal branches in the aVCN terminated in endbulbs of Held with the usual range of sizes and shapes in both controls (Fig. 3F) and mutants (Fig. 3H), indicating that despite their defective branching patterns and trajectories, Npr2 mutant SGN axons are still able to find appropriate targets and make specialized synapses with normal morphology. Thus local interactions seem to be able to govern synaptogenesis independent of the changes in axon trajectory.
Since the peripheral organization of SGN projections in the cochlea appeared unaffected in Npr2 mutants, SGNs are predicted to receive sharply tuned frequency information from hair cells. However, given the disorganization of SGN central axons in Npr2 mutants, we wondered whether mutant SGN axons preserve the tonotopic order of their projections as they exit the cochlea and find their way into the cochlear nuclei. Crystals of the lipophilic dyes, DiI and DiD, were inserted into the apical and basal turns of the cochlea in fixed E16.5 mouse heads and the dye was allowed to diffuse anterogradely through SGN axons to the hindbrain (Fig. 4A). In control animals (n = 2 wild-type and 2 heterozygote), SGN axons were tonotopically segregated within the eighth nerve, and their bifurcation points fanned out in tonotopic order within the developing cochlear nuclei, with axons from more basal SGNs bifurcating more dorsally than apical SGNs (Fig. 4B). The gross tonotopic segregation observed in control embryos was maintained in Npr2 mutants (n = 4) (Fig. 4C,C′). However, in some Npr2 mutant embryos (n = 2/4), intermingling of apical and basal projections was observed (Fig. 4C,C′, arrowheads), suggesting imprecise tonotopy. Additionally, mutant axons appeared to project more strongly towards what will become the pVCN and DCN, quantified by comparing the fluorescence intensity of the branches projecting rostrally vs. caudally (P<0.05) (Fig. 4D). Apical SGNs were more strongly affected than basal SGNs and showed a stronger bias towards the developing pVCN and DCN.
To determine whether the blurring of tonotopy persists through the onset of hearing, similar dye labeling of SGNs was performed at P14 by placing DiI and DiD crystals in the apical and mid-turns of the cochlea, respectively (Fig. 4E). In controls (n = 2 wild-type), clear segregation of the two dyes was observed in the eighth nerve (Fig. 4F) and this segregation was maintained both in the aVCN (Fig. 4F′) and pVCN (Fig. 4F″), as assessed using confocal imaging. In Npr2 mutants (n = 4), the axons from apical and mid-turn SGNs were also appropriately segregated within the eighth nerve (Fig. 4G, H), confirming that the dyes labeled distinct populations of neurons in the cochlea. However, the projections overlapped extensively in the aVCN (Fig. 4G′, H′) and/or pVCN (Fig. 4G″, H″). Some overlap was apparent in all of the mutants; variability in precise size and location of the dye crystals prevented quantification of the degree of mixing. Thus, tonotopic segregation appears normal in the auditory nerve, but is degraded within the cochlear nuclei of Npr2 mutants.
The abnormal tonotopic organization of SGN projections raised the question of whether intrinsic neuronal circuits within the cochlear nuclei are similarly disrupted. Tuberculoventral (TV) cells are glycinergic neurons that reside in the deep layer of the DCN and innervate targets in the aVCN and pVCN, forming a negative feedback circuit. They are tonotopically arranged, receiving input from the same auditory nerve fibers as their targets and therefore exhibit similar tuning [1], [2], [34]. The pattern of TV cell connectivity was examined by injecting biocytin into the aVCN, which normally labels TV cell bodies in the DCN as well as SGN afferent fibers that project to that isofrequency band [1] (Fig. 5A–B). Labeling follows the tonotopic organization of the cochlear nuclei: dorsal injections labeled bands of TV cells dorsally in DCN, whereas ventral injections labeled TV cells in a more ventral position in the DCN. In control animals (n = 16 wild-type and 33 heterozygote), a few labeled cells in the DCN were located ventral to the isofrequency band, because their axons crossed the injection site to innervate more ventral regions of the aVCN (Fig. 5A, arrowhead). However, labeled cells dorsal to the band in the DCN were not observed in normal animals, indicative of the sharp tonotopic organization (Fig. 5A,D). In contrast, in Npr2 mutant mice (n = 28), the labeled cell bodies were found over a large span of the DCN, even when the injections were made ventrally in the aVCN (Fig. 5B, D). To quantify this result, the distribution of labeled cells along the tonotopic axis was measured in reconstructions of 37 slices with injections into the ventral half of aVCN (Fig. 5C). In control mice (n = 26 cochlear nuclei from 12 heterozygote and 4 wild-type mice), the distribution of labeled cells aligned at their peaks showed a sharp peak that tapered ventrally toward the granule cell lamina, with an average half-width of 132±80 µm. In Npr2 mutants (n = 11 cochlear nuclei from 7 mice), the distributions lacked sharp peaks. The average distribution of labeled cells, aligned on the median, was significantly broader, with an average half-width of 276±130 µm (P<0.001), reflecting the more diffuse organization observed within individual cochlear nuclei. These findings indicate that the tonotopic organization of the TV cell projection in mutant cochlear nuclei is less precise than in control animals, consistent with the overall disruption in SGN axon topography shown by genetic and dye labeling.
Our anatomical studies show that although the innervation of the cochlea is not altered noticeably, there is a consistent and striking change in SGN central axonal innervation patterns in Npr2 mutant mice. While changes in the periphery are well-known to diminish auditory sensitivity, how a loss of precision in the organization of SGN inputs to the cochlear nuclei might affect hearing is unclear. To address this question, we compared auditory brainstem responses (ABR) in six-week old wild-type and Npr2 mutant mice. ABRs are generated by the synchronous firing of groups of aligned axons. In cats, the first large positive and negative waves reflect the firing of axons of SGNs, the second positive wave reflects the firing of neurons in the VCN that lie near the nerve root, the third positive wave reflects activation of the VCN rostral and caudal to the nerve root and the superior olivary complex, and later waves reflect the summation of activity at many stages of the auditory pathway [35]. Similar waveforms are observed in mice; it is broadly accepted that the first two peaks reflect activity in the nerve and cochlear nuclei as in cats [36], [37]. No significant difference was observed between control (n = 6 wild-type) and Npr2 mutant (n = 15) mice in the shape or amplitude of the early peaks in responses to 16 kHz tones, which activate the most sensitive regions of the cochlea in mice [38] (P>0.3 for the amplitudes of peaks one and two at all sound pressure levels) (Fig. 6A, B). The normal average ABR waveforms confirm the absence of obvious peripheral defects and suggest that the timing of firing of SGNs and of their targets in the VCN is also apparently normal. In addition, ABR thresholds did not differ significantly between wild-type and Npr2 mutant mice (P = 0.38) (Fig. 6C). Thus, within the resolution of these measurements, the sensitivity and timing of firing of auditory neurons in the brainstem seem normal in Npr2 mutants.
It should be noted that Npr2 mutant mice exhibit additional abnormalities, including dwarfism and cardiac deficits [39] that compromise their health and often cause them to die within the first postnatal month. Thus, it is possible that no obvious ABR phenotype was observed because the animals that survived to the testing stage were the healthiest and least abnormal. However, cochlear nuclear innervation defects were fully penetrant and varied only in severity. Moreover, since it is difficult to establish behavioral baselines in these animals, we were unable to use pre-pulse inhibition of the acoustic startle reflex to test for deficits in specific hearing tasks, such as frequency discrimination, gap detection, and sound localization.
Although ABRs did not reveal any significant differences in auditory responsiveness in Npr2 mutant mice, this method assesses the overall activity of the population, leaving open the possibility that individual cells may not transmit signals normally. To determine whether Npr2 mutant axons are indeed able to develop normal synapses despite the change in branching patterns and trajectory, we made intracellular recordings in slices. Cochlear nuclear neuronal responses to sound depend on the pattern of convergence of synaptic inputs, the physiological properties of those inputs, and the electrical properties of target neurons that shape the voltage responses to synaptic currents. Whole-cell patch recordings in slice preparations of the cochlear nuclei confirmed that the three principal cell types of VCN (bushy, octopus, and T stellate cells), recognizable by the differences in their intrinsic electrical properties, are present in Npr2 mutants. In Npr2 mutants as in control animals, bushy and octopus cells fire transiently in response to depolarizing current pulses, whereas T stellate cells respond with trains of action potentials that last for the duration of the depolarization, in both wild-type and mutant animals [10], [12], [40], [41] (S1 Figure). Comparison of wild-type (n = 3) and mutant (n = 4) bushy cell properties revealed no change either in the resting potential (−65±1.7 mV in wild-type vs. −66±2.1 mV in Npr2 mutant) or input resistance (92±8 MΩ in wild-type vs. 97±9 MΩ in Npr2 mutant); the properties of T stellate (n = 2) and octopus cells (n = 2) were also within the normal range. The absence of any measurable differences in the intrinsic properties indicates that mutant neurons are capable of signaling as rapidly and precisely as the wild type.
Another important determinant of acoustic signal transmission is the number of SGN inputs that contact each target neuron, which ranges from few (for bushy and T stellate cells) to many (for octopus cells). Given the abnormal trajectories of SGN axons seen within the cochlear nuclei, we asked whether SGNs would still converge normally on principal cells of the VCN in Npr2 mutants. The number of excitatory inputs that converge on a recorded cell can be estimated by measuring the growth of synaptic responses to shocks of fiber bundles as the shock strength is gradually increased because the synaptic response grows in steps as additional fibers are brought to threshold [16]. The number of steps in the increase in synaptic current is thus an estimate of the number of excitatory inputs. Bushy cells receive converging input from a small number of SGNs. In the mutants, as in control animals, some jumps were small and others were large, reflecting the fact that bushy cells in mice receive input from small bouton endings as well as large endbulbs of Held [7], [16], [42] (Fig. 7A). The number of converging inputs to bushy cells in Npr2 mutants fell into the normal range, between 1 and 6 [16]. However, a surprisingly large proportion had only a single input (5/7 in Npr2 mutants, compared with 5/21 similar bushy cells in a wild type strain [16]. Interpretation of these findings is complicated by the fact that there are multiple types of bushy cells: globular bushy cells that project to the medial nucleus of the trapezoid body and spherical bushy cells that project to the lateral or medial superior olivary nuclei. The sole electrophysiological distinction between these cell types in slices is the number of converging inputs [7], [16]. Especially when the aVCN is disorganized, it is impossible to know whether populations of different types of bushy cells were sampled equally. However, our results suggest that bushy cells in the aVCN in Npr2 mutants likely receive input from fewer SGNs than normal.
To see whether other target neurons in aVCN also receive fewer inputs, we performed a similar analysis of T stellate cells, which are present both in aVCN and pVCN. Shock-evoked synaptic responses in T stellate cells normally grow with between five and eight steps, each delivering roughly equal steps of current of between 100 and 300 pA [16], [43] (Fig. 7B). Many of these responses are likely to arise from SGNs but some could also arise from other T stellate cells [43]. In control mice, no differences have been reported between T stellate cells in pVCN, where they are most abundant, and in the aVCN [40], [44], [45]. In Npr2 mutants, 6/10 of the T stellate cells we recorded were in the pVCN, near the octopus cell area. Convergence of inputs in these cells was normal, averaging 7.5±1 (n = 6). In contrast, in the 4/10 T stellate cells that were recorded more anteriorly, evoked responses grew in significantly fewer current steps (3.5±0.6, n = 4) (P<0.001) (Fig. 7B).
Octopus cells reside in pVCN and would therefore not be expected to show a similar change in the number of SGN inputs. However, these cells are so heavily innervated by SGNs that it is not possible to estimate the actual number from the growth of synaptic responses with shock strength [12], [13]. Instead, the synaptic responses generally grow in steps so small that the growth appears graded. In Npr2 mutant mice, the growth of synaptic responses showed more irregularity than we have observed in CBA or ICR mice [12], [16], but this irregularity was also observed in control mice. No differences in convergence between wild-type and mutant mice could be resolved in octopus cells (Fig. 7C).
Together, these data suggest that in Npr2 mutants, convergence of SGNs onto bushy and T stellate cell targets in the aVCN is reduced, while in the pVCN, convergence onto T stellate cells and octopus cells is normal. This subtle change in circuit organization is consistent with our finding that SGN projections are biased towards the pVCN and DCN at embryonic stages.
To determine whether the observed changes in the pattern of connectivity are accompanied by changes in the nature of transmission between SGNs and their cochlear nuclear targets, we examined the pattern of synaptic responses to trains of shocks. In wild-type mice, repeated stimulation of the auditory nerve consistently evokes synaptic responses, although when driven at high rates, synaptic responses show depression, with a stronger effect in bushy than in T stellate cells [15], [16], [46]. Synapses between SGNs and principal neurons in the VCN in Npr2 mutants (22 cells from 22 animals) exhibited the expected synaptic depression observed in wild-type and heterozygous animals (21 cells in 10 wild type and 11 heterozygote mice) (Fig. 8A, A′). However, Npr2 mutants differed from control animals in that shocks intermittently failed to evoke any responses in some neurons. For instance, in 7/12 bushy cells, some of the shocks in a train failed to evoke a response (Fig. 8A). Failures were sporadic and complete, with no synaptic response at all in the target neuron (Fig. 8A″). A similar phenotype was also detected in T stellate cells (Fig. 8B), with 3/10 cells sporadically failing to respond to shocks; in contrast, 0 of 10 wild type and heterozygote responses failed. One reason failures may have been detected in relatively fewer T stellate cells than in bushy cells is that failures of small inputs are difficult to detect. Indeed, in T stellate cells failure was often incomplete in that small (<10%) synaptic current remained, presumably because the larger of two inputs failed while the smaller one did not.
For both bushy and T stellate cells, failures occurred even after the first shock in the train, when depletion of neurotransmitter is not an issue (Fig. 8C). Together with the all-or-none character of the failures, these findings suggest that action potentials sometimes fail to invade the SGN synaptic terminals. To test whether a conduction block could be overcome by making action potentials in the parent axon taller and/or wider, we applied a low concentration of 4-aminopyridine (4-AP), a non-specific blocker of K+ channels used to relieve conduction block in patients with multiple sclerosis [47]. Indeed, 0.1 mM 4-AP eliminated synaptic failures reversibly in both bushy and T stellate cells (Fig. 8D, E). These results support the idea that Npr2 mutant auditory nerve axons suffer from blocks in action potential conduction. Importantly, the responses that did occur showed normal, precise temporal tracking of inputs (Fig. 7). Thus, our data indicate that Npr2 mutant mice exhibit altered spatial organization of the auditory circuit and less reliable action potential conduction, yet still maintain the overall temporal precision of auditory signal transmission.
The sense of hearing depends on accurate transmission of frequency and timing information from the cochlea to the brain by SGNs. Hence, SGN projections are organized spatially according to frequency and form synapses that preserve the timing of sound stimuli. Using a combination of anatomical and physiological methods, we find that these two fundamental characteristics are differentially affected in Npr2 mutant mice, which exhibit blurred tonotopy in the cochlear nuclei but still form functional connections with largely normal electrophysiological features. Although there is a slight reduction in the convergence of inputs and occasional failures in transmission, the timing of neuronal firing at a population level and sound detection thresholds, as assessed by ABR, do not differ significantly between Npr2 mutant and control mice. Taken together, these data indicate that central auditory circuits with defective spatial organization are still capable of normal signal transmission and hence auditory responsiveness, though it is unlikely that auditory processing in Npr2 mutant mice is entirely normal. These findings highlight the importance of Npr2 for central auditory circuit assembly and underscore the challenges of understanding the genetic basis of central auditory processing disorders.
Although recent studies have uncovered a number of genes required for cochlear wiring [48], how SGN central axons navigate to the cochlear nuclei is poorly understood. SGN axons reach the hindbrain and start to bifurcate by E12 in mice [21]. When the aVCN is not present, SGN axons still project to the brainstem and bifurcate [49], likely because the Npr2 ligand CNP is expressed along the entire rostral-caudal axis of the hindbrain [23]. In addition, since SGN axons enter the hindbrain at the level of rhombomere 4, which gives rise to the pVCN and DCN [50], this region may provide attractive cues that are primarily responsible for early SGN guidance decisions. Indeed, we find that Npr2 mutant axons preferentially extend towards the developing pVCN and DCN, indicating that when required to make a directional choice without bifurcating, SGN axons show a caudal bias. Nevertheless, their projections follow aberrant trajectories, suggesting that Npr2 is also required for normal responsiveness to cues in the environment. A direct role for Npr2 in axon guidance has not been clearly shown; although Npr2 mutant DRG axons make occasional guidance errors upon entering the spinal cord, they follow grossly normal paths towards targets in the dorsal and ventral horns [24]. Moreover, expression of CNP is restricted to the dorsal neural tube embryonically [23], [25], making it improbable that a CNP-Npr2 interaction directs growth of SGN axons deeper within the developing cochlear nuclei. It therefore seems more likely that the observed guidance defects are secondary to the loss of bifurcation.
After bifurcating, developing SGN axon branches must navigate towards distinct regions of the cochlear nuclei while retaining the tonotopic organization established in the cochlea. This fundamental feature of the auditory pathway is established during embryogenesis [33], [51] and does not require hearing, although it is later refined by activity-dependent mechanisms that need not be driven by sound [52], [53]. We still know very little about the molecular mechanisms that direct this critical feature of central auditory circuit assembly, and the few examples of mutant mouse strains with disrupted tonotopy are complicated to interpret. For example, tonotopy is abnormal in mice lacking the transcription factor Neurod1, which acts early during SGN development [54]. However, in these animals, auditory afferents are intermingled with projections from the vestibular endorgans, suggesting that disrupted tonotopic organization is secondary to a general change in neuronal identity. Eph/Ephrin signaling may play a more specific role in central topographic projections, with ephrin-B2 mutants exhibiting abnormally broad frequency bands in the DCN [55]. However, since EphA4/ephrin-B2 signaling is also involved in bundling of peripheral SGN projections extending towards the organ of Corti [56], the change in frequency responses in these mutants could also arise from peripheral disorganization.
Npr2 mutant mice are especially interesting because defects in the tonotopic organization in the cochlear nuclei occur without obvious defects in cochlear organization. SGN axons are topographically ordered in the eighth nerve in Npr2 mutants, but exhibit disorganization in the nerve root and blurred tonotopy in the cochlear nuclei, indicating that trajectories become disarrayed as they enter the auditory brainstem. The phenotype was fully penetrant, with abnormal topography apparent in genetically labeled SGNs, by labeling of the SGN axons with lipophilic dyes, and by biocytin labeling of second order neurons in the cochlear nuclei. Since SGN bifurcation points are tonotopically ordered within the nerve root, with small bundles of SGN axons bifurcating together, it is possible that the abnormal guidance of Npr2 mutant axons exiting the eighth nerve disrupts this bundling, thereby perturbing the local tonotopic order. Indeed, proper fasciculation during axon guidance is known to play a key role in topographic mapping of axons in other systems [57], [58]. Alternatively, mutant axons may be unable to detect guidance cues in the environment, perhaps because key receptors are not trafficked properly to the branches that do form. In fact, tonotopy worsens over time, with only a few misrouted axons at E16.5 but apparent overlap at P14, as would be expected if the primary problem is defective guidance of the collaterals that sprout from the unbifurcated axons. Higher resolution labeling methods will be needed to discern whether individual fibers extend their primary axons to the tonotopically appropriate location, with blurring due to abnormal guidance of collaterals outside of this area. It is tantalizing to consider whether the disorganization of TV cell projections might also be secondary to the changes in SGN axon branching patterns. However, although expression is initially restricted to sensory neurons [24], [25], Npr2 also appears to be transcribed in the cochlear nuclei later in development [59] (Allen Mouse Brain Atlas) and could act independently in other populations of neurons. Analysis of cochlear-specific Npr2 conditional knock-outs will be necessary to resolve this issue.
Unexpectedly, the loss of tonotopic organization in the central auditory circuits of Npr2 mutant mice results only in subtle changes in auditory function. ABRs, which are generated by the summation of coherent currents and thus largely reflect synchronous firing in bundles of axons [35] are normal in the mutants; intracellular recordings, which assay transmission to individual post-synaptic targets, show that evoked excitatory postsynaptic responses are normal when present but occasionally fail. The absence of any obvious ABR defect in Npr2 mutants is consistent with our anatomical studies, which revealed no abnormalities in peripheral wiring, myelination, axon diameter, or synaptic morphology. It is also consistent with whole-cell patch-clamp recordings from individual neurons, which show that the principal cells of the VCN in Npr2 mutants retain the ability to signal rapidly and with temporal precision. Indeed, the basic features of synaptic transmission were unaffected; EPSC amplitudes, kinetics, depression, and delays in the VCN of Npr2 mutants were in the normal range [16]. The only salient defect observed was an occasional failure in transmission, which would not be expected to alter the timing of signaling in the population of neurons. Moreover, since amplitudes of the first wave did not differ significantly between mutants and controls, either roughly similar numbers of neurons are activated or the smaller heads of mutants compensate for slightly reduced numbers of active neurons. Normal ABRs are also observed in animals in which reorganization of central tonotopic maps is induced by persistent, moderate noise [60]. While this might mean that ABRs are not sensitive enough to detect such changes, it may also reflect the plasticity of central auditory circuits, as has also been described by others [52]. Overall, these physiological studies suggest that Npr2 mutant SGNs are still able to respond to sounds with normal sensitivity and timing, despite the disrupted spatial organization. Thus, in contrast to what has been observed for development of the calyx of Held [17], functionally normal synapses can form even when SGN axons follow abnormal paths.
Although our physiological tests revealed no significant change in auditory responsiveness, it is still unclear whether the blurring of spatial organization represents a functional blurring at the level of frequency discrimination in Npr2 mutants. In wild-type animals, the tuning of bushy and T stellate cells shows similar sharpness to that of auditory nerve fibers [11], indicating that SGNs that converge onto a single target neuron are similarly tuned. Since frequency coding is likely intact in the cochlea, activity-dependent synapse elimination, not only in the aVCN but also in TV cells of the DCN, could select for appropriate inputs with similar tuning in Npr2 mutant mice even when they are not in the correct spatial location. Thus, the broadening of TV cell isofrequency mapping in Npr2 mutants might reflect appropriate functional connections between cohorts of neurons that transmit similar frequency information, but that are no longer spatially confined to a tight band due to the disorganization of SGN afferents. Although pre-pulse inhibition of the acoustic startle response can be used in mice to test for frequency discrimination [61], such experiments are not possible with Npr2 mutants, which have dwarfism and cardiac defects, and therefore await generation of Npr2 conditional knockout animals.
Although the basic features of synaptic transmission were unaffected by the loss of Npr2 function, auditory signal transmission became less reliable, with some shocks to SGNs failing to produce any response in post-synaptic targets. Since the failures sometimes occurred in the first response of a train, they could not have resulted from the depletion of neurotransmitter. Furthermore, EPSC failures were all-or-none, indicating that some action potentials did not reach the SGN terminals. Additionally, Npr2 mutant axons showed no significant loss of myelin and did not exhibit signs of the increased spike latency or jitter associated with dysmyelinated SGNs [62]. Given the changes in SGN axon branching patterns in Npr2 mutants and our ability to reverse failures with a K+ channel blocker that strengthens action potentials, it is likely that failure occurred at branch points, which have long been recognized as being weak points in conduction [63]. Although a similar functional phenotype has not yet been described in other sensory neurons in Npr2 mutants, DRG neurons do exhibit mildly impaired ability to activate target neurons upon capsaicin treatment [24], indicating the need for a more detailed analysis of these neurons.
Overall, our results suggest that the development of the auditory circuit is robust enough that surprisingly normal synaptic connections can be made even in the face of disorganized topography. It is unlikely that Npr2 mutant mice have completely normal hearing, but more subtle behavioral tests will be required to reveal deficits. Notably, it is estimated that ∼1% of people with normal hearing sensitivity, and therefore normal cochlear function, have defects in their ability to process sound [64]. Unambiguously identifying and characterizing patients with these central auditory processing disorders has been challenging, because multimodal sensory, language, and attention deficits can accompany or mimic central auditory processing disorders, thereby complicating diagnosis [65]. Interestingly, NPR2 mutations cause achondroplasia in humans [30], suggesting that closer examination of auditory function may be warranted in such patients. The identification of central wiring defects in Npr2 and other mutant mouse strains may lead to more directed clinical analysis of hearing in human patients carrying analogous mutations and therefore improve the diagnosis and classification of these disorders in the future.
The following mouse strains were used: Neurog1-creERT2 mice [33], AI14-tdTomato mice (Jackson Laboratories, Stock Number 007908), and Npr2cn mice which carry a missense point mutation (L885R) in the guanylyl cyclase domain of the Npr2 gene that prevents the protein from catalyzing cGMP formation [39] (Jackson Laboratories, Stock Number 003913). Animals were maintained on a mixed genetic background. Mice were genotyped using previously described PCR protocols (Jackson Laboratories, [33], [39]). For timed pregnancies, embryonic day 0.5 (E0.5) was defined as noon on the day of a copulatory plug. In most cases, animals were euthanized using C02 exposure followed by cervical dislocation or anesthetized with ketamine/xylazine and then either cervically dislocated or perfused transcardially with fixative. For physiological studies, young animals were decapitated with colostomy scissors. All mice were maintained in accordance with institutional and National Institutes of Health (NIH) guidelines approved by the Institutional Animal Care and Use Committees (IACUC) at Harvard Medical School (Protocol 03611) and University of Wisconsin (Protocol M00449-0-12-12).
E16.5 embryo heads were fixed in 4% paraformaldehyde (PFA) in PBS overnight and rinsed in PBS. The cochlea was exposed so that basal and apical turns were visible. In some cases, a small crystal of DiI (Life Technologies) was placed in the base of the cochlea, while a crystal of DiD (Life Technologies) was placed in the apex. In other cases, a picospritzer was used to inject a small amount of DiI or DiD dissolved in DMSO into the base or apex of the cochlea, respectively. Tissue was incubated at 37°C in PBS for 3–4 days to allow the dye to diffuse along axons. The hindbrain was then dissected out, cleared in ScaleA2 [66] at 37°C for 1 hour, mounted on a slide, and imaged by confocal microscopy to obtain z-projection images. To determine caudal/rostral bias of projections, the bifurcation zone was demarcated with a 100 pixel (px) diameter circle, and the intensity of caudal and rostral projections was measured by defining 100 px diameter circles adjacent to this zone. The ratio of caudal to rostral projections was calculated for each image and averaged for controls (n = 2 wild-type+n = 2 heterozygote embryos) and Npr2 mutants (n = 4 embryos). Student's t-test was used to assess statistical significance.
For P14–P18 animals, mice were perfused transcardially with 4% PFA in PBS (n = 2 control and 4 Npr2 mutants). Their heads were bisected sagittally and fixed overnight in 4% PFA in PBS at 4°C. Tissue was rinsed in PBS and dissected so that the cochlea was exposed, with the brain still attached, then decalcified in 0.1 M EDTA in PBS at room temperature for 3 days. The decalcified bone covering the organ of Corti was removed so that mid and apical turns were visible, and small crystals of DiI and DiD were placed in the apical and mid-turns of the cochlea, respectively, using a 30-gauge needle to first create a small slit into which the dye crystal could be inserted. The tissue was incubated at 37°C in 4% PFA in PBS for 1 week, at which point most of the dye had diffused along projections. Since the axons are heavily myelinated at this stage and require a large amount of dye to reach the central projections in the cochlear nucleus, an additional crystal of DiI or DiD was at this time placed in the same slit, and allowed to diffuse for another week. The cochlea and cochlear nuclei were then dissected out, embedded in 5% low melt agarose in PBS, and sectioned by vibratome at 150 µm. For the cochlea, transverse sections of the cochlear nerve were collected, and for the cochlear nucleus, transverse sections of the aVCN and pVCN were collected. These were mounted on a slide and imaged by confocal microscopy (Leica SP8 X).
To examine the overall pattern of peripheral projections in the cochlea, wild-type (n = 2) or Npr2 mutant (n = 2) animals were perfused transcardially with 4% PFA in PBS, and the cochleae were dissected out and fixed overnight at 4°C, then subjected to whole-mount immunofluorescence with chick anti-Neurofilament antibody (1∶1000, Abcam), using Alexa488-conjugated goat anti-chick secondary antibody (1∶1000, Life Technologies). Cochleae were mounted in Vectashield (Vector Labs) and imaged on a Leica SP8 X confocal microscope. To visualize individual SGN peripheral processes in the cochlea, Npr2cn/+ mice were crossed with Npr2 heterozygotes also carrying the Neurog1-creERT2 and Ai14: tdTomato alleles. Since leaky Cre expression results in random, sparse recombination of the Ai14: tdTomato allele even without tamoxifen administration, this allowed us to label relatively few SGNs with the red fluorescent protein tdTomato. Cochleae from perfusion-fixed P14 animals (n = 3 heterozygous control, n = 5 Npr2 mutant) were collected and further fixed overnight in 4% PFA in PBS, then mounted on a slide in Vectashield for confocal microscopy (Leica SP8 X).
To label SGN central axons at E16.5, embryo heads were fixed in 4% PFA in PBS overnight and the cochlea was exposed. A picospritzer was used to inject DiI dissolved in DMSO into the cochlea, and then treated as described above for tonotopic dye labeling. To visualize the overall pattern of SGN projections in the cochlear nuclei at postnatal stages, Npr2cn/+ mice were mated with Npr2 heterozygotes also carrying the Neurog1-creERT2 and Ai14: tdTomato alleles. P14–P18 animals (n = 5 heterozygous control, n = 6 Npr2 mutant) were perfused transcardially with 4% PFA in PBS, and their brains were drop fixed overnight in 4% PFA in PBS. Cochlear nuclei were then dissected out and cleared overnight in ScaleA2. The entire cochlear nucleus was mounted in ScaleA2 on a glass slide and imaged using a Leica SP8 X confocal microscope. Tiled confocal stacks (∼300 µm thick) were obtained at 10× so that the entire cochlear nucleus was covered. These tiled images were stitched together by ImageJ and z-projected to generate a single, large image of the cochlear nucleus including aVCN, pVCN, and DCN. For examination of projections in just the aVCN and pVCN, animals were processed as above, and then cochlear nuclei were embedded in 5% low melt agarose in PBS and cut sagittally at 150 µm using a vibratome. Regular confocal stacks were obtained at 20× and 40× and z-projected.
To assess the morphology of auditory nerve fibers and topographic organization of tuberculoventral cell projections, biocytin injections were made into the aVCN in parasagittal slices in mice aged between P14 and P26. With a single, parasagittal cut, the cochlear nuclei were removed from the brainstem in a single “slice” of up to 400 µm either with a vibratome or with scissors. The slice was maintained in vitro as in electrophysiological experiments. With a picospritzer, normal saline containing 1% biocytin (Sigma) was injected into the aVCN through a pipette with a tip diameter of ∼5 µm. Movement of the pipette through the slice disrupted processes that crossed the injection site as pulses of pressure released biocytin. Biocytin was allowed to spread through the tissue for 1.5 to 2 hours as slices continued to be superfused with warmed, oxygenated saline. Slices were then fixed in 4% PFA, stored at 4°C, embedded in a gelatin-albumin mixture, and resectioned at 40 to 60 µm in frozen sections. Biocytin in cells and fibers was visualized with horseradish peroxidase (Vectastain ABC Elite Kit, Vector Laboratories) [12]. Photomicrographs were taken through a Zeiss Axioskop with a Zeiss Axiocam.
After being processed histologically, sections were analyzed with a camera lucida. Each section was reconstructed and marked with the locations of labeled neurons and landmarks. Landmarks were used to reconstruct slices as illustrated in Figure 5D. The distribution of labeled cells in the reconstructed slice was measured by means of a transparent grid that was laid parallel to an isofrequency band. Cells were then counted within parallel rows of squares as illustrated by the histograms in Figure 5D. Comparisons between genotypes were made by lining up peaks in histograms and summing cells in bands. Half widths were statistically compared using a one-way ANOVA test with Origin (v 7.5) software.
P21 animals were perfused transcardially with 4% PFA in PBS, and bisected heads were fixed overnight at 4°C in fixative (2.5% PFA, 5% glutaraldehyde, 0.06% picric acid in 0.2 M sodium cacodylate buffer). The cochlear nuclei were dissected out with the eighth nerve attached, and the region where the eighth nerve enters the cochlear nuclei was cut into a 1–2 mm cube in the fixative. The tissue was washed in 0.2 M sodium cadocylate buffer three times, followed by incubation in 1% osmium tetroxide/1.5% potassium ferrocyanide in water for 1 hour in the dark at room temperature. After three washes in malelate buffer (pH 5.15), the tissue was placed in 1% Uranyl Acetate or maleate buffer for 30 minutes, washed in water three times, and then dehydrated through an ethanol series (70% ethanol for 15 min, 90% ethanol for 15 min, and 100% ethanol twice for 15 min). Tissue was incubated in propyleneoxide solution for 1 hour, and then infiltrated with Epon resin mixed 1∶1 with propyleneoxide for 2–3 hours at room temperature. Samples were embedded in freshly mixed Epon and polymerized for 24–48 hours at 60°C. Thin sections were cut transverse to the eighth nerve using a Reichert Ultracut-S and were imaged using a Technai G2 Spirit BioTWIN transmission electron microscope with an AMT 2k CCD camera. To calculate the g-ratio, EM images of the eighth nerve were obtained for control (n = 3) and Npr2 mutant (n = 3) mice, and Fiji (ImageJ) was used to demarcate the area encompassed by each axon, as well as the area of the entire myelinated fiber. The g-ratio for each axon was calculated for ∼200 axons for each animal by dividing the diameter of the entire myelinated fiber by the diameter of the axon proper, and averages for controls and mutants were calculated. Statistical significance was assessed using Student's t-test.
Auditory brainstem responses (ABRs) were recorded in 6-week-old mice in a soundproof chamber, as previously described [67]. Average ABR waveforms were plotted using a MATLAB (MathWorks) script written by Ann E. Hickox in the laboratory of Dr. Charles Liberman (Eaton Peabody Laboratories, Massachusetts Eye and Ear Infirmary, Boston, MA). Statistical significance was assessed using Student's t-test.
Coronal slices of the cochlear nuclei were made from mice between P17 and P25. Slices (220 µm thick) were cut with a vibrating microtome (Leica VT 1000S) in normal physiological saline or in saline with reduced Na+ at 24–27°C, and then transferred to a recording chamber (∼0.6 ml) and superfused continually at 5–6 ml/min. Temperature was controlled with a Thermalert thermometer (Physitemp) the input of which comes from a small thermistor (IT-23, Physitemp, diameter: 0.1 mm) placed between the inflow of the chamber and the tissue. The output of the Thermalert thermometer was fed into a custom-made, feedback-controlled heater that heated the saline in glass tubing (1.5 mm) just before it reached the chamber to maintain the temperature at 33°C. Biocytin injections were made under the control of a Wild (M5) dissecting microscope. For electrophysiological recordings, the tissue was visualized through a compound microscope (Zeiss Axioskop) with a 63× water immersion objective and CCD Camera (Hamamatsu), with the image displayed on a video screen.
Whole-cell patch clamp recordings were made by using an Axopatch 200A amplifier (Axon Instruments, Burlingame, CA). Patch electrodes whose resistances ranged between 3.5 and 8 MΩ were made from borosilicate glass. All recordings of eEPSCs were digitized at 40 kHz and low-pass filtered at 10 kHz. The series resistance was compensated by 85–90% in recordings from octopus cells and by 70–80% in recordings from T stellate and bushy cells with a 10-µsec lag [16]. EPSCs were evoked by shocks through a Master-8 stimulator and Iso-flex isolator (AMPI, Jerusalem, Israel), delivered through an extracellular-saline-filled glass pipette (∼5 µm tip). Analysis of EPSCs was performed by using pClamp (Clampfit 9.0, Axon Instruments).
For solutions, all chemicals were from Sigma-Aldrich, unless stated otherwise.
Normal saline: The normal extracellular physiological saline comprised (in mM) 130 NaCl, 3 KCl, 1.2 KH2PO4, 2.4 CaCl2, 1.3 MgSO4, 20 NaHCO3, 6 HEPES, 10 glucose, and 0.4 ascorbic acid saturated with 95% O2-5% CO2, pH 7.3–7.4, between 24 and 33°C. The osmolality was 306 mOsm/kg (3D3 Osmometer, Advanced Instruments Inc, Norwood, MA). Cutting solution: Some dissections were performed in a special cutting solution that contained (in mM) 99 NaCl, 3 KCl, 1.2 KH2PO4, 1 CaCl2, 1.3 MgSO4, 20 NaHCO3, 6 HEPES, 10 glucose, and 72 sucrose.
Pipette solution: Recording pipettes were filled with a solution that consisted of (in mM) 90 Cs2SO4, 20 CsCl, 5 EGTA, 10 HEPES, 4 Mg-ATP, 0.3 GTP, 5 Na-phosphocreatine, 5 mM QX314, and was adjusted to pH 7.3 with CsOH (∼298 mOsm). Voltages were corrected for a −10 mV junction potential.
|
10.1371/journal.pcbi.1006119 | Assessing the public health impact of tolerance-based therapies with mathematical models | Disease tolerance is a defense strategy against infections that aims at maintaining host health even at high pathogen replication or load. Tolerance mechanisms are currently intensively studied with the long-term goal of exploiting them therapeutically. Because tolerance-based treatment imposes less selective pressure on the pathogen it has been hypothesised to be “evolution-proof”. However, the primary public health goal is to reduce the incidence and mortality associated with a disease. From this perspective, tolerance-based treatment bears the risk of increasing the prevalence of the disease, which may lead to increased mortality. We assessed the promise of tolerance-based treatment strategies using mathematical models. Conventional treatment was implemented as an increased recovery rate, while tolerance-based treatment was assumed to reduce the disease-related mortality of infected hosts without affecting recovery. We investigated the endemic phase of two types of infections: acute and chronic. Additionally, we considered the effect of pathogen resistance against conventional treatment. We show that, for low coverage of tolerance-based treatment, chronic infections can cause even more deaths than without treatment. Overall, we found that conventional treatment always outperforms tolerance-based treatment, even when we allow the emergence of pathogen resistance. Our results cast doubt on the potential benefit of tolerance-based over conventional treatment. Any clinical application of tolerance-based treatment of infectious diseases has to consider the associated detrimental epidemiological feedback.
| Conventional therapies improve patient health by eliminating the pathogen, or, at least, reducing its burden. Recently, alternative therapies that exploit host tolerance mechanisms have received attention from the medical community as a promising strategy. These treatments aim at reducing the level of illness due to the infection, rather than eliminating the pathogen directly. Using a mathematical model, we show that although these treatments are beneficial at the individual level, they can have undesired public health consequences. In particular we show that tolerance-based treatment gives more time for the disease to spread in the population, which in turn increase its prevalence. Moreover, in the case of a low coverage of the treatment of a chronic infection, the overall mortality can increase.
| Hosts can respond to infections in various ways. The host can reduce the pathogen replication or load and thus improve its health. In evolutionary ecology, such a response is called “host resistance”. Another possible host response is “disease tolerance”, that induces a state, in which the host, at a given pathogen load, suffers less from the negative consequences of being infected.
In evolutionary ecology, disease tolerance has received attention as a host strategy that has an impact on the evolutionary dynamics of host-pathogen systems very different to that of host resistance [1–3]. In a first scenario, when resistance is cost-free, resistance genes are predicted to fix in the population [4, 5]. Another possibility, if the pathogen population is not cleared fast enough, or if there is cost of resistance, is that resistance and susceptible genes will coexist in the host population [6–8].
Tolerance genes, on the other hand, do not drive the pathogen to extinction. They even increase the prevalence of the pathogen, thus increasing the selective pressure favoring themselves. This positive evolutionary feedback often leads to the fixation of tolerance genes [5, 9]—although scenarios explaining polymorphisms in tolerance have been considered [10].
The particular molecular and immunological mechanisms that confer host resistance are clinically relevant as they provide targets for therapeutic agents. The most common treatment agents, such as antibiotics or antivirals, aim at reducing pathogen replication or burden and are often based on host resistance mechanisms. A class of agents against HIV, for example, inhibits the coreceptor CCR5—a treatment strategy inspired by a naturally occurring polymorphism in the gene encoding CCR5 that reduces the susceptibility of individuals to HIV infection [11].
But tolerance mechanisms can also be exploited therapeutically. For example, widely used anti-inflammatory drugs reduce the negative impact of an infection without targeting the pathogen directly. Other examples of tolerance mechanisms have been found in the context of Plasmodium infections. Plasmodium infections lead to the release of heme from erythrocytes, which has proinflammatory properties. The inflammation triggers a reactive oxygen species response that can lead to liver failure in individuals with malaria. Some individuals, however, express an enzyme—heme oxygenase 1 (HO-1)—that prevents liver failure by limiting the reactive oxygen species response. Because the action of this enzyme does not affect the level of the pathogen it represents a tolerance mechanism. Inhibiting the reactive oxygen species response with a pharmacological agent that acts similarly to HO-1 has been shown to limit liver failure in mouse models [12]. HO-1 was also shown to provide host tolerance by preventing free heme from promoting severe sepsis [13]. Additional tolerance mechanisms against severe sepsis have recently been discovered [14], which involve the inhibition of cytokine production by anthracyclines. Similarly, based on the observation that sickle human hemoglobin confers tolerance to malaria, it was proposed that modulation of HO-1 via the transcription factor NF-E2-related factor 2 (Nrf2) might be a therapeutic target for treating cerebral malaria [15]. We refer to treatment strategies that are based on tolerance mechanisms as tolerance-based treatment.
Tolerance-based treatment is seen by some to have great promise [16]—in part, simply as a complement to more conventional treatment based on the exploitation of host resistance mechanisms. It is even hypothesized that tolerance-based treatment could hinder the development of drug resistance as it exerts a milder selection pressure on the pathogen [1, 17].
However, negative consequences of tolerance-based treatment have been considered too. Treated hosts could become healthy carriers of the pathogens, hence giving them more opportunity for transmission [9, 18–20]. Moreover, it was postulated that damage-limitation treatments—a form of tolerance-based treatment—might select for increased pathogen transmission [20, 21].
In an evolutionary context, theoretical studies also showed that pathogen virulence may increase in a tolerant host population [22]. In the context of public health, this translates into a risk of evolution towards higher virulence in response to tolerance-based treatment. But in the study by Miller et al. [22] this higher virulence did not affect host mortality because all hosts carried the tolerance gene.
Generally, the translation of insights from evolutionary ecology to the public health situation is hindered by the fact that tolerance genes were proposed to fix in the host population, and hosts are therefore protected from dying. Tolerance-based treatment, even if it confers great benefits to individual hosts, cannot be expected to be applied to the entire population. As a consequence, disease-induced mortality could increase when tolerance-based treatment is rolled out.
Here, we assess the promise of tolerance-based treatment (TOL) on the population level and investigate the public health impact of treatment at various levels of coverage. While most previous studies focused on changes in the incidence and prevalence of the disease, we specifically focus on the disease-induced mortality. To this end, we use mathematical models. Specifically, we extend the well-known epidemiological SIR model, featuring susceptible, infected, recovered individuals, by an additional compartment of treated individuals. We investigate the effect of TOL for two types of infections: acute and chronic.
While, at the individual level, TOL can be effective, and may even prevent the evolution of pathogen resistance in the long run, it is problematic epidemiologically. Indeed, we show that for chronic infections, the disease-induced mortality increases for low treatment coverage, because of the introduction in the population of asymptomatic carriers of the disease. Comparing TOL to conventional treatment, based on a reduction of load (ROL), we find that for both, acute and chronic infections, ROL always outperforms TOL even when we consider that pathogen resistance can emerge against ROL.
We investigated the impact of TOL on the population of hosts with a mathematical model. We based our model on the SIR model [23–25], which describes susceptible, infected, and recovered hosts. To describe treatment, we divided the compartment of infected individuals into a treated and an untreated compartment (see Fig 1). Treated individuals arise from infected untreated individuals at a given rate and remain infectious.
ROL treatment is implemented in this model as an increased recovery rate of treated individuals. Treated individuals are less infectious than untreated ones because they recover faster. Per unit time their infectiousness is not assumed to be affected by treatment. Tolerance-based treatment, on the other hand, does not affect the rate of recovery but lowers the disease-induced death rate of treated hosts. Because TOL, by definition, does not affect the pathogen load we assume that treated individuals are equally infectious as untreated ones per unit time. However, they cause more infections than untreated individuals because they live longer and thus have an extended infectious period.
We assess the effect of treatment on three epidemiological quantities: prevalence, incidence, and disease-induced mortality. We determine these quantities in the endemic equilibrium for different levels of treatment coverage. In the Methods section, we show how we calculated these quantities from our model equations. Because TOL increases the infectious period, we expect the incidence and prevalence to rise. The effect of TOL on the population-wide disease-induced mortality depends on how the higher prevalence is balanced by the lower mortality of treated individuals.
We investigate the impact of TOL and ROL for acute and chronic infections. Acute infections are modeled with influenza in mind, and are characterized by a short period of infection and a high transmission. Specifically, an untreated infection is assumed to last two weeks and the case fatality proportion is set to 1/50. Chronic infections are assumed to last years and there is no recovery, as for HIV infection. For both types of infection, the basic reproduction number R0 = 2 in the absence of treatment. We neglect seasonal fluctuations in any of the model parameters because this would preclude an equilibrium analysis.
We assessed the efficacy of treatment by evaluating numerically the incidence, prevalence and disease-induced mortality for different levels of treatment coverage. We vary treatment coverage by changing the rate of treatment, θ. We define treatment coverage as the fraction of treated hosts, which is the product of the rate of treatment times the duration of an untreated infection: θ θ + δ + v + r.
ROL reduces the incidence, the prevalence and the disease induced mortality (Fig 2A–2F). Thus, ROL is unambiguously effective on the population level, which is owed to the fact that it shortens the infectious period. For the parameters chosen here, chronic infections can even be eradicated by ROL if the coverage exceeds 55% (Fig 2D–2F).
TOL, on the other hand, affects the epidemiology very differently. Unlike ROL, TOL increases incidence and prevalence (Fig 2A, 2B, 2D and 2E). This effect is due to TOL lengthening the infectious period. The rise is much less pronounced for acute than for chronic infections. For the acute infection, the incidence increases by only 2% with the treatment coverage (Fig 2A), while, for chronic infections, it rises from 140 to 260 new infections per week (Fig 2D). Similarly, the prevalence increases by 2% in acute infections (Fig 2B), as compared to the chronic infection where the prevalence rises from 0.11 to 0.93 (Fig 2E). The more pronounced effects of TOL for chronic versus acute infections are due to recovery rate being much lower than the death rate in chronic infection. An analytical explanation is provided in Text D in S1 Text.
In acute infections, the effect of TOL on disease-induced mortality is similar to ROL. For large treatment coverage, disease-induced mortality is reduced with both treatment approaches. In chronic infections, however, TOL can even increase the disease-induced mortality when the coverage is low (Fig 2F). This is due to the fact that there is no recovery from chronic infections in our model, and treated hosts keep infecting for life. This effect is maintained even when infected hosts can recover, provided that the time to recovery is long enough. Thus, for chronic infections, the population-level disadvantages of TOL outweigh the benefits to the individual.
To assess if TOL could be a useful addition of our treatment repertoire when combined with ROL, we implemented a strategy we call TOL+ROL. Individuals receiving this combination treatment experience both, faster recovery (due to ROL) and decreased mortality (due to TOL) (see the system of Eq (1) in the Methods section). Again, we calculate the endemic incidence, prevalence and disease induced mortality under TOL+ROL and compare it to these measures under TOL and ROL alone (Fig 2).
We find that TOL+ROL does not improve on ROL in terms of incidence and prevalence but it does not do worse either—a very conceivable outcome given that individuals receiving TOL+ROL live longer. Apparently, the gain in life expectancy of individuals receiving TOL+ROL does not translate into a substantial increase of incidence and prevalence (Fig 2A, 2B, 2D and 2E) because, due to fast recovery under TOL+ROL, this strategy does not produce many asymptomatically infected individuals that increase the force of infection. With respect to disease induced mortality, TOL+ROL outperforms ROL in acute infections (Fig 2C). This effect is a direct consequence of the lower mortality rate of individuals treated with TOL+ROL as compared to ROL.
Thus, TOL can be a useful additional treatment strategy, especially for acute infections, if combined with ROL. Unlike on its own, in combination with ROL it is certainly not predicted to have negative public health consequences. It can be shown that TOL+ROL has public health benefit across the entire range of possible treatment coverage if the faster recovery outweighs the increase in life expectancy. Formally, the duration of infection in treated individuals, 1/(δ + vT + rT), needs to be smaller than that in untreated individuals, 1/(δ + v + r).
Up to this point in our analysis, TOL did not have any advantage over ROL. One supposed advantage of TOL, which we have not yet taken into account, is that it does not provoke pathogen resistance. The reason for this is that TOL does prolong rather than shorten the infectious period in treated individuals and thus increases pathogen fitness. Reduction of the pathogen load that follows ROL treatment, on the other hand, imposes a negative selection pressure on the pathogen by reducing the duration of infection. In response, the pathogen might evolve resistance.
To assess the promise of TOL more comprehensively, we included pathogen resistance to ROL into our model. To this end, we added a compartment for individuals infected with resistant pathogen strains (Fig 1). Individuals enter this compartment either after being infected and receiving treatment, which may trigger de novo resistance emergence. Resistant pathogen strains are also assumed to be transmitted. We further assume that resistant pathogens render ROL ineffective, i.e. individuals infected with resistant pathogen strains have the same recovery and disease-induced death rates as untreated infected individuals. Lastly, we allow resistant pathogens to carry a fitness cost in terms of a lower transmission coefficient.
We do not implement the emergence of resistance as a stochastic process. While this would be admittedly more realistic, treating resistance deterministically is more favorable to TOL. We thus present the best case scenario for TOL.
We find that resistance outcompetes the wildtype pathogen strain if the treatment coverage exceeds a threshold, φROL. For the parameters we chose, φROL = 0.2 for acute and φROL = 0.4 for chronic infections (Fig 3A and 3E). Below this threshold, wildtype and resistant pathogen strains coexist.
Below the threshold φROL, the incidence, prevalence and disease-induced mortality are very similar to the model without pathogen resistance. Above the threshold, the three measures under ROL and TOL+ROL become independent of treatment coverage because treatment is assumed not to affect resistant pathogen strains. The incidence and prevalence under ROL and TOL+ROL still remain below the levels they attain under TOL even if we allow pathogen resistance to evolve. However, the disease-induced mortality under TOL can become smaller than under the other treatment strategies for high treatment coverage (Fig 3D and 3H). For chronic infections, the levels of treatment coverage for which TOL becomes advantageous is much larger than the threshold φROL: TOL becomes beneficial if treatment coverage exceeds approximately 60% (Fig 3H). Some of these results depend critically on our equilibrium assumption (see Discussion).
Tolerance mechanisms are currently considered to be exploited therapeutically [16]. Undoubtedly, once developed, tested and approved, TOL will benefit treated individuals. This clear benefit of TOL should not cloud our view for what is most important: the direct public health consequences of such treatment as measured by the number of lives saved population-wide. In this paper, we assessed the potential of TOL from the public health perspective.
We find that TOL is beneficial at the population level for acute infections. However for chronic infections, irrespective of whether we take pathogen resistance into consideration, TOL on its own is not a promising treatment strategy from a public health perspective. The levels of coverage required to make mortality lower for TOL than in the absence of treatment—above 50% in our simulations—could probably be attained only for chronic infections in countries with an excellent public health infra-structure. However, if treatment is not perfect (Text C and Fig C in S1 Text), consequences of TOL are even more dramatic. The epidemiological feedback leads to an higher mortality with treatment than in the absence of treatment.
TOL is also of limited use as a supplement to other interventions. We considered the combination of TOL and ROL. Except for acute infections and low coverage, this combination is not better than ROL on its own for the population, although the addition of TOL can improve individual host health.
Because specific agents for tolerance-based treatment are still in development, a mathematical modeling approach is currently the most appropriate way to assess the public health impact of TOL. Mathematical modeling also enables us to gain insights into a wide range of different infections and to apply TOL and ROL at different levels of coverage, separately and in combination. Having said that, experimental study systems are being developed that will allow the direct assessment of the epidemiological effect of TOL. A recently developed transmission model, involving Salmonella infection of tolerant and non-tolerant mice, highlighted the role of tolerant mice in the spread of the infection [26]. While being certainly more biological, such transmission models cannot easily be scaled up to the population sizes a pathogen encounters during an epidemic. Furthermore, the treatment agents might have multiple effects that do not easily fall into the categories of TOL or ROL. Thus, even with transmission models, mathematical modeling will play a role in assessing the public health consequences of treatment.
There is a rich literature on the evolutionary ecology of disease tolerance (as reviewed by Boots et al 2009). Most mathematical models in this context focus on the evolution of tolerance as a host trait. In these models, tolerant hosts are competed against resistant hosts under pathogen pressure. The fact that tolerant hosts lead to an increase of pathogen prevalence leads to a benefit of tolerant over resistant or non-tolerant hosts. Thus, unlike in our model of tolerance-based treatment, the epidemiological feedback “biases competition” [5] in favor of tolerance as a host strategy.
A few of the evolutionary models consider the evolution of pathogens in response to host resistance or tolerance [5, 20, 22, 27, 28]. Miller et al, 2006 for example, investigate pathogen evolution in host populations that exhibit a variety of tolerance mechanisms. They find that the “absolute mortality”, i.e. the total number of deaths per unit time, can increase for virulent pathogens with the level of tolerance for both, unevolved and evolved pathogens. But the “relative mortality”—that is adjusted for the host population size and is comparable to the mortality we consider—is not found to increase (see their Fig 6). It is important to note, that in their study all hosts are tolerant, which in our model would correspond to 100% treatment coverage. The same applies to the study by Vale et al, 2014 that focused on the evolutionary response of the pathogens to various damage-limitation treatments and found that prevalence can increase and the pathogen can evolve higher virulence. While identifying important negative evolutionary consequences of TOL, these studies did not identify the conditions under which the application of tolerance-based treatment would lead to an epidemiological increase in mortality under realistic levels of treatment coverage. Of note, some models [29] consider the coevolution of tolerance in hosts and pathogen virulence. In this case, they obtain that in some cases that tolerance is not a trait shared by the whole population, but is variable in the host population.
Formally, our modeling of TOL bears most similarities with studies on the epidemiological and evolutionary aspects of imperfect vaccines [30, 31]. In particular, the anti-toxin vaccines discussed in these studies reduce the virulence in the vaccinated hosts without affecting transmission, recovery, or infection probability, and are therefore identical to TOL in terms of their effect on the infection parameters of an individual. However, anti-toxin vaccines differ from TOL in that they are given also to uninfected hosts. Thus, these vaccines are equivalent to prophylactic TOL. The major difference between prophylactic tolerance-based treatment and TOL is the number of people that need to be treated to reach similar effects. In particular, to reach a substantial decrease in the overall mortality in the context of chronic infections, the prophylaxis of a large fraction of the entire population is required. This is described in more details in Text E in S1 Text, where we establish a model of prophylactic TOL.
In contrast, our analysis focuses on a treatment given only to infected individuals, and is thus not prophylactic. Moreover, the treatment is given to a fraction of the infected population. Thus, our model addresses the impact of treatment coverage on the epidemiological feedback of TOL. In particular, we focused on the disease-induced mortality—the most relevant entity for public health for severe infections. This feedback is most pronounced at low treatment coverage where the increase in prevalence due to TOL meets a sufficient frequency of untreated, and hence vulnerable, hosts.
TOL could be useful when linked with transmission control, or if the tolerance induction goes hand-in-hand with a reduction of transmission. Sometimes such transmission reductions are a side effect of TOL [32]. Our model allows us to calculate the transmission reduction required for TOL not to increase disease prevalence: transmission has to be reduced by 2% in acute and by 88% in chronic infection (see Text F in S1 Text). The substantial reduction required for chronic infections is difficult to achieve and, once again, TOL, even with transmission control, might be useful only against acute infections (see Text G and Fig D in S1 Text). However, such transmission control might reduce pathogen fitness, and pathogen could evolve in response to TOL. While we did not consider the potential evolutionary consequences of TOL in our study, recent papers [20, 21] suggest that TOL could increase pathogen virulence, especially when virulence and pathogen fitness are tightly linked [33].
But these considerations about transmission control might still be too optimistic. It has been shown that antipyretic treatment—that aim at reducing fever during flu infection—can increase transmission [34]. This is due in part to more frequent contacts between an individual with reduced fever and susceptibles, and also to the lengthening of the infectious period [35]. Fever, although invalidating for the patient, reduces pathogen replication rates and enhances the adaptive immune response [36]. It has even been suggested that the use of aspirin during the 1918 outbreak might have caused an increase in mortality [37]. Even in the case of non-lethal infection, we can hypothesize that TOL, by increasing contact rates, will raise transmission and thus the incidence and prevalence of the infection.
Tolerance treatments might, however, be applied in hospitals, where transmission can be curbed. Of particular interest are antivirulence drugs [38, 39], because of their potential for the management of antimicrobial resistance. Tolerance mechanisms involving free heme regulation also received broad interest in the last years, as they suggest promising applications for treating severe sepsis [13], especially since available treatments are limited [40]. The targeting of free circulating heme, which contributes to pathogenesis indepently of pathogen load, is a case of tolerance-based treatment that can be applied after bacteria have been cleared. The treatment we discussed above may be applicable not only because transmission can be controlled in a hospital setting, but also because sepsis is treated after bacterial clearance.
Our results suggests that pathogen evolution could have a dramatic effect in chronic infections, where infected hosts are infectious for a long time. The mortality, that already increases because of the epidemiological feedback, will be amplified by the higher virulence of evolved pathogens. Further studies are needed to assess the public health implications of pathogen evolution in response to TOL. To go beyond the insights of Vale et al., such studies should focus on the non-prophylactic use of TOL and consider a broad range of treatment coverage.
The main disadvantage of TOL that we identified does not rely on the evolution of the pathogen but arises simply through the epidemiological feedback that is amplified by TOL. Our results raise serious doubts about the promise of tolerance-based strategies, especially when treating chronic infections.
We consider an extended SIR model in which infected patients can be treated. The model is depicted in Fig 1. Susceptible (uninfected) hosts S enter the population at a rate Λ, and die at a per capita rate δ. They can be infected by individuals that are either untreated (I) or treated (T). Susceptible hosts become infected at a rate that depends on the number of susceptible S, the total number of infected I + T and the transmission coefficient β. This reflects contact-dependent transmission from infected hosts to uninfected hosts. Infected hosts die at a rate δ + v, with v ≥ 0 indicating a higher mortality of infected than susceptible. Infected can recover at a per capita rate r and become part of the recovering population R. Similarly, treated infected die at a rate δ + vT and can recover at a per capita rate rT. Untreated infected individuals are treated at a rate θ, and become immediately treated infected. The model is summarized by the system of ordinary differential equations
S ˙ = Λ - δ S - β S ( I + T ) I ˙ = β S ( I + T ) - ( δ + v ) I - θ I - r I T ˙ = θ I - ( δ + v T ) T - r T TR ˙ = r I + r T T - δ R . (1)
We assume that infected individuals enter instantaneously the treated infected compartment at a rate of treatment θ. Thus, a fraction
φ = θ θ + r + v + δ (2)
of them become treated infected. This rate ratio can be regarded as the probability for an infected individual to be treated instead of leaving to another state (death or recovery).
For the implementation of TOL, we assume that the death rate is lower for treated than untreated individual (vTOL = vT < v), but that the hosts are still infectious (rTOL = rT = r).
For a ROL the hosts recover faster than without a treatment, but the mortality rate is the same (rROL = rT > r and vT = v).
In addition, we consider a treatment that combines the properties of TOL and ROL (TOL+ROL). We translate this effect by combining the benefits of the two treatments, the increased recovery rate of ROL and the decreased mortality rate of TOL (rTOL+ROL = rROL and vTOL+ROL = v).
In this study, we will evaluate the benefits of a treatment by considering three measures. First, the incidence, defined as the rate of new infections, is
β S ( I + T ) . (3)
Second, the prevalence of the disease, which is the ratio of infected individuals over the total population, is
I + T S + R + I + T . (4)
Finally, we will measure the disease-induced mortality, which is the fraction of deaths that are due to the disease over the total number of deaths, given by
v I + v T T δ S + δ R + ( v + δ ) I + ( v T + δ ) T . (5)
This measure quantifies the population-level consequences of treatment on the death rate. Interestingly, it is possible to express mortality from an individual perspective. The probability that a given individual starting in the susceptible compartment dies from the infection is given by the product of the probability of entering an infected compartment (treated or untreated), and then to die from the disease. It is thus
p = β ( I + T ) β ( I + T ) + δ ( v δ + v + r + θ + θ δ + v + r + θ v T δ + v T + r T ) .
A simple mathematical derivation shows that this measure is the same as the population-level disease-induced mortality.
Throughout this study, we evaluate how these three measures vary at equilibrium with the fraction of treated population. Model (1) has two equilibria: the disease free equilibrium (DFE) and the endemic equilibrium (EE). Whether one equilibrium or the other is attained depends on the relative value of the model parameters (See Text A in S1 Text for details on the equilibrium values and the equilibrium stability analysis).
Moreover, the efficacy of the treatments are evaluated for two types of infection: an acute infection, which is highly transmissible, with fast death and recovery rates, and a chronic infection, with a slower dynamics than the acute infection, and for which no recovery is possible in the absence of treatment.
In an extension of the model, we assumed that ROL can induce pathogen resistance to treatment, and added to the compartment model the number IR of hosts that are infected by pathogens resistant to treatment. Pathogens infecting treated individuals can develop resistance to treatment at a rate sR, called acquired (or de novo) resistance. Susceptible individuals can be infected by treated, untreated, or resistant hosts. If a susceptible individual is infected by a non-resistant, it becomes infected non-treated, but becomes resistant if it is infected by a resistant individual. Individuals infected by the resistant pathogen die at a rate δ + vR. The model is an extension of model (1) with an additional compartment for the infected resistant and the corresponding transition rates (Fig 1) and is described by the set of equations
S ˙ = Λ - δ S - β S ( I + T R O L + ( 1 - c R ) I R )I ˙ = β S ( I + T R O L ) - ( δ + v ) I - r I - θ R O L I T ˙ R O L = θ R O L I - ( δ + v R O L ) T R O L - r R O L T R O L - s R I R O LI ˙ R = β ( 1 - c R ) S I R + s R T R O L - ( δ + v R ) I R - r R I R R ˙ = r I + r R O L T R O L + r R I R - δ R . (6)
In our model, resistance can appear in two ways. First, individuals receiving ROL can acquire resistance at a rate sR, and second, susceptible individuals can be infected by resistant infected. Transmission of the resistant pathogen is described by a mass action law with a transmission coefficient β(1 − cR), where cR > 0 represents the fitness cost of transmission associated to pathogen resistance to treatment.
Similarly to the case without resistance to treatment, the system of equations has a disease-free and an endemic equilibrium. Additionally, there is a third equilibrium where all the infected individuals are resistant to the treatment. It can be shown analytically that there is always a threshold value of the treatment coverage such that only resistant pathogens exist at equilibrium, provided that the cost of transmission of the resistant pathogen cR is small enough. Details of the equilibrium values for the model of pathogen resistance are given in Text B in S1 Text.
Parameters in the absence of treatment: The death rate without a disease is set to δ = 1/70 years−1 and Λ is determined so that the ratio Λ/δ is equal to a fixed population size in the case of no epidemic. We choose Λ = 106δ = 1.43 104 years−1. The transmission coefficient β is chosen so that the basic reproductive ratio R 0 = Λ β δ A = 2 when no treatment is applied. We assume that TOL reduces the death rate of the disease, and that ROL increases the recovery rate.
For an acute infection: The parameters r and v are chosen so that one out of fifty infected individuals dies without treatment. Hence, v r + v + δ = 1 50. Moreover, we set the duration of the infection to be 1 r + v + δ = 2 weeks.
For a chronic infection: Infected untreated individuals die from the infection in 10 years. They do not recover from the infection in the absence of treatment, and in one year with ROL.
In S1 Text, we provide results for various sets of parameters (Figs A and B in S1 Text). We consider an extension of the model with a cost associated with treatment (Fig A in S1 Text). We also consider various values of R0 (Fig B in S1 Text). In both cases, the main conclusions of this study remain unchanged. In another extension, we treat the case of imperfect treatment, characterized by 0 ≤ vTOL ≤ v (Fig C and Text C in S1 Text).
|
10.1371/journal.pbio.1001452 | Order–Disorder Transitions Govern Kinetic Cooperativity and Allostery of Monomeric Human Glucokinase | Glucokinase (GCK) catalyzes the rate-limiting step of glucose catabolism in the pancreas, where it functions as the body's principal glucose sensor. GCK dysfunction leads to several potentially fatal diseases including maturity–onset diabetes of the young type II (MODY-II) and persistent hypoglycemic hyperinsulinemia of infancy (PHHI). GCK maintains glucose homeostasis by displaying a sigmoidal kinetic response to increasing blood glucose levels. This positive cooperativity is unique because the enzyme functions exclusively as a monomer and possesses only a single glucose binding site. Despite nearly a half century of research, the mechanistic basis for GCK's homotropic allostery remains unresolved. Here we explain GCK cooperativity in terms of large-scale, glucose-mediated disorder–order transitions using 17 isotopically labeled isoleucine methyl groups and three tryptophan side chains as sensitive nuclear magnetic resonance (NMR) probes. We find that the small domain of unliganded GCK is intrinsically disordered and samples a broad conformational ensemble. We also demonstrate that small-molecule diabetes therapeutic agents and hyperinsulinemia-associated GCK mutations share a strikingly similar activation mechanism, characterized by a population shift toward a more narrow, well-ordered ensemble resembling the glucose-bound conformation. Our results support a model in which GCK generates its cooperative kinetic response at low glucose concentrations by using a millisecond disorder–order cycle of the small domain as a “time-delay loop,” which is bypassed at high glucose concentrations, providing a unique mechanism to allosterically regulate the activity of human GCK under physiological conditions.
| Glucokinase is a key metabolic enzyme that functions as the body's principal glucose sensor. Glucokinase regulates the rate at which insulin is secreted by the pancreas by using a unique but poorly understood cooperative kinetic response to increasing glucose concentrations. The physiological importance of this enzyme is underlined by the fact that mutations in the glucokinase gene lead to maturity-onset diabetes of the young type II (MODY II), permanent neonatal diabetes mellitus (PNDM), and hypoglycemic hyperinsulinemia of infancy (HI). In this study, we use cutting-edge high-resolution nuclear magnetic resonance methods to understand how the kinetic properties of glucokinase contribute to glucose homeostasis. We also seek to understand how a class of recently discovered small-molecule drugs, which hold promise as therapeutics for type 2 diabetes, function to enhance glucokinase activity. Our results suggest that glucokinase samples a range of conformational states in the absence of glucose. However, in the presence of glucose or a small-molecule activator, the enzyme population shifts towards a more narrow, well-structured ensemble of states. Our findings provide a new model for glucokinase cooperative kinetics, which relies on a slow order–disorder transition in response to glucose concentrations. These results also reveal a universal mechanism of glucokinase activation, which may inform the development of new antidiabetic agents.
| Human pancreatic glucokinase (GCK) is the body's principal glucose sensor [1]. GCK is a 52 kDa monomeric enzyme that catalyzes the formation of glucose-6-phosphate from glucose and ATP [2]. This chemical transformation represents the rate-limiting step of glucose catabolism in the pancreas, allowing GCK activity to regulate the rate at which insulin is secreted from β-cells [3]–[5]. The importance of this enzyme in maintaining glucose homeostasis is emphasized by several disease states associated with GCK dysfunction. Loss-of-function mutations, of which more than 600 have been described, cause either maturity onset diabetes of the young type II (MODY-II) or permanent neonatal diabetes mellitus (PNDM) [6]. A small number of gain-of-function mutations have also been identified in patients with the potentially fatal disease, persistent hypoglycemic hyperinsulinemia of infancy (PHHI) [7]–[19]. Functional characterization indicates that most PHHI-associated variant enzymes display a reduced level of positive cooperativity toward glucose. Interestingly, many of the PHHI-associated substitutions co-localize to a common region of the GCK scaffold that is distant from the active site [7]–[19]. In recent years, pharmaceutical research has directed significant efforts toward developing synthetic allosteric activators that target GCK as a strategy to lower blood glucose levels in type 2 diabetic patients [20]. At least one GCK activator has advanced to phase II clinical trials as an antidiabetic agent, yet the functional role and molecular basis of action of such molecules is poorly understood [21]. Similarly, the mechanism of hyperactivity associated with PHHI-linked GCK variants is unclear.
GCK's unique kinetic features are responsible for its physiological role in maintaining glucose homeostasis. The steady-state velocity shows a sigmoidal dependence upon glucose concentration, with a Hill coefficient of 1.7 [2]. This positive cooperativity enables GCK to sensitively respond to small perturbations in blood glucose concentrations, whereby the midpoint of this steady-state response, K0.5, approximates physiological blood sugar levels (4–10 mM). The kinetic cooperativity of GCK is mechanistically distinct from other modes of allostery, which involve polypeptide oligomerization or require multiple ligand binding sites [22]–[25]. Two theoretical mechanisms—the ligand-induced slow transition (LIST) and mnemonic models—have been put forth to explain kinetic cooperativity in monomeric GCK [26],[27]. In both models, cooperativity is postulated to result from slow conformational transitions that accompany glucose binding and/or product release. The models differ, however, in the relative degree of conformational heterogeneity displayed by the enzyme at various stages along the reaction coordinate. According to the LIST mechanism, two separate catalytic cycles are possible resulting from two distinct conformations of the enzyme. In contrast, the mnemonic model postulates a single catalytically active enzyme conformation with the ability to “remember” its active state for only a short period of time following turnover. Although data have accumulated over the years in support of each mechanism, a consensus has yet to be reached [28]–[33].
X-ray crystallographic structures reveal that GCK adopts a prototypic hexokinase fold, consisting of a small and large domain separated by a variable opening angle dependent upon substrate association (Figure 1A) [28]. The kinetic mechanism of GCK is strictly ordered [2], with glucose binding first, accompanied by a large structural rearrangement from an open to a closed structure (Figure 1B). Binding of ATP does not induce significant additional changes to the structure of the glucose-bound state (Figure 1C) [34]. The crystal structures also identify a common binding site for synthetic GCK activators at a location within the small domain that is approximately 20 Å removed from the glucose binding site.
Differences observed between available crystal structures demonstrate the ability of GCK to sample discrete conformations, but they do not explain the dynamic basis of allosteric regulation as set forth in the LIST and mnemonic models [28],[29],[34]–[38]. Herein, we describe the results of the first investigation of the functional dynamics of GCK at atomic detail by high-resolution NMR of specifically labeled side chains. We characterize the conformational dynamics of the wild-type enzyme as it progresses along the reaction coordinate and explain the molecular mechanism of kinetic cooperativity. We also uncover the molecular basis of activation observed in PHHI-associated variants or when a synthetic activator associates with the enzyme. Our findings suggest a model for GCK cooperativity involving a slow disorder–order cycle of an intrinsically disordered domain that is operational under low glucose concentrations but that is bypassed at elevated glucose concentrations.
The large size and dynamic nature of GCK, combined with the poor spectral dispersion even upon perdeuteration, prevented the use of traditional NMR approaches for backbone resonance assignment and structural analysis [32]. Instead, we focused on selected side chains only, a strategy that has proven effective for studies of other challenging protein systems and large complexes [39],[40]. For this purpose, we introduced 13C spin probes in the Cδ1 methyl groups of 17 isoleucine side chains and 15N spin probes at the Nε sites of three tryptophan side chains [39]–[41]. The labeled residues are quite uniformly distributed throughout the enzyme's structure. The large domain contains one Trp and 10 Ile residues, while two Trp and seven Ile residues reside in the small domain (red and yellow spheres in Figure 1A–C). Together these probes permit the study of internal dynamics throughout the protein. Importantly, the methyl resonances can retain narrow line widths even in high-molecular weight proteins through the methyl-TROSY effect manifested in the 2D 1H-13C HMQC experiment used here [39],[40]. Site-specific assignment of Ile and Trp resonances was achieved by single-site substitutions with Val and Phe, respectively, followed by the recording of 2D 1H-13C HMQC and 1H-15N HSQC spectra to identify missing cross-peaks (Figure S1).
To achieve a global portrait of the enzyme's structure and dynamics prior to glucose association, we collected 1H-13C HMQC spectra of 13CH3-Ile labeled GCK. The 1H-13C HMQC of unliganded GCK displays a high degree of cross-peak overlap and heterogeneous peak intensities (Figures 1D and S2). This behavior is consistent with previously reported, unassigned 1H-15N TROSY spectra of GCK specifically labeled on Ile backbone atoms [32]. Nine Ile residues located in the large domain and two Ile side chains situated in the small domain are observable in unliganded GCK. The 2D 1H-15N HSQC spectrum of unliganded 15N-Trp labeled GCK reveals only a single cross-peak. Mutagenesis demonstrated that this cross-peak is dominated by contributions from W167 (Figure S3). Notably, the only small domain residues observable in the unliganded spectra—I159, I163, and W167—all reside in a loop that is absent in the X-ray structure of unliganded GCK. The sharp cross-peaks originating from these three residues, along with the nearly degenerate chemical shifts of the Ile 159 and 163, reflect the presence of extensive rapid motional averaging and indicate that this loop is disordered, both in the crystal and in solution (Figure S4) [42],[43].
To investigate whether the absence of the resonances of five isoleucines belonging to the small domain results from motions of these side chains occurring on the intermediate exchange regime (µs – ms time scale), we collected 1H-13C HMQC spectra of unliganded GCK at variable temperatures. Across a range of temperatures (283–313 K) compatible with retention of enzyme activity, no additional cross-peaks were observed that could be attributed to the five missing small domain isoleucine side chains (Figure S5). We tested whether the five missing isoleucine cross-peaks might reside in the intrinsically disordered side chain region of the 1H-13C HMQC spectrum, underneath the sharp, intense peaks of I159 and I163, by constructing a variant in which I159 and I163 were replaced with leucines. These substitutions had a minimal impact upon the kinetic properties of the enzyme (Table S1); however, the removal of I159 and I163 did not eliminate all peak intensity in the region of the spectrum where disordered side chains are expected. Therefore, residual intensity observed in the I159L–I163L double variant could stem from some of the remaining isoleucine side chains in the small domain (Figure S6). Together these results show that the majority of the isoleucines of the small domain are subject to conformational exchange with an exchange rate kex that fulfills the NMR coalescence condition kex≈2.2Δv, where Δv is the chemical shift change between conformational substates of unliganded GCK. At least two isoleucines, I159 and I163, remain fully disordered in the unliganded form of GCK.
Upon addition of glucose, the intensity of the intrinsically disordered region of the 1H-13C HMQC spectrum decreases and new cross-peaks originating from I110, I126, I130, I189, and I211 become clearly visible (blue peaks in Figure 1E). In contrast, residues in the large domain are much less affected by glucose binding (gray peaks). In the presence of glucose, the cross-peaks of I159 and I163 dramatically increase their line widths and shift to new positions (Figures 1E and S2), reflecting the formation of the 151–179 β-hairpin, a well-defined structural element in the crystal structure 1V4S [28]. In the 1H-15N HSQC spectrum of the 15N-tryptophan-labeled GCK–glucose complex, three well-resolved cross-peaks corresponding to W99, W167, and W257 are observed (Figure 2G), consistent with the glucose-induced structural organization observed in the 1H-13C HMQC spectra of 13CH3-Ile labeled enzyme. Addition of the ATP analogue AMP–PNP to the binary complex perturbs the spectrum only slightly, indicating that AMP–PNP weakly impacts the structure and dynamics of the binary complex (Figure 1F).
Although PHHI is usually associated with single amino acid replacements, we employed our previously identified hyperactive quadruple variant to fully characterize the structural and functional impacts of activating mutations [44]. This hyperactive variant contains a redesigned α13-helix with sequence ALIAAAV. Similar to other activating PHHI-linked variants, the α13-helix variant displays a decreased glucose K0.5 value, a reduced level of cooperativity, and an increased equilibrium affinity for glucose compared to wild-type GCK [44]. Specifically, the glucose KD value of the α13-helix variant is 50 µM, a value similar to glucose Km values of the noncooperative GCK isozymes, hexokinases I–III. In contrast to wild-type GCK, the 1H-13C HMQC of the unliganded α13-helix variant displays cross-peaks originating from Ile residues in both the small and large domains (Figure 2E). Under saturating glucose conditions, the α13-helix variant exhibits a spectrum that closely resembles the wild-type glucose-bound spectrum (Figure S7). These spectral characteristics are not specific to the hyperactive α13 variant, as similar effects were observed with the single-site activating variant, S64P (Figure S8B). In general, the 1H-13C HMQC spectra of activated variants reveal varying degrees of structural stabilization relative to the wild-type enzyme. We observed a correlation between increases in glucose affinity caused by individual activating GCK variants, a systematic sharpening of cross-peaks, and the appearance of a larger number of resolved cross-peaks in the 1H-13C HMQC spectra in their unliganded state (Figure S8). These data demonstrate that activating PHHI-associated substitutions alter enzyme dynamics and promote a conformational ensemble that more closely resembles the glucose-bound state.
Binding of the allosteric activator, which reduces the glucose K0.5 value and eliminates cooperativity, induces changes in the 2D 1H-13C HMQC spectrum similar to that produced by activating variants (Figure 2). Well-dispersed cross-peaks originating from Ile residues in both the large and small domain are observed in the presence of the activator. The only differences between the NMR spectra of glucose-bound GCK in the absence and presence of the activator arise from residues near the activator binding site, such as I211 (Figure S9), suggesting local perturbations only. From these data, we conclude that allosteric activators stabilize the small domain and alter GCK dynamics. Our results also demonstrate that small-molecule-mediated GCK activation does not require prior formation of the binary GCK–glucose complex, a finding that could have important therapeutic implications.
The 151–179 loop is disordered in unliganded wild-type GCK, as evidenced by the fact that the chemical shifts and peak intensities of I159, I163, and W167 fall within the disordered side chain region (Figure S4). This loop becomes structured in the unliganded α13-helix variant as shown by the 2D 1H-15N HSQC and 2D 1H-13C HMQC spectra in Figure 2E and H. Notably, the chemical shift of W167 in the unliganded α13-helix variant is identical to that observed in the wild-type GCK-glucose complex (Figure 2G and H), indicating that the loop conformation adopted by the variant is similar to that produced upon glucose binding. The crystal structure reveals a possible mode of communication between the activator, the α13-helix, and the 151–179 loop (Figure 2A) [28]. One face of the activator binding site is comprised of the α13-helix residues V452, V455, and A456, and two of these residues, V452 and A456, interact with the loop residue I159. In the presence of the activator, or when an activating substitution is introduced into the α13-helix, the interactions between I159, V452, and A456 are stabilized. In turn, this stabilization is relayed to the glucose binding site via two additional loop residues, T168 and K169, which form hydrogen bonds with the O1 and O6 atoms of bound glucose. Structural organization of either the α13-helix or the 151–179 loop can be achieved by glucose binding, activator association, or a hyperactive substitution.
PHHI variants or the allosteric activator promote a population shift toward a more structured state, which involves a major reorganization of the small domain. This effect, which is evidenced by increased spectral dispersion and loss of line broadening compared to the unliganded wild-type enzyme, abrogates kinetic cooperativity. The differential glucose binding affinities of wild-type GCK and the α13-helix variant (KD = 5 mM and 50 µM, respectively) suggest that a substantial thermodynamic barrier separates the closed GCK conformation from the unliganded state. In the wild-type enzyme, this barrier is overcome by glucose binding, which triggers the disorder–order transition. In an activated variant, or in the presence of a synthetic activator, the small domain is stabilized, which decreases the free energy penalty of the conformational change and allows glucose binding energy to manifest itself in the form of a lower KD value. The differential glucose binding affinities of wild-type and activated GCK also suggest that the high-affinity conformation represents ≤1% of the total unliganded ensemble.
The experimental data presented here provide direct evidence that the small domain of GCK is highly dynamic in the absence of ligand. Some evidence of GCK structural plasticity has emerged from X-ray structures determined in the presence and absence of various ligands; however, since crystal structures represent high-resolution snapshots of protein states, they neither reveal time-scale information nor how different states are dynamically related to each other. Optical spectroscopy, on the other hand, provides time-scale information at single sites without reporting on local structure. Indeed, the application of transient state fluorescence spectroscopy to GCK revealed the existence of glucose-induced structural rearrangements that span multiple time scales, but these studies failed to provide a unified, structure-based mechanism of GCK cooperativity [45]–[48]. Here we used all native Ile and Trp side chains to probe the structural-dynamic behavior of GCK at 20 different sites distributed throughout the protein. These data, obtained in solution under physiological conditions in the absence and presence of glucose and activator, provide temporal information for wild-type GCK and its variants with high spatial resolution. Based on this new type of information, we propose a refined model for allostery in monomeric GCK.
The model that emerges from our data demonstrates that GCK cooperativity may result from dynamic structural modulations of the intrinsically disordered small domain. In the absence of ligand, the enzyme exists as an ensemble of conformations, which interconvert on a millisecond time scale that coincides with enzyme turnover (kcat∼60 s−1) (Figure 3). Upon glucose association the small domain becomes ordered, as reflected in the sharpening of the NMR resonances and the increase of chemical shift dispersion of Ile residues. After formation of the GCK–glucose binary complex, ATP binds and catalysis proceeds with little additional reorganization. Following product release, ordered unliganded GCK persists until, on the millisecond time scale, the small domain undergoes an order–disorder transition, allowing access to a “time delay loop.” Under low glucose concentrations, the delay loop is operational, leading to slow turnover and kinetic cooperativity (Figure 3, red). Under high glucose concentrations, or when GCK is activated, the delay loop is effectively bypassed, turnover is fast, and cooperativity is eliminated (Figure 3, green).
The putative involvement of slow conformational changes in the generation of kinetic cooperativity in enzymes has long been appreciated [49]. In the case of GCK, the generation of a sigmoidal steady-state response, which is key to this enzyme's sensitivity to oscillating physiological glucose levels, requires a time delay slower than 1/kcat. The absence of resonances from Ile residues located in the small domain prevents quantitative analysis of the underlying rate constants associated with the disorder–order transition. However, typical Ile and Trp side-chain chemical shift ranges point to a rate constant kex between 5 s−1 and 100 s−1 in order to explain the disappearance of the cross-peaks from the small domain by coalescence. This disorder–order transition rate satisfies the conditions needed to account for the existence of kinetic cooperativity in GCK [26],[27]. Protein structural organization occurring in the millisecond time regime is not uncommon [50], and here this phenomenon appears to contribute to the generation of GCK cooperativity. As with other intrinsically disordered proteins [51],[52], it is possible that the dynamic nature of the small domain also plays a role for the emergence of new functional attributes, including regulation of GCK activity by multiple interacting partners [53]–[55] and posttranslational events [56],[57].
Recombinant human pancreatic GCK was produced as an N-terminal hexa-histidine-tagged polypeptide in Escherichia coli strain BL21(DE3). Bacterial cultures were inoculated to an initial OD600 nm of 0.06 and were grown at 37°C in minimal medium supplemented with ampicillin (150 µg/mL), thiamine hydrochloride (25 µg/mL), 15NH4Cl (1 g/L), Ca(OH)2 (0.1 mM), MgSO4 (1 mM), and glycerol (1%) (w/v). When the OD600 nm reached 0.85, IPTG (1 mM) was added to induce gene expression and growth was continued for 12 h. The specific incorporation of 15N and 13C labels in the Trp and Ile residues, respectively, was achieved following the protocols by Muchmore et al. [41] and Tugarinov et al. [39]. Cells were harvested by centrifugation at 8,000 g, and 5 g of wet cell pellet was resuspended in 17 mL of buffer A containing HEPES (50 mM, pH 7.6), KCl (50 mM), imidazole (40 mM), dithiothreitol (10 mM), and glycerol (25% w/v). Cells were lysed using a French Press and subjected to centrifugation at 25,000 g at 4°C for 1 h. The supernatant was immediately loaded onto a 5 ml HisTrap Fast Flow Affinity Column (GE Healthcare) previously equilibrated in buffer A. Following loading, the column was washed with 10 column volumes of buffer A followed by 5 columns of buffer A containing 65 mM imidazole. GCK was eluted with buffer A containing 250 mM imidazole, and the enzyme was dialyzed at 4°C against 1L of potassium phosphate buffer (25 mM, pH 8), containing KCl (25 mM) and dithiothreitol (10 mM). GCK was then concentrated to ∼600 µM using an Amicon centrifugal concentrator (MWCO = 10,000). Protein was injected onto a Superdex 200 16/60 gel filtration column (Amersham-Pharmacia) pre-equilibrated with potassium phosphate buffer (25 mM, pH 8.0), containing KCl (25 mM) and DTT (10 mM). The gel filtration column was run at a flow rate of 0.12 mL/min, and fractions that contained the highest A280 nm readings were pooled and used immediately in the NMR experiments.
Site-directed mutagenesis was performed using the QuikChange protocol (Stratagene). Mutagenesis reactions contained human glk template DNA (400 ng), Pfu Turbo enzyme (2.5 units), cloned Pfu Turbo reaction buffer (1×), and mutagenic primers (125 ng).
GCK activity was measured spectrophotometrically at 340 nm by coupling the production of ADP to the oxidation of NADH via the combined action of pyruvate kinase and lactate dehydrogenase. Assays were conducted at 25°C in reaction mixtures containing HEPES (250 mM, pH 7.6), KCl (50 mM), NADH (0.25 mM), dithiothreitol (10 mM), pyruvate kinase (15 units), lactate dehydrogenase (15 units), ATP (0.1–50 mM), MgCl2 (1.1–51 mM), and glucose (0.05–100 mM). Data were fitted to the Hill equation or the Michaelis-Menten equation depending on the substrate under investigation. Assays were initiated by the addition of ATP and were conducted in duplicate for each substrate concentration. The kinetic constants reported are the average of data obtained from at least two independent preparations of enzyme. Assays were conducted before and after NMR measurements to verify retention of enzyme activity during the time course of the experiment.
All NMR data were collected with a Bruker Avance III spectrometer operating at 800 MHz proton field and equipped with a TCI cryogenic probe. The GCK NMR samples were prepared in potassium phosphate buffer (25 mM, pH 8.0), containing KCl (25 mM), DTT (10 mM), deuterated glycerol (5% v/v), and D2O (10% v/v). For all experiments of glucose-bound GCK, glucose was added to a total concentration of 200 mM. 1H-13C HMQC experiments were recorded as matrices of 2048×390 (in Figure 1) or 2048×256 (in Figure 2) complex data points. Unless otherwise stated, 1H-15N HSQC experiments were recorded as matrices of 2048×128 complex data points. All spectra were apodized with a cosine function in each dimension prior to zero-filling. NMR data processing and spectral analyses were performed with NMRPipe [58] and the CCPN-Analysis software [59]. As a control, the enzymatic activity of GCK was measured before and after each NMR experiment.
Site-specific assignments of Ile and Trp resonances were mainly achieved by single-site substitution with Val/Leu and Phe, respectively, followed by the recording of 2D 1H-13C HMQC and 1H-15N HSQC spectra to identify missing cross-peaks (Figure S7). Assignments of Ile residues in the α13-helix variant could be directly transferred from their wild-type assignments due to identical peak positions, except for I159 and I163, which were assigned by site-directed mutagenesis. Ile residues in the GCK–activator complex were assigned based on identical cross-peak positions with respect to the wild-type spectrum. Assignments for W99, W167, and W257 in wild-type GCK and for W167 in the α13-helix variant were performed by individually replacing Trp by Phe.
|
10.1371/journal.pcbi.1005282 | Predicting Structure-Function Relations and Survival following Surgical and Bronchoscopic Lung Volume Reduction Treatment of Emphysema | Lung volume reduction surgery (LVRS) and bronchoscopic lung volume reduction (bLVR) are palliative treatments aimed at reducing hyperinflation in advanced emphysema. Previous work has evaluated functional improvements and survival advantage for these techniques, although their effects on the micromechanical environment in the lung have yet to be determined. Here, we introduce a computational model to simulate a force-based destruction of elastic networks representing emphysema progression, which we use to track the response to lung volume reduction via LVRS and bLVR. We find that (1) LVRS efficacy can be predicted based on pre-surgical network structure; (2) macroscopic functional improvements following bLVR are related to microscopic changes in mechanical force heterogeneity; and (3) both techniques improve aspects of survival and quality of life influenced by lung compliance, albeit while accelerating disease progression. Our model predictions yield unique insights into the microscopic origins underlying emphysema progression before and after lung volume reduction.
| Surgical and, more recently, bronchoscopic lung volume reduction is the only available treatments for patients with advanced stage emphysema. Several large-scale, clinical studies have outlined appropriate selection criteria based on patient outcomes; however, the underlying mechanisms determining disease progression and response to these treatments are not well-understood. To answer this question, we have developed a network model of the lung to compare immediate and long-term response to each treatment. This approach allows us to directly study macroscopic changes in function related to microscopic changes in the local structural and mechanical environment. In addition, it facilitates direct comparisons between surgical and bronchoscopic lung volume reduction given identical initial conditions, which is not feasible in a clinical study. We propose here a mechanism suggesting that lung volume reduction efficacy is intimately linked to changes in microscopic force heterogeneity within the lung. Such an understanding of the mechanisms driving emphysema has the potential to greatly improve current therapies for this condition through more rationalized, patient-specific treatment strategies.
| Emphysema, a subtype of chronic obstructive pulmonary disease (COPD), is a progressively destructive lung tissue disease characterized by abnormal and permanent enlargement of airspaces distal to the terminal bronchioles. This largely preventable, yet presently incurable disease is associated with high morbidity and mortality, presenting a substantial burden on resource utilization [1,2]. While pharmacological treatment can have limited benefits for patients, especially in advanced stages of disease, surgical and bronchoscopic treatments aimed at reducing hyperinflated lung volumes have been shown to ameliorate symptoms of dyspnea, improve quality of life, and in certain cases provide a survival advantage [3–6].
Proposed mechanisms by which lung volume reduction results in functional improvements primarily involve a reduction in hyperinflation leading to restored chest wall and diaphragm mechanics, and an increase in elastic lung recoil (hence a decrease in lung compliance) and radial traction on airways leading to improved expiratory flow rates and lung emptying [7,8]. In lung volume reduction surgery (LVRS), emphysematous tissue is removed from the upper lung by wedge excision typically via median sternotomy or video-assisted thoracoscopic surgery. Although LVRS has proven efficacy in patients with predominantly upper-lobe emphysema and low baseline exercise capacity [9], many still do not meet the strict indications for this procedure. Moreover, widespread implementation of LVRS is hindered by high costs, a limited number of highly experienced centers, and significant post-procedural morbidity and mortality [8].
Recently, a growing number of non-surgical techniques have been developed with the goal of providing less invasive alternatives [10]. Among the more extensively investigated of these bronchoscopic lung volume reduction (bLVR) treatments are (1) nitinol coils, which return to a predetermined shape after placement in target airways while retracting the surrounding diseased lung parenchyma [5,11,12]; (2) one-way valves, which facilitate lobar collapse by blocking regional ventilation but permitting emptying of affected areas [6,13,14]; and (3) biomaterial-based lung sealants, which function to block small airways and prevent collateral ventilation pathways inducing absorption atelectasis in the delivered region [15–17]. While recent evidence appears to be trending in support of bLVR and ongoing studies are expected to further define optimal implementations for such approaches [8], the immediate and long-term effects in the lung have yet to be fully understood.
Previous work has focused primarily on the benefits in clinical function and survival outcomes for these treatments and thus, provides no insight into the micromechanical origins leading to such improvements. The aim of this study was to investigate the structure-function relationships before and after lung volume reduction using a computational elastic network model [18–22]. Here, we simulate emphysema progression by removal of elastic elements in the network, and then track changes in network compliance and structure following LVRS and bLVR interventions. Our simulations shed light on the factors contributing to immediate and long-term treatment efficacy as well as predicted outcomes spanning several years. We show that macroscopic improvements in compliance following lung volume reduction are correlated with the microscopic distribution of mechanical forces in the local lung environment. Furthermore, our model predicts similar survival and quality of life benefits for both interventions, indicating how bLVR may be implemented as an effective, less invasive treatment for advanced emphysema.
The elastic behavior of lung tissue was modeled using a two-dimensional (2D) computational network of linearly elastic elements arranged in a hexagonal lattice under the influence of gravity. Disease progression was driven by elimination of network elements carrying a high force, representing in vivo tissue failure in regions of high local mechanical stress. Fig 1A shows a representative simulation beginning with healthy tissue followed by gradual deterioration. As tissue failure progressed (i.e., high force elements were removed), smaller airspaces were observed to coalesce into larger neighboring airspaces representing the process of airspace enlargement characteristic of emphysema progression. Fig 1B illustrates the two lung volume reduction techniques initiated in parallel from the same network configuration. LVRS was simulated by removing the upper portion of the network, while bLVR was simulated by reducing specific affected regions of the network.
To characterize functional changes within the network at each configuration, the network compliance, C was calculated as the inverse of the 2D bulk modulus. To characterize structural changes, network heterogeneity was quantified as the coefficient of variation of individual airspace sizes, CVarea. Prior to lung volume reduction, values of C and CVarea increased from baseline consistent with the loss of elastic recoil and enlarged airspaces observed in emphysema (Fig 1C). In the representative network shown, both the non-specific LVRS and the region-specific bLVR yielded comparable immediate and long-term improvements in network elasticity as characterized by similar reductions and recoveries in C, respectively. In contrast, upper network resection with LVRS was less effective in other networks as characterized by differences in reduction and recovery in C (S1 Fig). Thus, variability in initial conditions among the N = 14 large networks, representing inter-subject variability, simulated disease distributions with treatment-specific responses to lung volume reduction.
To distinguish networks demonstrating a benefit from LVRS, we defined the predictive index β characterizing network heterogeneity below the line of LVRS resection. Here, β was calculated as the coefficient of variation of airspace sizes below the line of LVRS resection. Less heterogeneity in this lower region (i.e., smaller β) corresponded to relatively spared tissue due to predominately upper lung emphysema. More heterogeneity in this region (i.e., larger β) corresponded to emphysematous tissue destruction extending into the lower lung regions. Based on this index, an arbitrary threshold of β = 3.5 was defined to divide the networks into LVRS responder (N = 8) and marginal-responder (N = 6) groups. As shown in Fig 2A, the immediate drop in C following LVRS was inversely related to β (R2 = 0.600), with larger changes observed for responders compared to marginal-responders (36.4±8.6% vs. 20.1±4.4%; p = 0.001). Representative networks for both groups are shown in the supplement (S2 Fig). In addition, β was inversely related to the overall network heterogeneity quantified by CVarea (S2 Fig). That is to say, smaller values of β implied heterogeneous network structure amenable to LVRS, whereas larger values of β implied homogeneous disease patterns less effectively treated by resection of the upper network.
To better characterize the mechanical differences between networks with different β values, we compared the distribution of forces carried by individual network elements before and after intervention. As shown in Fig 2B, LVRS skewed the distribution to the right (i.e., toward larger forces) for responders, whereas it did not significantly alter the skewness relative to before intervention for marginal-responders. Interestingly, bLVR skewed the distribution of forces to the right for all networks and reduction conditions. As shown in Fig 2C, plotting values of C after bLVR against the coefficient of variation of forces (CVforce) revealed that a simple power law function fitted the data well: C∼CVforce−0.7 (R2 = 0.816). This observation demonstrates that macroscopic functional changes immediately after bLVR are linked to the underlying microscopic force redistribution. Taken together, these findings suggest that the introduction of high-force elements plays a critical role in lung volume reduction efficacy and is dependent on pre-treatment structure for LVRS but not bLVR.
In addition to the immediate response following lung volume reduction, we also evaluated the long-term response for both LVRS and bLVR. Disease progression was modeled as consecutive stages of increasing network deterioration characterized by the cumulative number of broken elements at each stage. Fig 3 shows the average changes in C and CVarea before and after intervention.
Prior to lung volume reduction, average values of C and CVarea increased with worsening disease severity consistent with a softening network and expanding emphysematous regions. Comparing networks classified as either responders or marginal-responders, statistically significant differences between groups were detected for C and CVarea just before intervention, suggesting that responders were slightly less elastic with more heterogeneous disease progression. Following LVRS, average values of C and CVarea initially returned to near baseline levels. Responders demonstrated smaller increases in C and CVarea at advanced disease stages indicating more sustainable functional and structural improvements compared with marginal-responders. This highlights the improved long-term treatment efficacy in networks with smaller vs. larger β, given similar stages of tissue deterioration prior to intervention. Following bLVR, no differences were detected between networks classified as either responders or marginal-responders. Instead, long-term response was related to bLVR reduction size. Networks with affected regions reduced to 20% of their original size yielded smaller increases in C at each stage of disease progression compared with those reduced to 40%, while average values of CVarea were not statistically different between groups.
Based on the long-term simulations in this computational model, we compared the predicted outcomes for LVRS and bLVR as influenced by lung compliance. We found that all treatments actually accelerated network deterioration, more than doubling the rate of increase in C prior to intervention (Fig 4A). Despite the accelerated rate of increase, treatments restoring C to near baseline levels lengthened the expected survival estimated as the number of broken elements required to reach a 60% increase from baseline. LVRS and bLVR reduction to 20% yielded increases of 1.51±0.13 and 1.51±0.11 times longer than without treatment, respectively (Fig 4B). We also calculated a combined index related to the area enclosed by the threshold and the curve defined by values of C (schematic shown in Fig 5). By incorporating both the rate of increase and the number of broken elements, this index termed “Relative Benefit” represented a measure for quality of life. LVRS and bLVR reduction to 20% yielded similar increases of 11.7±3.7 and 10.6±3.2 relative to without treatment, respectively, while bLVR reduction to 40% yielded a considerably smaller increase of 4.7±1.8 (Fig 4C). These model predictions indicate that bLVR can yield similar outcomes as LVRS when affected regions are appropriately reduced in size.
Lung volume reduction represents the primary therapeutic strategy for advanced emphysema. LVRS is a well-established surgical treatment, but is limited by strict indications and significant post-procedural complications. Several non-surgical bLVR approaches are on the rise providing less invasive alternatives with the potential for considerably lower post-procedural morbidity and mortality. While previous work has evaluated improvements in clinical function and survival advantage provided by these techniques, little is known about the corresponding micromechanical mechanisms responsible for improvement in survival and quality of life. In this study, we constructed 2D elastic networks to simulate lung volume reduction with LVRS and bLVR in a force-based model of emphysema progression. Our main findings include: (1) analysis of network structure using a simple measure of disease heterogeneity prior to lung volume reduction can predict LVRS efficacy; (2) macroscopic functional improvements following bLVR correspond to microscopic changes in force heterogeneity; and (3) lung volume reduction improves aspects of the predicted survival and quality of life influenced by contributions of lung compliance, albeit while accelerating disease progression.
Mechanical forces have long been suggested to play a role in emphysema progression [23]. Previous work [24] provided early evidence demonstrating alveolar wall rupture in elastase-treated tissue slices as a direct result of local mechanical forces. Subsequently, it was shown that increases in lung compliance paralleled changes in airspace heterogeneity associated with force-induced failure of the extracellular matrix (ECM) [25]. Inflammatory processes, concerted action of proteases, and ECM remodeling also likely contribute to emphysema progression [26–29]; however, their role has been proposed more broadly within self-propagating dynamic loops of enzymatically initiated but mechanically driven tissue destruction [19]. Previous network model simulations have demonstrated that emphysematous tissue breakdown cannot be reproduced by a purely chemical process, such that the inclusion of local forces are critical in developing observed emphysema patterns [18,19]. Alternative models based on uniform softening or random cutting have also been found to poorly characterize these progressive changes [25]. Thus, we consider the network model described here to provide a suitable description of emphysema progression with the capacity for studying the structure-function relations following lung volume reduction.
It is known that patients with predominately upper-lung emphysema have more favorable outcomes following LVRS [3,9]. In this study, greater improvements in C after LVRS were observed in those with less affected lower network regions (Fig 2A) as resection of the upper diseased airspaces allowed for the remaining tissue to restore function. These networks also displayed smaller increases in C, especially at advanced disease stages (Fig 3A), as well as considerably smaller decreases in total network stress, σ, that may reflect transpulmonary pressure in the lung (S3 Fig). This is consistent with previous experimental data demonstrating that improvements after LVRS are correlated with increased ratios between upper to lower zone emphysema (determined by computed tomography, CT), but are not well predicted by pre-surgical measurements of static lung compliance or elastic recoil [30,31]. The observed improvements in C may also be related to differences in force distribution between LVRS groups (Fig 2B). One possible explanation is that emphysematous areas remaining after LVRS act as “shock-absorbers” contributing to a softer overall tissue, whereas networks with relatively low structural heterogeneity better facilitate force propagation and a stiffer overall tissue. Moreover, we found that a power law distribution could be used to characterize the distribution of forces before and after intervention (S4 Fig). The tail of the distribution varied with the specific intervention and may indicate the emergence of complex network behavior [32], which would be a unique case when increased lung heterogeneity potentially contributes positively to treatment outcome. Nonetheless, these findings suggest a mechanism that may explain how functional changes evolve based on intrinsic structural differences prior to LVRS.
Motivated by the benefits observed with LVRS, however, non-surgical bronchoscopic alternatives have been the focus of recent investigations, where patient outcomes have improved as bLVR techniques have become more proficient [8,10]. A recent study using endobronchial valves [14] reported that improvements in measured FEV1 were correlated with effective collapse of the affected lobe, a finding confirmed to be enhanced by fissure completeness and absence of interlobar collateral ventilation [6]. This is conceptually similar to our model predictions that bLVR reduction size is inversely related to immediate and long-term improvements in C (Fig 2 and Fig 3). Comparing bLVR for multiple reduction sizes revealed that macroscopic functional improvements in C were linked to underlying microscopic changes in force heterogeneity. Although radiographic evidence indicates near complete reduction is currently not achievable [8,12,15], our findings highlight important structure-function interactions between network reorganization after bLVR and its effect on the local mechanical environment in the lung (observations which would otherwise be impossible to detect via imaging or functional studies). This unexpected relationship demonstrates that bLVR can be an effective treatment for advanced emphysema, but also suggests a mechanism by which elevated forces in close proximity to reduced areas may promote local tissue destruction.
By simulating LVRS and bLVR in parallel from the same configuration, we were able to directly contrast outcomes and disease progression after treatment. In general, lung volume reduction led to more rapid tissue failure as a result of increased mechanical forces on elastic elements. This is consistent with clinical reports of accelerated deterioration of lung function (relative to pre-surgery) observed in patients following LVRS [33,34]. Despite elevated rates of tissue failure, LVRS and bLVR are predicted to lengthen survival and improve quality of life by restoring lung function to levels closer to healthy tissue (Fig 4). Statistically comparable outcomes were observed for LVRS and bLVR reduction to 20%, suggesting that bLVR is capable of similar treatment efficacy as current surgical standards. The modality-specific reduction in bLVR efficacy highlighted by Ingenito et al. [10] might also be explained by the less proficient treatment modeled by bLVR reduction to 40%. Nonetheless, these computational findings support bLVR application across an even broader treatment population, as suggested by Deslee et al. [35]. This is of particular interest given the potential for substantially less invasive bronchoscopic techniques to extend treatment options to those who do not qualify for LVRS.
Changes in relative lung volumes are also correlated with treatment efficacy. Fessler et al. [7] have shown that decreases in residual volume (RV) relative to total lung capacity (TLC) contribute to improvements in FEV1 after LVRS. Our results support these findings illustrating that LVRS in heterogeneous networks and bLVR applied to affected regions represent treatments that come closest to the removal of pure RV and allow for expansion of more normal regions. Interestingly, related studies evaluating the success of bilateral lung transplant somewhat counterintuitively observed significantly better outcomes in cases with donor lungs larger than the recipient thorax (as estimated by the donor-recipient predicted TLC ratio) [36,37]. It was proposed, however, that a decrease in lung compliance post-transplant was likely a contributing factor for survival and performance, which would agree with the benefits of lung volume reduction modeled here. The importance of these anatomic considerations also suggests the potential for coupling this computational approach with CT imaging in the future. The non-invasiveness of such an analysis would be uniquely suited to infer patient outcomes prior to treatment and aid clinical decision-making.
There are several limitations of the network model that must be considered when associating computational findings with clinical outcomes. (1) Our model considers the lung tissue to be a collection of interconnected acinar compartments; however, recent work has highlighted the involvement of the small airways in COPD development. Narrowing and loss of terminal bronchioles are believed to increase small airway resistance in COPD patients [38], and may even precede emphysema development [39]. Hiorns et al. [40] have further demonstrated the dynamic and spatially heterogeneous nature of these airway-parenchyma tethering interactions in precision cut lung slices. Although interactions at this scale are not included here, the stiffer airways would likely be associated with local parenchymal destruction. The hexagonal network units might alternatively reflect the mechanics of secondary pulmonary lobules, approximating the coalescence of destroyed lobules and enlarged lesions characteristic of emphysema, as described by Hogg et al. [41]. (2) The number of broken elements may not have a direct temporal correlation with progression in vivo even though tissue destruction is clearly associated with more developed disease severity. Moreover, emphysema progression is modeled by breaking a specified number of elements at each stage of disease. An alternative approach would be to define a global threshold, as investigated for ventilator-induced lung damage [42], above which elements are considered to fail. Both approaches yield similar disease patterns, but could influence the apparent rate of tissue failure differently. Nonetheless, observed inter- and intra-subject variability in vivo, along with only few data from follow-up studies, validate the general interpretations of our results presented here. (3) Ventilatory dependencies associated with incomplete lung fissures are not captured by our network models; however, the bLVR simulations presented here are markedly similar to the mechanical action of nitinol coils and biomaterial-based lung sealants believed to function independently of collateral ventilation pathways [12,15]. (4) Chest wall mechanics, irregular lung boundaries, nonlinear dynamics, 3D interactions, enzyme kinetics, and ECM remodeling were also not included in this study, but could improve the interpretation of factors contributing to disease progression. Future work expanding bLVR in a true multiscale model of emphysema in 3D [43] that incorporates airway-parenchymal interactions, inflammation, and enzyme kinetics may provide additional insights and enhance the potential for clinical application. Despite these limitations our network model has been shown to generate disease patterns with strong correlation to those observed using CT imaging [18] by including contributions of mechanical forces that likely drive emphysema progression [19].
While these computational simulations represent a simplified view of emphysema progression, this model provides new perspective into the structure-function relations underlying the progressive nature of emphysema before and after lung volume reduction. Immediate and long-term responses to these interventions appear to be intimately linked to changes in microscopic force heterogeneity within the lung, which could explain known structural limitations for surgical approaches and emphasize pertinent implications in disease progression for bronchoscopic approaches. Furthermore, our findings suggest that effective bronchoscopic reduction of affected lung tissue can achieve similar if not better functional improvements, survival advantages, and quality of life benefits as currently established surgical techniques. These insights have the potential to inform more rationalized design of lung volume reduction techniques and patient-specific treatment strategies.
We constructed N = 14 networks to model the elastic behavior of the lung parenchyma with different initial conditions simulating inter-subject variability. Networks were progressively degraded by eliminating elements carrying the highest forces and then finding the network configuration with minimal elastic energy for five sequential iterations. LVRS and bLVR were then applied to the same network configuration and the treated networks were subsequently degraded as before. Structural and functional parameters were tracked for each network configuration to characterize changes at each stage of disease progression. Finally, predicted survival and quality of life outcomes were compared for both treatments.
The 2D network model used in this study has been described previously [18–22]. Briefly, elastic elements inter-connected via pin joints were allowed to rotate freely while nodes bordering the perimeter of the network were kept fixed to ensure the network was initially pre-stressed and hexagonal units, representing individual acini, were not collapsed. For each configuration, the total elastic energy, Etot was calculated as the sum of the energies for individual elastic elements, Ei:
Etot=∑iEi=12kiΔli2
where ki are the linear spring constants and Δli are the individual element displacements from their resting length. Each network consisted of 6,987 elastic elements and 2,310 hexagonal cells (85x56 nodes). Uniform distributions (Mean ± SD) of spring constants (1.0 ± 0.4) and resting lengths (0.5 ± 0.1) were assigned to the elastic elements to introduce a degree of initial heterogeneity.
The minimum energy corresponding to the equilibrium configuration of the network was obtained using the equation above with a variant of the simulated annealing technique [44,45]. Here, the position of each node was displaced by a small amount proportional to and in the direction of the local resulting force on the node. If the change in total energy compared to the previous position was negative (ΔE < 0) the new configuration representing a lower energy state of the system was accepted. Alternatively, for ΔE ≥ 0, the new configuration could be accepted with probability P = exp(−ΔE/T) where T was a control parameter that was sequentially reduced until a pre-defined convergence criterion was reached.
Gravity dependence in the network was simulated by applying additional downward forces at each node with magnitude proportional to the number of dependent nodes below. This relatively weak influence represented the net effect of gravity over long time-scales proposed to enhance tissue destruction in the upper lung [23]. In the absence of this term, emphysema would be expected to progress with equal probability in any region of the network.
Emphysema was initiated in the network model by randomly breaking ~4% of all the elastic elements. Tissue failure was then simulated using a force-based destruction approach. Elastic elements were sorted by their corresponding force, and the top 0.7% of elements were broken with probability P = 0.40. Individual elements were not considered to experience fatigue behavior. The modified network was solved to yield a new configuration and distribution of forces, which corresponded to a later disease stage with different elastic elements at risk for failure. This discretized approach generated a disease progression driven by the spatial distribution of forces while the probabilistic elimination of elements introduced a degree of stochasticity to each network, limiting the deterministic nature of each simulation. These steps were repeated for a total of five iterations simulating progressively more developed disease severity.
LVRS and bLVR were applied in parallel to reduce affected emphysematous regions. To simulate lung resection in LVRS, the upper 30% of the network was removed and affected regions intersected by the threshold were stretched to form a continuous, fixed horizontal upper border. To simulate reduction of enlarged airspaces in bLVR, nodes encompassed by a selected perimeter, corresponding to the region to be reduced, were moved toward their geometric center of mass. Regions including a fixed border were asymmetrically reduced in size parallel to the axis of the border. For each network, affected regions were selected and then reduced to 1, 20, or 40% of their original size.
Mechanical stress σ was calculated for each network configuration by numerically differentiating the total energy, Etot of the system at the equilibrium configuration and after stretching the network by a small bi-axial strain, ε = ±0.01. Here, the equilibrium configuration was assumed to correspond to FRC, representing a static measurement of lung function. Networks were then stretched with a sinusoid of amplitude ε = ±0.04 around the equilibrium configuration, such that the 2D bulk modulus was defined as the slope of the corresponding stress-strain curve. The compliance C was calculated as the inverse of the estimated network bulk modulus at each stage of disease progression. To facilitate comparisons with baseline, σ and C are reported as percent changes from the initial network configuration prior to emphysema destruction.
Network structure was quantified by considering the sizes of individual airspace units. Each network configuration was converted to a binary image and the number of pixels enclosed by connected spring elements represented the individual airspace area. Overall structural heterogeneity was then assessed as the coefficient of variation of airspace sizes, CVarea. For network configurations directly before intervention, we also considered the coefficient of variation for airspaces below the line of LVRS resection. This predictive index, referred to as β, subsequently characterized disease heterogeneity in the network not resected by LVRS. Note that β was calculated as a single predictive index before treatment, whereas CVarea was calculated for each stage of disease progression to track changes in overall network structure.
The rate of tissue failure was estimated before and after intervention as the increase in C over four stages of disease progression. The number of broken springs required to reach a 60% increase in C was calculated for each network as an estimate of survival. However, since network deterioration prior to treatment was typically less than this threshold a second order polynomial was fitted to values of C to estimate survival in the absence of any lung volume reduction. The relative benefit of treatment was then calculated as shown in the schematic (Fig 5). The area between the survival threshold and the compliance curve represents a composite index for quality of life, incorporating both the rate and sub-threshold duration of disease progression. Larger values of this area correspond to lower values of C over a longer period of time and hence represent better quality of life. To compare the benefits provided by lung volume reduction, data are reported as normalized by the estimated values in the absence of any treatment.
Network simulations were completed using custom-developed software, which has been utilized previously to generate and analyze networks in conjunction with other experimental studies [18–22]. Network manipulations involving LVRS and bLVR were implemented cooperatively with this program using original scripts developed in MATLAB (MATLAB r2013a, MathWorks, Natick, MA).
Two-way repeated measure analysis of variance (ANOVA) was used to compare network values of C, CVarea, and σ between treatment groups at each stage of disease progression, as well as the skewness of force distributions for each treatment group. One-way ANOVA was used to compare estimates of disease progression rate, survival, and relative benefit. Post-hoc Holm-Sidak and Tukey tests were used to determine differences between groups. The average change in C after LVRS for responder and marginal-responders were compared using a t-test. For all comparisons, p<0.05 was considered significant. Statistical analyses were performed using SigmaPlot (SigmaPlot v12.3, Systat Software, Inc., San Jose, CA) and MATLAB.
|
10.1371/journal.pbio.1000429 | Cryptic Variation between Species and the Basis of Hybrid Performance | Crosses between closely related species give two contrasting results. One result is that species hybrids may be inferior to their parents, for example, being less fertile [1]. The other is that F1 hybrids may display superior performance (heterosis), for example with increased vigour [2]. Although various hypotheses have been proposed to account for these two aspects of hybridisation, their biological basis is still poorly understood [3]. To gain further insights into this issue, we analysed the role that variation in gene expression may play. We took a conserved trait, flower asymmetry in Antirrhinum, and determined the extent to which the underlying regulatory genes varied in expression among closely related species. We show that expression of both genes analysed, CYC and RAD, varies significantly between species because of cis-acting differences. By making a quantitative genotype-phenotype map, using a range of mutant alleles, we demonstrate that the species lie on a plateau in gene expression-morphology space, so that the variation has no detectable phenotypic effect. However, phenotypic differences can be revealed by shifting genotypes off the plateau through genetic crosses. Our results can be readily explained if genomes are free to evolve within an effectively neutral zone in gene expression space. The consequences of this drift will be negligible for individual loci, but when multiple loci across the genome are considered, we show that the variation may have significant effects on phenotype and fitness, causing a significant drift load. By considering these consequences for various gene-expression–fitness landscapes, we conclude that F1 hybrids might be expected to show increased performance with regard to conserved traits, such as basic physiology, but reduced performance with regard to others. Thus, our study provides a new way of explaining how various aspects of hybrid performance may arise through natural variation in gene activity.
| A major conundrum in biology is why hybrids between species display two opposing features. On the one hand, hybrids are often more vigorous or productive than their parents, a phenomenon called hybrid vigor or hybrid superiority. On the other hand they often show reduced vigour and fertility, known as hybrid inferiority. Various theories have been proposed to account for these two aspects of hybrid performance, yet we still lack a coherent account of how these conflicting characteristics arise. To address this issue, we looked at the role that variation in gene expression between parental species may play. By measuring this variation and its effect on phenotype, we show that expression for specific genes may be free to vary during evolution within particular bounds. Although such variation may have little phenotypic effect when each locus is considered individually, the collective effect of variation across multiple genes may become highly significant. Using arguments from theoretical population genetics we show how these effects might lead to both hybrid superiority and inferiority, providing fresh insights into the age-old problem of hybrid performance.
| Crosses between closely related species give two contrasting results [3]. One result is that species hybrids may be inferior to their parents, with reduced fertility or viability [1]. The other is that F1 hybrids may be superior (heterosis), with increased vigour [2],[4]. Hybrid inferiority is commonly explained through incompatible interactions between loci [5]–[8]. Hybrid superiority is either explained through accumulation of different recessive deleterious mutations in each species, or by loci exhibiting heterozygote advantage (overdominance) [9]. The deleterious recessives hypothesis has received support from studies on domesticated inbred varieties [10], although it is unclear how such deleterious mutations would become fixed in natural populations with larger effective population sizes (though see [11],[12]). The hypothesis of heterozygote advantage suffers from the problem that it is unclear why overdominance should be prevalent [13].
One approach to understanding the basis of hybrid performance is to analyse interspecific variation at a few interacting loci and then extrapolate these findings. Variation between closely related species mainly involves either loci with small quantitative effects, or loci conferring no detectable phenotypic effect, known as cryptic variation [14]–[20]. Gene expression studies have revealed extensive differences between species in both cis- and tran-regulation across the genome [21]–[25]. However, the relationship between such variation in gene expression and phenotype has not been extensively explored. To address this issue, we have analysed interspecific variation in expression of two interacting developmental loci in Antirrhinum.
The Antirrhinum species group of southern Europe comprises about 20 species with diverse morphologies [26],[27]. These species can be intercrossed in the laboratory to give fertile hybrids, allowing the genetic basis of the species variation to be studied. So far, these studies have largely addressed divergent traits such as flower shape and colour or leaf shape and size [28]–[32]. However, to determine whether cryptic variation may also be prevalent, we chose here to analyse a conserved trait – flower asymmetry. All the species in the group have asymmetric flowers with matching upper and lower petals. The asymmetry depends on four key dorsoventral genes, CYCLOIDEA (CYC), DICHOTOMA (DICH), RADIALIS (RAD), and DIVARICATA (DIV) [33]–[36]. CYC and DICH encode related proteins belonging to TCP family of transcription factors, whereas RAD and DIV encode members of the Myb transcription factor family. The interaction between CYC and RAD has been studied in detail [37],[38]. RAD is a likely downstream target of CYC and acts in parallel with CYC to control dorsal and lateral petal development. As their interaction had been well characterised, we chose this pair of genes for our studies on cryptic variation.
We show here that species of the Antirrhinum group exhibit variation in the levels of CYC and RAD expression. This variation is cryptic as it lies within a plateau in gene expression–morphology (GEM) space. However, phenotypic effects can be revealed by creating genotypes in which the species alleles are shifted off the plateau. By considering the consequences of such patterns of variation for multiple loci and in relation to possible gene expression–fitness (GEF) spaces, we conclude that F1 hybrids might be expected to show increased performance with regard to basic physiological traits such as growth. This finding provides an explanation for hybrid vigour that avoids some of the pitfalls of previous hypotheses. Hybrid inferiority may also be expected in the longer term for nonphysiological traits such as those involved in sexual reproduction.
To determine the extent of interspecific variation in CYC and RAD expression, a range of species was crossed with A. majus. Expression of the species allele relative to A. majus in the F1 hybrids was then determined by competitive reverse transcription (RT)-PCR. For this procedure, RNA was extracted from flower buds collected at the same developmental stage from individual plants (three individuals were used as replicates). The expression levels of the two alleles were distinguished by pyrosequencing [21],[39]. CYC and RAD were sequenced from each species and a region chosen that included differences between A. majus and the other Antirrhinum species. As a control, genomic DNA from the hybrids was also assayed and deviations from a 1∶1 ratio used to calculate PCR bias.
Expression ratios were represented relative to alleles from the reference species A. majus (i.e., CYCmaj = 1 and RADmaj = 1). For any comparison, the A. majus and other species allele will be in the same hybrid background and should thus only detect cis-acting differences (trans-acting variation should affect both alleles in the hybrid equally). Some species had alleles with significantly lower expression than A. majus (i.e., CYCtor, RADcha, RADpul), while others had higher expression levels (i.e., CYCpul, CYCcha, RADlin, RADlat; Figure 1). A. pulverulentum exhibited two different expression levels. The two expression categories correlated with different genomic DNA PCR biases and different DNA sequences, suggesting that they reflected a polymorphism within A. pulverulentum. Taken together, the results show that there is significant cis-acting variation in expression levels for CYC and RAD among the Antirrhinum species.
To determine the relationship between variation in CYC and RAD expression and developmental phenotype, a mapping between gene expression and morphology for A. majus was established. In what follows, we make the simplifying assumption that expression for each gene can be represented along a single axis, ignoring factors such as spatial or temporal variation in expression pattern. We also represent morphology along a single axis as this allows the GEM space to be more readily visualised. The advantage of taking such a simplified view is that it allows the key interactions and principles to be identified.
Plants with various combinations of CYCmaj and RADmaj activity were generated by crossing A. majus to lines carrying cyc and/or rad mutant alleles. These mutant alleles carry transposon insertions that reduce gene expression levels [33],[36]. Genotypes were confirmed using allele-specific cleaved amplified polymorphic sequences (CAPS). The resulting nine genotypes exhibited a range of phenotypes, consistent with previous studies (Figure 2) [40]. Three genotypes looked wild type (Figure 2B, 2C, 2F), as expected from the recessive nature of the mutants. The double mutant (Figure 2G) had fully ventralised flowers. Other allele combinations had intermediate phenotypes ranging from very strongly ventralised (Figure 2D, 2H), semi-ventralised (Figure 2A, 2I), to near wild-type flowers with a gap or notch between the lower and upper petals (Figure 2E).
To establish a more quantitative mapping between expression and morphology, gene expression and morphometric measurements were made for each genotype. Expression levels for CYC and RAD were determined by quantitative RT-PCR, using UBIQUITIN (AmUBI) as reference gene (Table 1). Compared to wild type, the fully ventralised cyc rad double mutant had CYC and RAD expression levels of less than 1% (Table 1, row G). In single rad homozygotes (Table 1, row I) expression of RAD was less than 0.14% of wild type, whereas CYC expression remained unaffected. In single cyc homozygotes (Table 1, row A), expression of CYC was down to 1%, whereas RAD was reduced to 20%. The reduced expression of RAD in these plants was consistent with RAD being a downstream transcriptional target of CYC. The residual RAD expression of 20% was presumably driven mainly by DICH, which acts redundantly with CYC [34]. Single heterozygotes for CYC (Table 1, row B) or RAD (Table 1, row F) showed about 50% expression of the relevant gene, indicating that there was little dosage compensation. The other genotypes gave further combinations of expression levels. Taken together the genotypes defined nine positions in CYC-RAD expression space.
To allow a GEM space to be visualised, a single morphometric measure was needed for each genotype. To obtain this measure, the corolla was first dissected and flattened. 112 points were then placed around the petal outlines to capture their overall shape and size (Figure 3A). Some of these points (primary landmarks) were placed at recognisable features such as petal junctions, whereas others (secondary landmarks) were regularly spaced between the primary landmarks. This procedure was followed for eight wild-type and eight fully ventralised (cyc rad/cyc rad) flowers (alleles that are linked in coupling, i.e. are on the same chromosome, are shown underlined). The resulting 16 sets of coordinates were aligned (Procrustes alignment) and subjected to principal component analysis (PCA). A statistical model was obtained yielding one PC that captured most of the variation (90%) between the wild-type and the ventralised flower phenotypes (Figure 3B). For convenience, the PC values were scaled such that the mean ventralised mutant had a value of 0 and the mean wild-type a value of 1. The PC could therefore be considered as a dorsalisation index (DIcor) that provided a quantitative measure of variation in corolla morphology. Projection of the wild-type and fully ventralised petals onto DIcor yielded two distinctive groups, separated according to phenotype.
The DIcor for each of the nine genotypes was determined by flattening their petals, placing landmarks, and projecting their coordinates onto DIcor, which revealed that all genotypes had a DIcor between 0 and 1 (Table 1). The single heterozygotes (Table 1, rows B and F) had a DIcor of slightly less than 1. This difference from wild type was reproducible and observed in families segregating for the alleles. This finding indicates that the mutants are not fully recessive when assayed by this quantitative measure. The remaining genotypes gave lower DIcor values, reflecting their degree of ventralisation.
A GEM space was constructed by plotting DIcor for each genotype against its gene expression levels for CYC and RAD (Figure 4A). To get a better impression of the shape of the space, a continuous function was used to capture the main trends of the observed values. The resulting smooth GEM space gave a DIcor that climbed from where values of CYC and RAD expression were low to a plateau where gene expression was high. Plotting the gene expression levels for the species relative to wild-type A. majus (coordinates CYC = 1, RAD = 1) within the same space showed that they were all located on the plateau of high DIcor (Figure 4B). This finding is consistent with all species having asymmetric and fully closed flowers. Thus, even though there is variation in expression between species, the variation is cryptic at the morphological level because of the plateau in GEM space.
One way of revealing the cryptic variation would be to shift the species off the plateau onto a steeper part of the GEM space by creating double heterozygotes. In A. majus double heterozygotes, gene expression levels are shifted to position (CYC≈0.6, RAD≈0.5), which corresponds to a DIcor of 0.76 and lies just below the plateau in GEM space. If a similar shift is applied to the species, several distinct DIcor values would be expected as the species are unlikely to fall on exactly the same DIcor contour as A. majus (Figure 4B and 4C). Because DIcor can only be strictly determined within the A. majus background, testing this prediction would require alleles from the species to be introgressed into the A. majus background followed by creation of the double heterozygotes. As an introgression programme was already underway for A. charidemi, this species was chosen for further analysis.
Using CAPS and amplified fragment length polymorphism (AFLP) markers CYC and RAD alleles from A. charidemi were introgressed into the A. majus background. At the backcross 5 (BC5) generation, plants with genotype CYCmajRADmaj/CYCchaRADcha were crossed to the double mutant cyc rad/cyc rad generating two main genotypes. CYCmajRADmaj/cyc rad showed the expected notched morphology corresponding to a DIcor of 0.78 (Figure 5A, 16 plants). By contrast, CYCchaRADcha/cyc rad had a morphology more similar to wild type (Figure 5B) and had a significantly higher DIcor of 0.86 (Figure 5C, 11 plants). This finding indicates that the previously observed expression difference between A. charidemi and A. majus alleles had a phenotypic effect in a double heterozygous background.
To confirm that this effect was significant, a larger population was analysed (131 plants). Rather than using the entire corolla for calculating the DI, only the lateral lobe was used as this could be processed more readily. A new DI index, DIlat, was constructed by placing 25 points around the lateral lobe of wild-type and ventralised mutant flowers, capturing 96% of the variation (Figure 3C). This DI index was shown to be strongly correlated with the DIcor index for flowers in which both were determined (Pearson product moment correlation R = 0.91, p<0.001). CYCcha RADcha/cyc rad (69 plants) had a DIlat value of 0.68, which was significantly higher than the DIlat value of 0.59 for CYCmaj RADmaj/cyc rad (63 plants) (Figure 5D). These results confirm that alleles from A. charidemi confer greater dorsalisation than those from A. majus in a doubly heterozygous background.
To determine the individual contribution of CYC and RAD to the observed difference in DIlat, recombinant CYCcha RADmaj and CYCmaj RADcha chromosomes were obtained by screening BC5 progeny with CAPS markers (CYC and RAD are 3cM apart [41]). The recombinants were crossed to the double mutant cyc rad. CYCcha RADmaj/cyc rad had a DIlat of 0.74, greater than CYCmaj RADmaj/cyc rad, whereas CYCmaj RADmaj/cyc rad had a DIlat of 0.68 similar to CYCmaj RADmaj/cyc rad (Figure 5E). This indicates that the shift in DIlat between A. majus and A. charidemi mainly reflects a change in CYC activity, consistent with the observed higher levels of CYCcha expression in A. majus/A. charidemi F1 hybrids (Figure 1).
To confirm that the differences in gene expression were maintained in the A. majus background, allele expression was compared in the introgression lines. Consistent with the expression analysis on the F1 hybrid, expression of CYCcha was about 30% higher than that of CYCmaj in BC6 CYCmajRADmaj/CYCchaRADcha flower buds (Figure 6A, stage 11). This finding confirmed that the variation in CYC expression observed in the F1 hybrid was due to cis-regulatory differences. There was no significant difference between expression of RADcha and RADmaj (Figure 6C, stage 11), irrespective of whether plants carried CYCcha or CYCmaj (unpublished data). Thus, the position of A. charidemi in gene expression space was very similar for the F1 and BC6 plants (Figure 4B, 4C).
Expression analysis was also carried out at various developmental stages for F1 and BC6 plants to determine whether the relative expression levels were maintained. At all stages tested, expression of CYCcha was about 30% higher than that of CYCmaj in F1 or BC6 plants (Figure 6A, 6B). Expression of RADcha was also found to be higher than that of RADmaj but this difference was only observed at earlier developmental stages (Figure 6C, 6D). The early enhancement of RADcha was observed irrespective of whether plants carried CYCcha or CYCmaj (unpublished data). This difference in RAD expression appears to make little contribution to the phenotype because as previously shown, CYCmaj RADcha/cyc rad had a similar DIcor to CYCmaj RADmaj/cyc rad (Figure 5F).
Given the similarity between the results obtained for F1 and introgressed A. majus backgrounds, double heterozygotes with further Antirrhinum species were generated by crossing them to cyc rad/cyc rad plants of A. majus. A strong notch was observed in flowers of hybrids with A. tortuosum (Figure S1), as might be expected from its relatively low levels of CYC and RAD expression (Figure 1) and predicted position in GEM space (Figure 4C). A mild notch was observed with A. latifolium and no notch with A. braun-blanquetii, again consistent with their position in GEM space. However, a notch was observed in A. pulverulentum even though its CYCpul expression was higher than CYCmaj. This difference from expectation may reflect alterations in timing of expression or contribution of other factors in the genetic background of these F1s.
The observed pattern of interspecific variation most likely reflects the interaction between gene expression and fitness. This relationship can be represented by a GEF space, which is similar to GEM space except that fitness instead of morphology is plotted on the vertical axis. As with GEM space, we make the simplifying assumption that gene expression for each gene can be represented as a single axis, ignoring variation in expression pattern (which would correspond to further axes). GEF space is related to GEM space because fitness depends on how particular morphologies influences survival and reproduction. However, GEF and GEM spaces are unlikely to have precisely the same form because a small change in morphology may have a dramatic effect on fitness, and because morphologies assessed in the laboratory may not be precisely the same as those found in nature. To account for the observed pattern of variation in CYC and RAD expression, we consider various possibilities for GEF space.
If we assume that GEF space has a peak around the centre of our observed expression values (Figure 7A), then there will be a zone around the peak in which variation will be effectively neutral. The extent of the neutral zone will depend on the shape of the peak and the effective population size Ne. If Ne = 500, then the neutral zone will occupy a region with fitness values ranging from 1 at the centre of the peak to about 0.999 (i.e., 1−1/2Ne = 0.001) [42] at its rim. For a radially symmetrical peak, this zone would form a circular domain in gene expression space (Figure 7B). Gene expression values would be expected to drift within this domain, accounting for the clustered distribution of observed expression levels.
This random drift away from the optimum would reduce mean fitness, generating a “drift load.” For a wide range of models, this load is ∼1/4Ne for each degree of freedom (df), and is independent of the selection strength (see Materials and Methods). This load would have little impact on a moderately large population for one or two loci. However, if this scenario applies to many loci, say n≈1,000, then the fitness cost could be substantial. Under this scenario, F1 hybrids would gain a major fitness benefit, because each species would represent a different sample of gene expression space around the fitness peak. An F1 would represent the mean of two samples (assuming that gene expression shows additive inheritance) and is therefore likely to be nearer the peak than each individual sample (see illustrated genotypes in Figure 7B). More precisely, the variance around the optimum of the mean of two independent populations is half that of either one, and so the “drift load” is half as great (i.e., 1/8Ne per degree of freedom). For example, with 1,000 loci and an effective population size of 1,000, the fitness benefit would be 0.125, which is very substantial (i.e., 1,000×1/(8×1,000)).
This major fitness benefit would break down in the F2, as genes segregate to give a fitness for each offspring that is the same, on average, as that of the parents. If variation in expression for each locus is determined by k trans-acting factors with similar effects, then the F2 variance is δ2/(8k), where δ is the difference between parental means [43]. However, there may be substantial cryptic variation, with divergence due to alleles acting in opposite directions, which can produce a high F2 variance, and strong dysgenesis (see [44]). If variation is determined by cis-acting differences, as described here, then the F2 breakdown follows Mendelian segregation. Under this scenario with radially symmetrical peaks, or peaks that are elongated parallel to one or other axis of GEF space (i.e., ridges with elliptical neutral zones oriented parallel to the gene axes; Figure 7B), the F2 is spread over a region bounded by the parental values, and so the fitness is intermediate between the F1 and the parentals.
Another scenario is that if the peak in GEF space is elongated in a direction that is tilted with respect to the gene expression axes (Figure 7C), then, the effectively neutral zone of such ridges would form a tilted ellipse (Figure 7D). This scenario seems the most likely for the CYC and RAD genes, as it matches the orientation of the plateau edge in GEM space and also matches the distribution of gene expression coordinates for most of the species. When applied to multiple pairs of loci, the consequence of tilted neutral zones for F1 fitness would be similar to that for untilted zones described above – F1s would be expected to be nearer to the centre of the peak and have half the drift load as the parents. However, in contrast to the untilted zones, the fitness of many F2 progeny would be expected to be lower than for the parents. This lower fitness is because segregation in the F2 would lead to many gene combinations ending up on the more steeply declining slopes and thus lying below the parents in fitness (see example in Figure 7D).
Finally, we consider a scenario in which the region of high fitness is curved in GEF space (Figure 7E). In this case, the effectively neutral zone forms a banana shape. F1 hybrids between genotypes lying at different ends of the banana would have lower fitness than the parents because they would fall in the groove of the fitness surface (Figure 7F). Such loci would therefore lead to dysgenic effects in both the F1 and F2. In its extreme form, when the neutral zone is bent to form an L-shape, the fitness distribution corresponds to Dobzhansky-Muller incompatibility.
Our use of a GEM landscape is similar to Rice's framework, which maps phenotype onto a set of developmental characters [45], in our case, gene expression. Rice [45] shows how stabilising selection on the phenotype can lead to canalisation, such that the phenotype tends to be buffered against fluctuations in the underlying traits. This finding has much in common with the evolution of dominance, where buffering can evolve in a similar way. The process is driven by selection to reduce the variance of the trait, which we do not consider here. Also, “characters” may themselves be polygenic traits, whereas we focus on cis-acting variation at single genes. Despite these differences, there are intriguing parallels that would reward further study.
Species from the Antirrhinum group exhibit significant cis-acting variation in levels of CYC and RAD expression. This variation is cryptic, having no obvious phenotypic consequences in a wild-type genetic background. The lack of phenotype arises because the species lies on a plateau in GEM space. However, phenotypic effects can be revealed if the species are shifted off the plateau by construction of double heterozygotes. In such backgrounds, species carrying alleles with relative high levels of gene expression exhibit near wild-type phenotypes, whereas species with lower expression tend to give notched flowers with reduced dorsalisation. Recombination analysis allows the contribution of each locus to be determined. In the case of A. charidemi, the main source of phenotypic variation is due to a difference in expression at the CYC locus.
Our findings bridge those from several other studies. Comparative analysis of Caenorhabditis species has revealed changes in genes controlling the conserved trait of vulval development [18]. These changes are more substantial than those we describe, most likely because of the different timescale of these studies: divergence times are about 14–18 million y for Caenorhabditis species [46], compared to 1–5 million y for Antirrhinum [47]. Cryptic variation underlying vulval development has also been described within Caenorhabditis species [20], although the genetic basis of this variation has yet to be determined. Extensive variation in gene expression has been observed between Drosophila sibling species [21],[22] and its phenotypic consequences studied for divergent traits, such as denticle pattern [48]–[50]. Divergent ecological traits have also been shown to reflect variation in gene activity within Arabidopsis thaliana [51]–[54]. Our results suggest that variation in gene expression may also be found to underlie highly conserved traits and that this variation can have phenotypic consequences in certain genetic backgrounds, depending on the structure of the GEM spaces involved.
Although cryptic variation of the kind we describe would be expected to have little effect on species hybrid performance for each locus, we show that the cumulative effect of such variation at many loci could have a major effect. The magnitude and direction of this effect depends on the population size and topography of the GEF spaces in the various species habitats. For example, if GEF spaces have a radial fitness peak that is preserved across habitats, then for 1,000 loci and an effective population size Ne of 1,000 the fitness benefit in species hybrids would be 0.125, which is very substantial. This benefit arises because each species is expected to drift around its fitness peak within a radial neutral zone. For multiple loci this causes each species to lie significantly below the optimum (i.e., there is a drift load). As the species diverge, they come to represent separate samples of the neutral zone, each carrying a different combination of alleles contributing to drift load. A species hybrid will then represent an average of two samples and is therefore expected to lie nearer the peak than either sample alone, creating hybrid superiority. This hybrid benefit will be lost in the F2 as the alleles segregate, creating genotypes with fitness similar to those of the parents.
GEF spaces with elongated peaks and elliptical neutral zones are also expected to show similar benefits in hybrid fitness. However, in this case the F2 genotypes are expected to have lower fitness than the parents when the elliptical zones are tilted in gene expression space. Hybrid inferiority arises for GEF spaces that have curved or L-shaped neutral zones (L-shaped zones correspond to the standard Dobzhansky-Muller incompatibility).
The phenotype and fitness of species hybrids will reflect the extent to which these various GEF scenarios apply to the many thousands of genes in the genome. Radial or elliptical neutral domains, centred around a common position in GEF space, would be expected for loci that are under similar normalising selection in multiple environments. This situation likely applies to the CYC and RAD genes as all species in the Antirrhinum group have similar asymmetric closed flowers. It would also be expected for many loci controlling basic physiology and growth. F1 hybrids would therefore be expected to show higher fitness and increased performance with respect to these traits. This provides an explanation for hybrid vigour that avoids the pitfalls of previous models that require fixation of loci with major deleterious effects or that invoke special mechanisms for heterozygote advantage. A similar explanation has been proposed to account for the origin of hybrid vigour between domesticated inbred lines [10]. Hybrid vigour is usually lost in F2s or recombinant inbred lines, indicating that many of the loci involved interact to give tilted rather than untilted neutral zones.
Although hybrid vigour is commonly observed for physiological traits, the overall fitness of species hybrids is often lower than that of the parents, with sterility or other dysgenic effects being observed. This observation may partly reflect adaptation to different environments and thus shifts in the shape of fitness surfaces that drive changes in genotype. However, it may also reflect loci that interact to give curved or L-shaped neutral zones [8]. Such zones will be prevalent for traits that involve more complicated epistatic interactions, perhaps accounting for the dysgenic effects observed in F1s. The negative contribution of loci with curved neutral zones is likely to increase with time, as loci drift towards the extremities of the banana-shaped neutral domains.
The overall fitness of an F1 hybrid will depend on the relative contribution of superior and inferior effects across the genome. In this paper we have concentrated on variation in gene expression levels. Other forms of variation, such as in gene expression patterns, protein activities, or chromosome arrangements are also likely to play an important role in species divergence. The corresponding fitness spaces may be more difficult to visualise because variation for each gene may no longer be represented along a single axis. Nevertheless, the principles may be similar to those described above, with both hybrid inferiority and superiority reflecting effectively neutral variation at multiple loci, but differing with respect to the topography of the fitness spaces involved.
Total RNA was extracted using RNeasy Plant Mini-kit (Qiagen). Total RNA was treated with DNaseI Amplification Grade (Invitrogen), and cDNA synthesized using Superscript III (Invitrogen), priming with oligo dT. Genomic DNA contamination was verified using the oligos tatgtaatttcactttaatttcgtctg and tgcttcgtttattatctgaacgatt spanning from the intron towards the 3′UTR in RAD (annealing temperature 55°C). Absence of a 1,010-bp PCR product after 30 cycles was considered as evidence of genomic DNA-free cDNA samples.
For expression analysis with competitive RT-PCR and quantitative sequencing, standard procedures were followed [21],[39]. For CYC, an assay was designed to detect a polymorphism G/A conserved in CYC sequences. CYC competitive PCR was done with the oligos [5′Btn]gcagcagccaaagagtcgag and cctgctgatgaaacccgaaaa, giving a PCR product of 172 bp, and the sequencing oligo aacaaacgcctcacg. RAD sequences (∼1,452 bp) from species were obtained using the oligos tccaacaagaccttttgattcc and tgcttcgtttattatctgaacgatt, spanning from the 5′UTR towards the 3′UTR, including the two exons and the intron. Sequences were aligned to the RADmaj (GenBank AY954971 http://www.ncbi.nlm.nih.gov/nuccore/61652984), and a conserved G/A polymorphism was identified. An assay was designed using the oligos aagtccgccaaggagaacaaa and [5′Btn]acggccctagccacgtta giving a PCR product of 89 bp, and the sequencing oligo ccaaggagaacaaagc. For both assays, competitive PCRs were done with an annealing temperature of 55°C to saturation (55 cycles). Genomic DNA from every plant was also included as control for allele-specific PCR biases. All PCR reactions were done in quadruplicate. For pyrosequencing sample preparation was done using the PSQ-kit (Biotage), and quantitative sequencing in a PSQ-96 sequencer (Biotage).
For quantitative RT-PCR, AmUbiquitin (AmUBI GenBank X67957 http://www.ncbi.nlm.nih.gov/nuccore/16070) was used as reference gene. Oligos were designed to flank both sides of the transposon insertion for Tam1 in cyc-608, and Ram1 in rad-609. The oligos pairs were gttcttgagtccaccgctttgttc and aatgccgatggataaacggactct for CYC, caccggtggtaacatgaaaactgac and tgcttgctatgtgattgaacaaaacc for RAD, ggccgactacaatatccagaaggag and gaaccgaaccatcagacaaacaaac for AmUBI. PCR programmes were as follow: initial denaturation at 95°C 2 min; 40 cycles of 95°C 15 s; 55°C (CYC), 58°C (RAD), 60°C (AmUBI) 30 s; 72°C 30 s; and 72°C 10 min, after which melting curve were recorded from 70°C to 95°C, every 0.5°C. PCR reactions were performed in quadruplicate in an Opticon real-time PCR instrument (MJ Research), using SYBR Green JumpStart (Sigma). Ct values were obtained with a threshold of 0.105 using the software Opticon Monitor 3.1 (MJ Geneworks).
Lines with combinations of CYC and RAD wild-type and mutant activities were obtained from by crossing single mutants cyc-608 (JI:608) and rad-609 (JI:609) and the double mutant cyc-608 rad-609 (JI:727) to the wild type. Genotyping for the CYCmaj wild-type allele was done using the oligos tcctcccttcactctcgcgc and tggcgcatagctggttcgac, spanning most of the coding region (annealing temperature 55°C). Presence of CYCmaj wild-type allele gave a PCR product of 790 bp. Genotyping for the cyc-608 mutant allele was done using the oligos tcctcccttcactctcgcgc and gtgacccatgcactcttgg spanning from the coding region to the Tam4 transposon insertion (annealing temperature of 57°C). Presence of cyc-608 mutant allele gave a PCR product of 327 bp. Genotyping for the RADmaj wild-type allele was done using the oligos tccaacaagaccttttgattcc and tgcttcgtttattatctgaacgatt, spanning from the 5′UTR to the 3′UTR, including the two exons and the intron (annealing temperature 55°C). Presence of the RADmaj wild-type allele gave a PCR product of 1,452 bp. Genotyping of the rad-609 mutant allele was done using the oligos tccaacaagaccttttgattcc and taaggaagcttcgggtccgg spanning from the 5′UTR towards the first exon, part of the intron, and the Ram1-like insertion (annealing temperature 60°C). Presence of the rad-609 mutant allele gave a PCR product of ∼1 kb.
Based on CYCcha and RADcha sequences obtained from A. charidemi, CAPS markers were designed. For CYC, A PCR product of 791 bp on the coding region was obtained using the oligos tcctcccttcactctcgcgc and tggcgcatagctggttcgac. The PCR product was digested with the restriction enzyme KpnI (Invitrogen) generating two fragments in the CYCmaj allele (675 bp and 116 bp), but did not digest CYCcha allele. For RAD, a PCR product of 796 bp, covering the second exon and further on down-stream the RAD stop codon, was obtained using the oligos tgcatgcaggttcagaaatc and tttgggctatttcgcttgac. The PCR product was digested with the restriction enzyme AluI (Roche) producing two fragments on RADmaj allele (444 bp and 352 bp), and three fragments on RADcha allele (444 bp, 200 bp, and 152 bp).
A collection of BC3 and SBC3 plants carrying CYCcha RADcha were screened using the CAPS markers. A line with genotype CYCmajRADmaj/CYCchaRADcha was selected and further backcrossed. The BC4 population of plants carrying CYCcha and RADcha was screened with AFLP markers to select the one with the most homogeneous A. majus genome. This plant was backcrossed to generate a BC5, the progeny of which was in turn used for morphological and gene expression analysis.
The GEM space was smoothed by fitting a 2-D function to the data. The fitting was achieved using the least-square method [55]. The function was performed using the MATLAB function fminsearch, which finds a local maximum depending on given initial parameter values. The function fitted for the GEM space was:where a, b, c, kc, kr, and A0 are the parameters of the function (Table 2), CYC and RAD are the gene expression levels. The parameter A has been fixed to ensure that the wild-type genotype (i.e., CYC = 1 and RAD = 1) has a DIcor = 1.
For a variety of models of selection, the expected loss of fitness due to drift around the optimum is ∼1/4Ne for each degree of freedom, independent of the selection strength: strong selection leads to smaller fluctuations, so that the net effect on fitness is the same as with weaker selection. This result arises from Wright's formula [56] for the stationary distribution under mutation, selection, and drift, which shows that the trait distribution is multiplied by the distribution in the absence of selection. The argument applies to stabilising selection on multiple polygenic traits, or to small fluctuations in allele frequency at balanced polymorphisms; if frequency-dependent selection maintains polymorphism, then the drift load is typically twice as large [57]. Matters are more complicated when variation is maintained by a balance between mutation and selection. If we focus on a single locus, the outcome depends on how mutation acts. With a continuum of allelic effects, drift has little overall effect on the mean fitness: though it reduces fitness by causing fluctuations in the mean around the optimum, which is counterbalanced by a reduction in variance. Nevertheless, there is still heterosis of ∼1/8Ne per locus, because an F1 individual is on average closer to the optimum. If variation involves rare deleterious alleles, then the drift load, and hence the heterosis, are smaller, in proportion to the frequency of deleterious alleles. In this case, heterosis is contributed only by those loci with 4Nes≈1, for which selection and drift have comparable strength. It is this latter case, in which weakly deleterious alleles can be fixed by drift, that has previously been discussed [58]–[59]. Detailed derivations are given in Text S1 and Figure S2.
|
10.1371/journal.pcbi.1005313 | Identifying T Cell Receptors from High-Throughput Sequencing: Dealing with Promiscuity in TCRα and TCRβ Pairing | Characterisation of the T cell receptors (TCR) involved in immune responses is important for the design of vaccines and immunotherapies for cancer and autoimmune disease. The specificity of the interaction between the TCR heterodimer and its peptide-MHC ligand derives largely from the juxtaposed hypervariable CDR3 regions on the TCRα and TCRβ chains, and obtaining the paired sequences of these regions is a standard for functionally defining the TCR. A brute force approach to identifying the TCRs in a population of T cells is to use high-throughput single-cell sequencing, but currently this process remains costly and risks missing small clones. Alternatively, CDR3α and CDR3β sequences can be associated using their frequency of co-occurrence in independent samples, but this approach can be confounded by the sharing of CDR3α and CDR3β across clones, commonly observed within epitope-specific T cell populations. The accurate, exhaustive, and economical recovery of TCR sequences from such populations therefore remains a challenging problem. Here we describe an algorithm for performing frequency-based pairing (alphabetr) that accommodates CDR3α- and CDR3β-sharing, cells expressing two TCRα chains, and multiple forms of sequencing error. The algorithm also yields accurate estimates of clonal frequencies.
| Our repertoires of T cell receptors (TCR) give our immune system the ability to recognise a huge diversity of foreign and self antigens, and identifying the TCRs involved in infectious disease, cancer, and autoimmune disease is important for designing vaccines and immunotherapies. The majority of T cells express a TCR made up of two chains, the TCRα and TCRβ, and high-throughput sequencing of samples of T cells results in the loss of this pairing information. One can identify TCRαβ clones using single-cell sequencing, but this is costly and typically probes only part of the diversity of T cell populations. Statistical approaches are potentially more powerful by sequencing the TCRα and TCRβ in multiple samples of T cells and pairing them using their frequency of co-occurrence. However, T cells involved in immune responses frequently share TCRα and TCRβ chains with other responding cells. This promiscuity, combined with a high prevalence of T cells with two TCRα chains and sequencing errors, presents significant challenges to frequency-based pairing methods. Here we present a new algorithm that addresses these challenges and also provides accurate estimates of the abundances of T cell clonotypes, allowing us to build a more complete picture of T cell responses.
| The ability of T cells to recognise antigens is conferred by a process of gene rearrangement that generates a diverse repertoire of T cell receptors (TCR), or clonotypes. Identifying the clonotypes involved in responses against pathogens and tumours or those involved in autoimmune disease can guide the design of vaccines and immunotherapies. In addition, the breadth of a T cell response correlates positively with the efficiency of control in many viral infections [1–3]. Thus, a method to characterise the diversity of antigen-specific responses—that is, the participating TCRs and their relative abundances—may yield potential correlates of protection.
The αβ TCR is a heterodimer, generated by a combination of ordered recombination of V, D, and J gene segments for the β chain and V and J gene segments for the α chain, together with random nucleotide insertions and deletions between the gene segments. The hypervariable CDR3α and CDR3β regions contact the peptide-loaded MHC (pMHC) most closely and so are considered the primary source of specificity in binding. From hereon we will use the term ‘chain’ interchangeably with the CDR3 region of the TCRα or TCRβ. Historically, the CDR3β has been thought to contribute more to the interaction with pMHC due to its greater theoretical diversity. However, studies of crystal structures have demonstrated that CDR3α loops can have equal or greater contact with pMHC, as measured by buried surface area [4]. Epitope-specific immune responses also show biases for certain V and J segments in both α and β chains [5, 6], suggesting both chains contribute to the binding affinity. The α chain may even play a dominant role in the recognition of certain antigens [7]. Characterising the true extent of clonal diversity within T cell populations therefore requires resolving the paired CDR3α and CDR3β sequences within them.
Standard methods of multiplex PCR and high-throughput sequencing lose this pairing information and as a result are commonly used to analyze either the α or β chains alone [8–11]. More recent studies have used single-cell sequencing approaches to identify TCRαβ pairs, and, analogously, the paired CDR3 sequences from the heavy and light chains of the B cell receptor. These approaches include using single-cell sorting and RT-PCR [12–14], also with barcoding [15–18]; and variations of emulsion techniques to isolate single cells and amplify with PCR [18–20]. Drawbacks of these techniques include limited scalability, the risk of undersampling rare clones and so underestimating diversity, imprecise information regarding clonal abundances, and the need to use customised equipment [18, 21].
An alternative strategy is to use statistical methods to associate the CDR3α and CDR3β sequences obtained from bulk sequencing of multiple subsamples of T cells taken from the parent population of interest [22]. This approach exploits the fact that paired chains will tend to appear together in samples and uses the frequencies of these co-occurrences to associate them. A similar approach has been used to pair the heavy and light chains of B cells [23]. Because frequency-based pairing can be applied to large samples of cells, it has the potential to recover antigen receptors in greater depth and more economically than single-cell approaches, as well as providing more precise estimates of clonal frequencies. However, several properties of antigen-specific T cell populations present difficult challenges to this method. First, there is accumulating evidence from single-cell sequencing studies that, within an individual, T cell clonotypes specific for a given pMHC can exhibit sharing of both α and β chains [13, 14, 17, 19]. Second, between 10–30% of T cells possess two productive α chains [13, 24, 25] and 6–7% of T cells possess two productive β chains [25, 26]. The combination of sharing of α or β chains, dual TCRs, and sequencing errors can confound frequency-based methods that assume unique pairings. To illustrate, frequent co-occurrences of the three chains α1α2β in samples may derive from a single clone possessing two α chains or two clones α1β and α2β present at similar abundances, and the two possibilities are difficult to distinguish.
Here we describe a novel approach to frequency-based pairing that addresses these issues and identifies TCRαβ clones and their relative abundances using high-throughput sequencing of CDR3α and CDR3β regions. Our approach is optimised for antigen-specific populations and designed for use with cells recovered from typically-sized human blood samples. It is specifically designed to deal with promiscuity in αβ pairing, dual TCRα clones, and high rates of sequencing errors. By drawing on bulk sequencing data, we increase the efficiency of detection of rare responding clones and reduce the costs associated with single-cell high-throughput sequencing methods. The method also goes beyond other currently available approaches, yielding estimates of the frequencies of clones within their parent populations.
Performing frequency-based pairing is in principle relatively straightforward if each clone is identified by two unique TCRα and TCRβ chains. However, single-cell analyses of epitope-specific T cell populations in mice and humans have revealed significant levels of sharing of both CDR3α and CDR3β sequences at the amino acid level across clones within individuals (Table 1).
The current upper limits on estimates of the number of unique TCRβ chains in the naive CD4 or CD8 pools are 106 in mice [27] and 108 in humans [28]. As a consequence, sequencing of samples of naive T cells typically results in nearly every cell possessing a unique TCRβ (see S1 Text, Section 1). Nevertheless, the true diversity of the naive repertoire may be even greater; due to the sequence of events involved in the generation of the TCR in the thymus, we expect each TCRβ to be shared with many TCRα within the naive CD4 and CD8 T cell pools. In mice, thymocytes undergo 6–9 divisions following TCRβ rearrangement at the DN3 stage [29–32], generating 64–512 cells which then undergo independent TCRα rearrangements. Assuming 5% of these TCRαβ precursors survive selection [33–36] leaves TCRβ clone sizes of 3–25 cells post-selection [27]. Thymocytes may undergo 1 or 2 divisions at the single-positive CD4 or CD8 stage before leaving the thymus [36]; if we assume a 2-fold expansion here on average, each αβ T cell precursor at DN3 generates 6–50 new naive cells with identical TCRβ chains, comprising 3–25 unique TCRαβ clones of typically 2 cells. Comparable estimates of TCRβ clone sizes have been obtained elsewhere [27, 32]. There is also evidence that TCRβ-clone sizes can be augmented by convergent recombination of the TCRβ chain [8, 37]. If a particular CDR3β contributes strongly to the affinity of binding to a given peptide-MHC, then because the recruitment of naive antigen-specific T cells appears to be highly efficient [38], our rough quantification of TCRαβ clonality in thymopoesis is consistent with the observation that TCRβ-sharing is commonly found within epitope-specific populations (Table 1).
Because the rearrangement of the TCRα follows that of the TCRβ, any sharing of CDR3α sequences across clones presumably arises from convergent recombination. Sharing then would be expected to arise most frequently for sequences that are close to germline, containing relatively few random N-nucleotide insertions. To examine this possibility, we immunised an HLA-A2 human volunteer with the live attenuated yellow fever vaccine YFV-17D, took a peripheral blood sample 15 days post-vaccination, and used dextramer staining and single-cell RNAseq to recover paired TCRαβ sequences from CD8+ T cells specific for the immunodominant epitope HLA-A02:01/LLWNGPMAV (see Methods; data provided in S1 Dataset). Out of 256 cells, we observed 169 unique CDR3α, with 15 (8.9%) of them shared between two or more clones (Fig 1A). We examined the numbers of nucleotide insertions at the V-J junction of the CDR3α and indeed saw significantly fewer in CDR3α sequences that were shared between two or more clones (mean 2.04 insertions, n = 23) than in sequences that were unique to a single clone (mean 3.62 insertions, n = 154; p < 0.005, Wilcoxon rank sum test; Fig 1B). In summary, it appears that convergent TCRα recombination may derive at least in part from the reduced junctional diversity of clones possessing CDR3 regions that are closer to germline.
Motivated by this promiscuity of TCRα and TCRβ pairings, we developed a semi-heuristic procedure alphabetr (ALgorithm for Pairing alpHA and BEta T cell Receptors) that recovers TCRαβ pairs from high-throughput sequencing data. Fig 2 shows the algorithm schematically. The experimental procedure is to sequence the CDR3α and CDR3β regions from multiple samples of T cells from the same parent population (Fig 2A–2C). The input to the algorithm is a list of these unpaired sequences (Fig 2C), each associated with the sample it belonged to (e.g. a given well in one or more 96-well plates). Fig 2C illustrates amino acid sequences as inputs, but the algorithm can be applied equally well to data comprising nucleotide sequences and/or the addition of V(D)J segment information. The number of cells in each well can be freely varied, and indeed as we describe below, varying the sample size across the plate(s) helps to increase both the number and accuracy of pairings. Given this information, alphabetr then calculates association scores between every α and every β chain found in a randomly chosen subsample of wells. This score is the sum of the number of co-occurrences of chains in each well, each weighted inversely by the total number of chains recovered from that well (Fig 2D(ii)). The weighting factor reflects the intuitive idea that our confidence that a co-occurring α and β pair derive from the same clone decreases as the number of unique chains recovered from that well increases. The algorithm then solves a linear sum assignment problem within each well based on these plate-wide association scores to generate a list of candidate pairs of α and β sequences within each well (Fig 2D(iii)). This is a list of αβ pairs in which each α is paired with only one β, and vice versa, such that the sum of the association scores is maximised. After repeating this assignment for every well in the subset, we generate a matrix of dimensions n × m where n and m are the total numbers of unique α and β chains recovered across the plate(s), respectively, and whose entries are the number of times that each candidate pair αi βj (i ∈ {1…n}, j ∈ {1…m}) have been associated. Sharing of chains across clones is now possible in this list. Those αβ pairs that appear in a number of wells greater than the mean of the non-zero elements of this matrix are retained as a refined list of candidate pairs. The pairing and filtering process is repeated on subsets of the data (Fig 2D), and a consensus list of putative paired CDR3 sequences comprises those appearing in more than a threshold proportion of these lists (Fig 2E). This pseudo-jacknife procedure acts to reduce the effect of very common clones pushing up the threshold for inclusion in the filtered list and increases the efficiency of pairing of rarer clones, while minimising the inclusion of incorrect αβ pairs. Steps A-D are described in more detail in Methods.
The algorithm then uses a maximum likelihood approach to estimate the relative frequencies of the clones associated with each candidate αβ pair (Fig 2F; Methods). These estimated frequencies are then used with the patterns of co-occurrences of chains to distinguish between β-sharing and dual TCRα clones (see Methods). This step also yields refined estimates of the frequencies of dual TCRα clones. The output of the algorithm is a list of single or dual TCRα clones together with estimates of their abundances within the parent population (Fig 2G).
To test the performance of alphabetr, we first used artificially generated datasets mimicking the bulk sequencing of CDR3α and CDR3β regions from polyclonal T cell populations. We assumed skewed distributions of clone sizes, with between 5 and 50 clones comprising the most abundant 50% of the population and the remainder, approximately 2000 clones, forming a flat tail at low frequency (see Methods). These distributions were chosen to reflect plausible immunodominance hierarchies within T cell responses, motivated by analysis of epitope-specific cells recovered from human subjects immunised with live attenuated yellow fever virus vaccine (our analysis and ref. [11]). We also analysed different sizes of parent populations (see S1 Text, Section 2). Within these hierarchies we allowed the virtual clones to exhibit sharing of CDR3α and CDR3β at ranges of frequencies consistent with published single-cell TCR sequencing studies (Table 1) and our own data (Fig 1A). We also allowed between 10% and 30% of clones to express two productive TCRα chains and 6% of clones to express two productive TCRβ chains. The sequences in each ‘well’ were then generated by sampling between 10 and 300 T cells from the parent population with replacement. Selecting an optimal pattern of sampling is an issue we return to below.
To assess the robustness of alphabetr, we simulated the properties of two forms of sequencing error: dropping of chains and productive in-frame sequencing errors. Dropping of chains represents the failure of CDR3α and/or CDR3β regions to amplify or be detected, a process which likely has both purely random and clone-specific elements [22]. To model this, each clone was assigned a drop rate at random from a lognormal distribution with mean 0.15 and standard deviation of 0.01, with the rate capped at 0.9. Each instance of a CDR3α and CDR3β from that clone was then removed from the well with probability equal to the drop rate. To model productive in-frame sequencing errors, every unique CDR3α and CDR3β was assigned an error rate randomly drawn from a lognormal distribution with mean 0.02 and standard deviation 0.005. Each instance of a sequence at the per-cell level was replaced at random by one of three erroneous ‘daughter’ sequences, unique and specific to the parent sequence, with probability equal to the sequence-specific in-frame error rate. Thus on average each CDR3α and CDR3β generated mutant offspring sequences at the rate of 2% per instance in each cell in the plate(s).
We then assigned identifiers to the remaining CDR3α and CDR3β sequences, associating them with the sample’s location in a virtual 96-well plate. The input to the algorithm is the list of these unpaired CDR3α and CDR3β sequences together with their well-identifiers. This process was repeated for different sampling strategies (varying the sample sizes within each well, and using one or five 96-well plates); different clonal size distributions; and different degrees of CDR3α and CDR3β sharing. Under these ranges of conditions, the algorithm was tested for the following:
alphabetr does not attempt to identify dual TCRβ expressing cells because dealing with this relatively infrequent phenomenon together with dual TCRα chains and sharing of both TCRα and TCRβ chains across clones is extremely challenging algorithmically. However, we include dual TCRβ cells in our simulated data at the level of 6% to establish their impact on the algorithm’s performance.
Applying high throughput single-cell sequencing technologies to very large numbers of T cells is becoming increasingly within reach, but smaller-scale solutions using frequency-based sampling potentially remain far more economical. While another implementation of this strategy exists [22], the promiscuous nature of TCRα and TCRβ usage within epitope-specific populations presents multiple challenges to frequency-based methods that have not been addressed to date, to our knowledge. The combination of alphabetr and relatively low-cost sequencing strategies addresses these issues, being capable of handling a wide range of clonal structures—skewed abundances, dual TCRα, sharing of both TCRα and TCRβ between clones—as well as providing estimates of clonal abundances. The algorithm is available as a documented package in R [44] from http://github.com/edwardslee/alphabetr.
Single-cell technologies clearly allow the identification of large clonal expansions within populations. Our algorithm offers the potential to both identify these common clones as well as achieve depths of coverage of rarer clones that far exceed those currently possible with reasonable levels of single-cell sequencing. Given the correlation between diversity of immune responses and protection, this characterisation of the full diversity of T cell responses may be a better prognostic indicator than simply identifying common clones. Further, establishing the levels of TCRα- and TCRβ-sharing within populations sheds light on mechanisms of antigen recognition, repertoire diversity, and the efficiency of recruitment into immune responses.
Our analysis demonstrates that the most difficult of these challenges is to reliably distinguish between abundant TCRβ-sharing or dual TCRα clones within highly skewed populations because the expected patterns of co-occurrences of the three chains under the two alternatives are very similar when sequencing samples of a few tens of cells per well; all three chains typically appear in nearly all the wells. The difference in patterns can be magnified to an extent by sampling very few numbers of cells per well, but this solution comes with the cost of a reduction in total sample size, sacrificing depth of recovery of rarer clones. One might suppose that the high prevalence of dual TCRα clones in the naive T cell pool favours that scenario over TCRβ-sharing. However, our immunological intuition here may be misleading. Naive T cell precursor numbers may be in the range 10–1000 cells in mice [45–47], which we estimate is comparable to or larger than the size of TCRβ-sharing populations exported from the thymus. If the sharing of a TCRβ between clones confers overlap in their TCR specificities, and if recruitment into immune responses is efficient, we might expect to see significant levels of TCRβ-sharing within expanded, epitope-specific populations. Indeed, as shown in Table 1, TCRβ-sharing has been seen to reach levels of up to 25% in responses to influenza epitopes in naive mice [13, 14] and almost 40% in secondary responses [14]. It also occurred at a level of 2% in our analysis of TCRα and TCRβ usage among CD8+ cells specific for a YFV epitope in a human volunteer. The TCRβ-sharing/dual TCRα ambiguity is therefore a robust feature of epitope-specific responses, and is challenging to unravel fully with statistical approaches.
There are at least three ways to address this problem. One solution is to pair alphabetr with, for example, one plate of single-cell samples. Since the ambiguity is only manifest strongly with common clones, this limited amount of extra information may serve to resolve the issue. A second approach is to exploit the fact that 30%-40% of clones will yield both an in-frame and an out-of-frame CDR3α sequence [13]. Currently, out-of-frame sequences are not utilised by alphabetr; one could extend it to include them and associate clones with their out-of-frame sequences. Clones possessing one in-frame and one out-of-frame CDR3α could then be excluded from the list of dual TCRα candidates, which would assist β-sharing/dual TCRα discrimination. A third possibility is to extend the algorithm to exploit the sequence information itself. If dealing with epitope-specific populations, we might expect more sequence similarity in the CDR3α in two β-sharing clones than in a dual TCRα case. In the latter, the two CDR3α sequences are likely unrelated because presumably only one of the TCRα chains is involved in antigen recognition and they rearrange independently.
In practice, one needs a strategy for implementing alphabetr on a given sample of T cells with no a priori knowledge of the number or size distribution of clones. Assuming the number of cells is not limiting, we advocate a high-mixed sampling approach that involves sampling 20–300 cells per well and deals efficiently with a wide range of clonal abundances. When alphabetr is implemented as described here, a standard desktop computer with 16 Gb of RAM is able to handle samples from parent distributions of up to 4000 clones. When sampling populations with much fewer clones, lower numbers of cells/well are needed to avoid high false pairing rates. Assuming cell numbers are not limiting, bulk sequencing of the β chain could be used to gain a rough estimate of the richness of the parent distribution and so indicate when a sparse sampling strategy would be beneficial. In situations where cell numbers are limiting, one approach could be to begin with a single plate of 10 cells/well to obtain a rough lower bound on the richness of the distribution and apply a low or high mixed sampling strategy with the remaining cells from the sample, as appropriate. The single plate of 10 cells/well is then still usable for the pairing process and for frequency estimation.
While we have framed our analysis around the sequencing of epitope-specific populations, alphabetr can equally well be applied more generally to T cell populations of restricted and potentially skewed polyclonality, such as tumour infiltrating lymphocytes or T cells extracted from sites of autoimmune responses. It therefore has immediate applications in cancer immunotherapy and other personalised immunomodulatory treatments. Until single-cell sequencing becomes more affordable, frequency-based pairing methods provide a rapid and economical means of characterising the clonal structure of T cell populations.
All experimental procedures were approved by the Regional Ethical Review Board in Stockholm, Sweden: 2008/1881-31/4, 2013/216-32, and 2014/1890-32.
Our approach exploits the fact that TCRα and TCRβ sequences (referred to as α and β chains) will tend to appear together in wells. Let Nα be the total number of unique α chains, Nβ be the total number of unique β chains, and the α and β chains found in the data set be labelled from 1 to Nα and from 1 to Nβ respectively. The degree of association between chains αi and βj is measured by a score Sij,
S i j = ∑ k = 1 W δ i j k c α k + δ i j k c β k , (1)
where the wells in the data are labelled from 1, 2, …, W, the numbers of distinct α and β chains in well k are c α k and c β k respectively, and δ i j k is 1 if both αi and βj are found in well k and 0 otherwise. Eq 1 sums the co-appearances in wells, each weighted inversely by the total number of α and β chains recovered from the well. The scaling accounts for the fact that the larger the number of unique chains in a well, the lower our confidence that a co-occurring α and β pair derive from the same clone.
The algorithm begins by sampling a proportion pJ of the wells in the data without replacement. For all analyses presented here, we used pJ = 0.75, which provided a good balance between depth and false pairing rate. The algorithm computes the association scores between every unique α and β chain using Eq 1 based on the sampled subset of wells. Let A k denote the set of A distinct α chains found in well k, that is A k = { α m 1 k , α m 2 k , … , α m A k }, where the m i k ∈ { 1 , … , N α } are integers that denote the labels of the A TCRα chains found in well k. Similarly, let B k denote the set of B distinct β chains found in well k, that is B k = { β n 1 k , β n 2 k , … , β n B k }, where the n i k ∈ { 1 , … , N β } subscripts denote the labels of the B TCRβ chains found in well k. The algorithm solves the following linear assignment problem using the Hungarian algorithm [39]:
maximize∑αi∈Ak∑βj∈BkSijxijsubject to∑αi∈Akxij=1 for βj∈Bk∑βj∈Bkxij=1 for αi∈Akxij≥0, αi∈Bk, βj∈Ak, (2)
where xij = 1 indicates that αi and βj are assigned as a candidate TCR pair and xij = 0 otherwise. A pair αiβj is defined as an assigned pair of well k if xij = 1 for Eq 2 associated with well k. The number of assignments made for every pair of α and β is recorded as Xij, i.e. Xij equals the number of times xij = 1 from the solutions of Eq 2 for each well in the subset. We then calculate a filter level F that determines the minimum number of assignments required for an assigned candidate pair of α and β chains to be determined as a true TCR pair. The filter-level F is chosen to be the mean of the elements of the set {N(i, j) : N(i, j) > 0, i ∈ 1, 2, …, Nα, j ∈ 1, 2, …, Nβ}, where N(i, j) is the number of times αi βj are assigned to each other, The output of this algorithm is then a list of candidate αβ pairs that may be associated with T cell clone. At this stage, dual TCRα cells are not identified; thus a clone α1α2β may be represented in this list as one or both of α1β and α2β.
The procedure above is performed Nr times on random subsets of the wells (all simulations in this paper use Nr = 100), and each replicate yields a list of candidate αβ pairs. We then perform a filtering or consensus step in which only αβ pairings that appear in more than a threshold proportion T of these lists are retained as candidates. The simulations we present in the text explore thresholds of T = 0.3, 0.6, and 0.9.
We use maximum likelihood to infer clonal frequencies based on the number of wells in which a pair of α and β chains both appear. Let N = {n1, n2, …, ns} be the set of s distinct sample sizes (ni cells per well) in all of the wells and W = {w1, w2, …, ws} where wi represents the number of wells with samples of size ni cells. Let cij denote the clone with chains αi and βj and let k i j l denote the number of wells of sample size nl cells per well that contain chains αi and βj. The likelihood of the observations k i j ( . ), given that the clone cij is present at frequency fij within the population, is
L ( observed incidence of clone c i j | f i j ) = ∏ l = 1 s w l k i j l 1 - q l k i j l q l w l - k i j l (3)
where ql is the probability of clone cij not being found in well l and is given by
q l = 1 - f i j n l + ∑ m = 1 n l 2 ϵ m - ϵ 2 m n l m f i j m 1 - f i j n l - m . (4)
Here ϵ is the average probability that a CDR3 sequence in a cell fails to be amplified and sequenced. For every clone cij, the algorithm maximises Eq 3 to estimate its frequency fij, and 95% confidence intervals are defined by the frequencies yielding log L = log L max - 1.96. Details of the derivation of Eqs 3 and 4 are given in Section 4 of S1 Text.
This procedure is applied to every αβ pair identified in the first phase of the algorithm. These estimated frequencies are used to distinguish TCRβ-sharing clone pairs from single TCR clones expressing two TCRα. This procedure is described in the following section. When a clone with two TCRα is identified, we revise the frequency estimate as follows. Let c(ij)t denote a clone with chains αi, αj, and βt, and k ( i j ) t l denote the number of wells of size nl that contain chains αi, αj, and βt. The likelihood of the observations given that clone c(ij)t has a frequency f(ij)t ∈ (0, 1] is
L ( observed incidence of clone c ( i j ) t | f ( i j ) t ) = ∏ l = 1 s w l k ( i j ) t l 1 - q l k ( i j ) t l q l w l - k ( i j ) t l (5)
where ql is the probability of clone c(ij)t not being found in well l and is given by
q l = 1 - f ( i j ) t n l + ∑ m = 1 n l 3 ϵ m - 3 ϵ 2 m + ϵ 3 m n l m f ( i j ) t m 1 - f ( i j ) t n l - m (6)
where ϵ is the mean drop rate as described above. Eq 5 is then maximised to estimate f(ij)t, and again log L = log L max - 1.96 is used to calculate 95% confidence intervals.
If the algorithm yields two clones that appear to share a TCRβ (α1β and α2β), we must decide whether this is indeed a β-sharing pair of clones or that the association derives from one dual TCRα clone (α1α2β). To do this, we use the likelihoods of observed co-occurrences of the three chains to assess the relative support for the two alternatives.
Let cij = (αi, βj) and ckj = (αk, βj) be two putative clones with a common TCRβ chain βj. We count the number of wells containing all three-way, two-way, and single appearances of the three chains. We then calculate the ‘full’ likelihoods of this pattern of occurrences under two hypotheses: (A) that cij and ckj are indeed two β-sharing clones, with frequencies fij and fkj estimated using Eq 3; and (B) that the chains derive from one dual TCRα clone c(ij)k present at frequency f(ij)k, estimated using Eq 5. If the difference log L B - log L A ≥ 10, we assume the three chains derive from dual TCRα clone.
The calculation of these full likelihoods is in Section 6 of S1 Text but is computationally tractable only for wells with less than 50 cells due to the need to calculate large multinomial coefficients. The full-likelihood method is therefore only appropriate for estimating frequencies of those relatively abundant clones that are commonly found in the wells with smaller sample sizes. We use a more restricted likelihood-based approach for discriminating β-sharing and dual TCRα among rare clones, which tend to appear only in larger samples. Let clones cij = (αi, βj) and ckj = (αk, βj) be two clones with a common beta chain βj, and let fij and fkj be their estimated frequencies. The algorithm calculates the ratio r i k j of the observed to the expected number of wells in which all three chains from the putative β-sharing pair cij and ckj co-appear, under the hypothesis that they are indeed two clones and not a dual TCRα:
R = r i k j = A c i j , c k j E c i j , c k j : i ≠ k , j ∈ 1 , 2 , … , N β (7)
where A(cij, ckj) is the number of times clones cij and ckj are observed to appear in the same well and Nβ is the number of distinct β chains, and the expected number is
E c i j , c k j = ∑ l = 1 s w l 1 - 1 - f i j n l - 1 - f k j n l + 1 - f i j - f k j n l (8)
(see S1 Text, Section 5 for a derivation and discussion of this equation). We then partition the set of ratios R into two groups C1 and C2 using k-means clustering, where the mean of ratios of C1 is greater than the mean of the ratios of C2 (see S1 Text, Fig G for an example). The clones associated with the ratios in C1 are chosen as dual TCR clones, such that if r i k j ∈ C 1, then clones cij and ckj are removed from the list of TCR pairs and replaced with a dual TCRα clone αi αk βj.
We created synthetic data sets reflecting the properties of antigen-specific T cell populations and sequencing errors. The data sets were sampled from a population of T cell clones where a significant proportion of α and β chains are shared and 10%-30% of clones have dual TCRα chains (e.g. three clones can have the following chains: αi βk, αj βk, and αj αh βl). The sharing of β chains was set such that 85.9% of β chains were uniquely from one clone, 7.6% shared by two clones, 3.7% shared by three clones, 1.9% by four clones, and 0.9% by five clones. The sharing of α chains was set such that 81.6% of α chains were uniquely from one clone, 8.5% shared by two clones, 2.1% shared by three clones, 0.7% shared by four clones, 3.3% shared by five clones, 0.5% shared by six clones, and 3.3% shared by seven clones. We determined these levels of sharing by averaging those from the published single-cell data shown in Table 1.
The frequencies of the N clones were drawn from a skewed distribution in which ns clones comprise a proportion ps of the population and the other N − ns clones evenly represent 1 − ps of the population. The clone ranked ith in abundance then has frequency fi where
f i = f 1 + r i - 1 if i = 1 , 2 , … , n s p s / ( N - n s ) if i = n s + 1 , n s + 2 , … , N (9)
where the frequency of the largest clone f1 and the step size r are determined by solving the equations
∑ i = 1 n s f i = p s , f n s = 1 . 1 × p s N - n s . (10)
The frequency of the smallest clone in the top 50%, fns, is set to be 10% higher than the frequency of the clones in the tail. All simulations were based on ps = 0.5. We varied the number of top clones ns between 5 to 50 to test how skewness in the antigen-specific T cell population impacts the performance of the algorithm.
In order to make the simulated data more realistic, experimental noise was included in the forms of ‘dropped’ chain errors and in-frame sequencing errors. Dropped chains are CDR3 sequences that fail to be sequenced due to PCR errors and/or sorting problems, and studies utilising both single-cell and many-cell techniques have reported average drop rates of 8% to 10% [17, 22]. In the simulations, each clone was assigned a drop rate from a lognormal distribution with a mean of 0.15 and standard deviation of 0.01, and every TCRα and TCRβ chain belonging to that clone was assigned that drop rate. In-frame errors cause a CDR3 sequence to be falsely identified with an incorrect productive nucleotide and/or amino acid sequence. In the simulations, each distinct sequence was assigned an in-frame error rate drawn from a lognormal distribution with a mean of 0.02 and a standard deviation of 0.005. The error model was simulated as follows: when a cell is sampled into a virtual well, each of its chains fails to be sequenced with probability equal to the pre-assigned, clone-specific drop rate. Every surviving chain produces one of three randomly chosen, distinct, and chain-specific false sequences with probability equal to that chain’s pre-assigned in-frame error rate.
A human volunteer was identified as HLA-A2+/HLA-B7+ and received the live attenuated yellow fever vaccine (YFV-17D). On day 15 post-vaccination, peripheral blood samples were taken, and live CD3+CD8+ T cells were isolated by negative selection using magnetic columns (Miltenyi Biotec, CD8+ T cell negative isolation kit). Cells were labeled with a panel of antibodies and the HLA-A02:01/LLWNGPMAV dextramer representing the immunodominant response. Single dextramer-specific CD3+CD8+ T cells were sorted into individual wells in 96 well plates containing a lysis buffer (0.4% Triton, RNAse inhibitor, dNTP, OligodT) and immediately stored on dry ice. Single cell transcriptome libraries were subsequently generated from these cells using an adapted version of the SMRT-Seq2 protocol [48]. Libraries were prepared for sequencing by tagmentation and labelling individual single cell transcriptomes with a custom Tn5 enzyme [49] and Nextera XT dual indexes. Pooled libraries were then sequenced using an Illumina Hiseq2500 on high output mode (2 × 100bp or 2 × 125bp reads), and individual TCRα and TCRβ chains were identified using the MiTCR algorithm with default parameters. The default settings for MiTCR were used to align the CDR3 sequences. These were then manually filtered to remove erroneous sequences (e.g. early stop codons and CDR3 sequences that were greater than 30 amino aids in length), and then BLAST was used on the remaining sequences to check for mapping to other parts of the genome, removing as appropriate. All clones used in the comparative analysis of CDR3α lengths were curated manually to exclude the possibility of contaminating TCR sequences.
CDR3 amino acid sequences are provided as a CSV file in S1 Dataset, and the raw reads are deposited in the Gene Expression Omnibus (GEO), GSE75659; Sequence Read Archive (SRA), SRP066963.
|
10.1371/journal.pcbi.1000440 | PoreWalker: A Novel Tool for the Identification and Characterization of Channels in Transmembrane Proteins from Their Three-Dimensional Structure | Transmembrane channel proteins play pivotal roles in maintaining the homeostasis and responsiveness of cells and the cross-membrane electrochemical gradient by mediating the transport of ions and molecules through biological membranes. Therefore, computational methods which, given a set of 3D coordinates, can automatically identify and describe channels in transmembrane proteins are key tools to provide insights into how they function.
Herein we present PoreWalker, a fully automated method, which detects and fully characterises channels in transmembrane proteins from their 3D structures. A stepwise procedure is followed in which the pore centre and pore axis are first identified and optimised using geometric criteria, and then the biggest and longest cavity through the channel is detected. Finally, pore features, including diameter profiles, pore-lining residues, size, shape and regularity of the pore are calculated, providing a quantitative and visual characterization of the channel. To illustrate the use of this tool, the method was applied to several structures of transmembrane channel proteins and was able to identify shape/size/residue features representative of specific channel families. The software is available as a web-based resource at http://www.ebi.ac.uk/thornton-srv/software/PoreWalker/.
| Transmembrane channel proteins are responsible for the transport of ions and molecules through biological membranes and are pivotal for the physiology of the cell. In fact, their incorrect functioning is involved or related to several diseases (diabetes, myotonia, Parkinson's disease, etc.). Moreover, their specificity and selectivity to different ions or molecules have been hypothesized and sometimes shown to strongly depend on the shape and size or amino acid composition of the channel. Therefore, computational methods to identify and quantitatively characterise channel geometry in transmembrane protein structures are key tools to better understand how they function. We have developed PoreWalker, a new method to detect and describe the geometry of these channels in transmembrane proteins from their 3D structures. The method is fully automated, very user-friendly, identifies the location of the channel and derives a number of channel features: diameter profiles at given heights along the channel, all the residues lining the channel walls, size, shape and regularity of the channel. These features can be very helpful in the study of how these channels might function. We have applied PoreWalker to several channel protein structures and were able to identify shape/size/residue features that were representative of specific channel families.
| Transmembrane channel proteins play pivotal roles in maintaining the homeostasis and responsiveness of cells and the cross-membrane electrochemical gradient by mediating the transport of ions and molecules through biological membranes [1]. For instance, aquaporins facilitate the flux of water and small uncharged solutes across cellular membranes and, in humans, are involved in several diverse functions, like concentrating urine in kidneys and participating in forming biological fluids [2]–[5]. In contrast, potassium channels are fundamental regulators of cell membrane potential and control the action potential waveform and the secretion of hormones and neurotransmitters [6]–[8]. Moreover, a family of transmembrane proteins, known as translocons, have been found to mediate protein transfers between different cellular compartments and consequently to be involved in the folding of membrane and secretory proteins [9]. Understanding the structure and function of transmembrane channel proteins and studying their properties and biochemical mechanisms is therefore a very important task in biological and pharmaceutical research [10],[11].
Transmembrane channel proteins usually show a cavity spanning the whole protein, herein designated as the pore, which forms the path used by ions and/or molecules to cross the membrane. The pore has two openings, one on each side of the membrane, and it has been hypothesized (and in some cases shown) that the specificity and selectivity to different solutes is strongly dependent on the particular structural or amino acid composition features of the channel [5],[8],[12]. Consequently, computational methods for the identification and description of pores in transmembrane protein 3D-structures represent key tools to gain insights into how these proteins function.
To our knowledge, several methods for the analysis of protein surface and cavities have been developed [13]–[19] but the only currently available method for the structural analysis and visualisation of transmembrane channels is HOLE, developed in 1993 and still widely used [20],[21]. This elegant algorithm implements a Monte Carlo simulated annealing approach to find the path that a sphere of variable radius can use to go through a channel and also provides pore anisotropy analysis and conductance prediction tools. The path is optimised so that it can be considered as the route of a plastic sphere squeezing through the channel, i.e. at each point of the path the channel can accommodate the largest possible sphere. Three more recent methods, developed for the detection of internal cavities and tunnels in any protein structure, CAVER [22], its improved version MOLE [23] and MolAxis [24] can be applied to identify pores in transmembrane proteins. CAVER explores the protein inner space using a grid-based approach, while MOLE implements an algorithm based on Voronoi polyhedra. Both approaches use an optimality criterion based on the minimization of a cost function, which depends on reciprocal atomic distances, and calculates the optimal way out from a user-specified starting point inside the protein to the outside environment. MolAxis exploits computational geometry techniques, in particular the alpha shapes theory and the medial axis concept, to detect possible routes that small molecules or ions can take to pass through channels and cavities. It is worth highlighting here that all the four programs, to be applied to transmembrane proteins, require user-defined specific information about the geometry of the channel that necessitate a fairly good knowledge of the location of the pore and/or of key residues lining the pore walls, like a starting point for the path search through the channel or a vector approximating the location of the pore within the protein 3D-structure. Moreover, they provide only a limited description of the channel geometry mainly consisting of diameter values and some of the residues lining the pore walls.
Herein we present PoreWalker, a method to provide a detailed description of the three dimensional geometry of a channel (or pore) through a transmembrane protein, given the coordinates of the protein structure. These 3D pore descriptors provide a quantitative description, including the size, shape and regularity of the pore, which we hope will help to explain pore specificity, the critical biological function of these molecules. PoreWalker is fully automated, requiring only the 3D protein coordinates from the PDB file, and so can be applied to any new structure or across all transmembrane proteins in the PDB. The method was applied to several structures of transmembrane channel proteins and was able to identify shape/size/residue features representative of specific channel families. The software is implemented as a web-based resource at http://www.ebi.ac.uk/thornton-srv/software/PoreWalker/ and its source codes will soon be available upon request to the authors.
The main goal of PoreWalker is to identify a channel in a transmembrane protein, through which a ligand (an ion or a small molecule) might pass. The channel can be defined by the pore-lining residues, which in most cases are accessible to solvent, and approximated by an axis, which ideally connects the two entrances of the pore and passes through the centre of the channel. The computational challenge is to identify the pore-lining residues from the protein 3D-structure and thus to define the axis and centre of the pore. Once this is done, the parameters defining the size and shape of the pore are easy to calculate.
The most direct approach to defining the channel would be to find the solvent accessible residues, and from these residues try to identify the pore-lining residues using geometric criteria. The most straightforward way to distinguish pore-lining residues from other accessible residues (on the outside of the protein) is to require that they ‘point towards’ the pore axis. However, since the latter cannot be defined without first knowing the pore-lining residues, it is necessary to adopt an iterative approach, whereby the axis is first approximated and then refined.
Ideally, the path taken by the ligand through the transmembrane protein will be linear and the pore will run approximately perpendicular to the plane of the membrane. Therefore, to make the first estimate of the channel axis, the algorithm takes into account the ‘special’ geometry of transmembrane proteins, in which the protein's secondary structures also tend to lie perpendicular to the transmembrane plane, running from one side of the membrane to the other. The channel axis is thus approximated as co-linear with these secondary structures and passing through their averaged centre of gravity.
In practice, paths can be convoluted and channels can be far from linear, as for the pores of some acid-sensing ion channels. Moreover, pores can be very narrow, with diameter values less than 1Å, so that pore-lining residues can not be straightforwardly detected as accessible to solvent. Therefore, the algorithm identifies a number of “local” pore centres at different pore heights (or slices) through the membrane so that the geometrically correct pore openings and path can be detected and a refined pore axis generated.
The algorithm is heuristic and iterative, and includes the following steps (see Figure 1):
All programs included in the PoreWalker pipeline are developed in-house in C and PERL programming languages. The web-server is based on PERL-CGI protocol and the results of the four step calculations are summarised in pictures and text files displayed on the website and downloadable.
In transmembrane proteins, the channel runs approximately perpendicular to the membrane plane and parallel to the bundle or barrel that makes up the transmembrane portion of the pore. The first step of the program consists in re-orienting the protein structure so that the origin lies at the centre of gravity of the transmembrane portion of the protein and the bundle/barrel lies perpendicular to the membrane plane. The main axis of the transmembrane bundle/barrel is calculated according to the position of the secondary structure elements that putatively form it. Each secondary structure element in the protein is identified from the separation of sequential C-alphas as described in Supplementary Text S1 and, if the helix or the strand is longer than 15 or 10 amino acids, respectively, it is approximated by a vector, which starts at its centre of mass and points toward the centre of mass of the terminal four and two amino acids of the helix or strand, respectively. The length threshold was applied because, on average, transmembrane helices and strands used for this calculation need to be sufficiently long to cross the membrane. This excludes small helices which often do not lie perpendicular to the membrane plane. The sign of all the vectors is selected so that they point in approximately the same direction and the averaged vector is calculated. However, outlying secondary structures found to be more perpendicular than parallel to the bundle/barrel axis are excluded from the averaging at this stage so that the transmembrane portion of the structure is orientated as parallel to the membrane axis as possible.
The whole protein 3D structure is then re-oriented so that its calculated main axis overlaps with the x-axis of the current 3D system and the centre of gravity of its transmembrane portion lies at the origin. In this way, the structure is moved into a new reference frame that approximately aligns the transmembrane secondary structure elements perpendicular to the membrane. The pore axis is then approximated as coinciding with the protein main axis (see Figure 2, step 2). This starting assumption, despite its crudeness, simplifies and speeds up the following steps of the method.
The centre of the pore is defined by iteratively maximising the number of detected putative pore-lining residues, i.e. water-accessible amino acids pointing towards the pore axis. At the beginning of the process, the centre of the pore and the pore axis, i.e. the linear vector going through the middle of the pore, are assumed to correspond to the centre of mass of the protein and to the x-axis, respectively. Putative pore-lining amino acids around the pore axis are then selected to satisfy three criteria: (1) the relative sidechain solvent accessibility calculated by NACCESS ([25], downloadable at http://www.bioinf.manchester.ac.uk/naccess/) must be higher than 5%; (2) the vector defined by the C-alpha-C-beta bond must point towards the pore axis; and (3) the distance of the C-alpha atom from the pore axis must be below a given threshold. The distance threshold is calculated at each iteration as the smallest distance between any pore-lining residue C-alpha and the current pore axis plus 6 Å. This prevents the inclusion of “second shell” residues in the selection of putative pore-lining residues and in the calculation of the final centre of the pore. Glycines lack C-betas and are therefore treated differently. For each Gly, a dummy atom is defined as the average of 3D-coordinates of its backbone carbonyl carbon and amide nitrogen. This atom can be considered a mirror image of the C-betas of a virtual side chain located between the two hydrogen atoms bound to its C-alphas and can therefore be used to evaluate the orientation of Gly backbone atoms. Glycines with a total relative accessibility higher than 5% and with the dummy atom pointing away from the pore axis are defined as pore-lining.
A new centre of the pore is then calculated from the selected putative pore-lining amino acids and the protein structure is translated so that the new pore centre and the x-axis corresponds to the origin of the 3D-system and to the new pore vector, respectively.
The above procedure is performed iteratively and stops when the number of newly selected putative pore-lining residues converges to its maximum, indicating that the pore centre has reached its optimal position. As a result of this first process, the protein structure is translated in space so that the x-axis goes through the current best-guess of the centre of the pore and a preliminary list of putative pore-lining residues is generated (see Figure 2, step 3).
The effectiveness of this step of the method was assessed by monitoring the distance of the selected pore-lining residues from the pore centre, as described in Supplementary Text S1 and shown in Figure S1.
To derive the best possible axis and cavity of the pore an iterative slice-based approach is used, in which the centre of the pore is systematically optimised for each slice and therefore eventual irregularities in the cavity can be detected. At each iteration, the protein structure is mapped onto a 3D-grid of 1Å steps and then sliced along the x-axis (i.e. the current pore axis) in 1Å thick layers. The pore centre of each slice is then identified by a grid-based approach so that it lies at the centre of the sphere with the maximum radius that the slice can accommodate. The maximum sphere and its centre are derived by expanding the sphere from the current centre until it clashes with a pore-lining atom, and systematically shifting the centre on the vertices of a 2D-grid so that the centre of the sphere of maximum volume for that slice can be identified.
The pore centre of the slice is initially set as the average of C-alpha and C-beta atoms of the putative pore-lining amino acids belonging to the slice selected in the previous step of the program, and the corresponding maximum sphere is calculated. A square 2D-grid perpendicular to the current pore axis (x-axis) is then built and used to optimize the location of the pore centre. The grid has 0.1 Å squares, it is centred at the pore centre, and its size depends on the sequence length of the protein and on the size of the pore. Grid vertices not surrounded by atoms in all the possible y and z directions are taken as located outside the pore and excluded from the optimization process. The sphere of maximum volume at a given centre is calculated by increasing its radius by 0.1 Å until it hits a vertex of the 3D-grid occupied by a backbone or C-beta atom. The current sphere radius is adjusted by subtraction of the atomic van der Waals radius (1.8 Å, corresponding to the average radius of all types of heavy atoms found in protein structures as in the AMBER united force field [26]) or approximate residue side chain radius (as in Levitt's amino acid ‘lollypop model’ [27]) if a backbone atom or a C-beta is hit, respectively. If the radius value is above any previously calculated radius, the current radius and corresponding sphere centre are taken as the maximum radius and pore centre for that slice.
At the end of the iteration, coordinates of the last four consecutive sphere centres at each end of the pore, that represent the two pore openings, are averaged to generate two points, which define the new pore axis. The structure is then re-oriented to align with the new vector (see Figure 2, steps 4–8). The last four consecutive spheres are used because the ends of the channels can be very irregular in term of shapes and therefore pore axes derived from the two very last sphere centres (one per end) often do not cross correctly one or both the pore entrances (the value 4 was derived on a trial-and-error basis in the range of values from 1 to 5).
The refinement process stops when the new pore vector “overlaps” to the old pore axis (i.e. when their angle is lower than 0.5 degrees) and the current pore axis and maximum sphere radii (i.e. those calculated in the previous iteration) are retained as optimal and used in the further analysis of the pore shape.
The last step of the method is the analysis and calculation of three main pore descriptors: the pore-lining atoms and residues (Section 4.1), and the shape of the pore cavity (Section 4.2) and its regularity (Section 4.3).
Pore descriptors calculated by PoreWalker for a submitted structure are summarised in the corresponding output webpage, which shows the features of the channel cavity and several visualizations of the pore based on the identified pore-lining residues. As an example, the output of the bovine aquaporin-1 (PDB code 1j4n) is summarised in Figure 3. The 3D shape of the pore is simplified in 2D as a stack of building blocks shaped as trapezia for funnel-like shapes (Figure 3B) going from the most negative to the most positive coordinate along the pore axis. In addition, the pore cavity is represented as a series of consecutive straight and wiggly lines representing channel areas where pore centres can (straight) or cannot (wiggly) be fitted to a line, respectively (Figure 3E). It is worth highlighting here that the approach does not take into account any chemistry (e.g. H-bonds) but just calculates the path of the pore centres. In practice, ions/molecules may well hop between low energy off-centre sites, within the channel, that optimize their interactions with pore residues during their passage through the channel.
Vertical and horizontal visualizations of the pore help to provide a better understanding of the channel features. Vertical sections (Figure 3A,D) are generated halving the protein structure along the pore axis, while horizontal sections (Figure 3G,I) are produced as 5Å slices of the protein structure perpendicular to the pore axis. Amino acids are coloured according to whether they are classified as pore-lining and red spheres represent optimal pore centres.
PoreWalker was tested on the 19 structures from the “Membrane Proteins of Known 3D Structure” resource (http://blanco.biomol.uci.edu/Membrane_Proteins_xtal.html) listed in Table 1, that include both ion and small molecule channels with straight and curve pores. Results are shown in Table 1, Figure 4 and Supplementary Figure S2. Although there is no fully comprehensive experimental data to assign with certainty the location and residue composition of channels in transmembrane protein 3D-structures, the position of the pore axis and of the pore centres, visually analysed in relation to the protein structure, and the minimum diameter value give a hint of the effectiveness of the method. From visual inspection, PoreWalker seems able to locate correctly the pore axis and the pore centres in most of the cases and therefore to identify fairly correctly the amino acids that line the pore walls with one or more atoms. In fact, the pore axis seems wrongly located only for the Amt-B (Figure 4K), Amt-1 (Supplementary Figure S2E) and the SecYE-beta translocon (Figure 4H) channels (PDB codes 1xqf, 2b2f and 2yxr, respectively). Both Amt-B and Amt-1 channels share a common hour-glassed shape with multiple exits at one of the pore gates and can therefore be thought to include more than one transmembrane tunnel of different length (Figure 5B). Likewise, the SecY-beta translocon shows two flexible loops at both sides of the channel that make a further narrower but longer cavity crossing the protein structure. Despite the misassignments of pore axis and pore centres, in these three examples most of the pore-lining residues still seem to be identified correctly because the calculated optimal cavities, indicated by red spheres, partially overlap with the “true” cavities, indicated by the black arrows.
In terms of pore shape, PoreWalker seems to recognise common sub-shapes across channel families. For instance, all aquaglyceroporins show a DU-like string shape (where D and U represent funnel-like shape of decreasing and increasing diameter, respectively), which represents a hour-glasssed shape confirmed by a few published data [5],[28],[29]. Likewise, potassium channels present a shared sub-shape, a DUD sub-string shape at the cytoplasmic side of the channel, that is in agreement with the channel features reported by Mackinnon et al., i.e. a constriction at the cytoplasmic side, the internal pore, widening into a larger water-filled void, the internal cavity, which leads towards the narrow selectivity filter located at the periplasmic side of the channel [12]. In addition, the linearity of the cavity seems to give some insights on the pore selectivity to different types of solutes (Table 1). In fact, 10 of the 13 channels for inorganic ions in the set showed a very regular cavity, with average percentage of co-linear pore centres of 91.9% (SD = 7.0%) and organic small molecule/ion channels had less regular pores, with percentage of co-linear centres lower than 60%.
For completeness, PoreWalker output was also compared with results obtained using HOLE and MolAxis on the same set of structures. A systematic comparison with MOLE results could not be performed because, probably due to the intrinsic looseness of some structures, like the MthK and the ASIC1 channels, many of the tunnels identified by MOLE lie parallel and not perpendicular to the membrane axis and could not be considered as transmembrane. Within the set of pore features produced by PoreWalker and HOLE, the only comparable quantitative measure is the diameter, calculated along the pore at given heights. Diameter profiles obtained at 1Å steps for the 19 transmembrane proteins in the set were compared using the R2 correlation coefficient (see Supplementary Text S1, Table 1, and Figures 6, S3, S4 and S5). Pore diameter analyses performed with the two methods showed good agreement for 12 of the 19 diameter profiles, with R2 higher than 0.75. However, the remaining 7 profiles showed very poor correlation coefficients, with R2 very close or equal to zero. This behaviour seemed to be strongly affected by the regularity of the cavity. In fact, R2 values showed a good correlation with the number of co-linear pore centres (Supplementary Figure S5) with a R2 of 0.70 and only one strong outlier, the sodium-potassium channel (PDB code 2ahy). The disagreement between the two profiles in this case was due to a completely different pore exit at the top channel side identified by HOLE that seems visually incorrect and makes the diameter trend in that area very peculiar.
As for MolAxis, the program does not calculate diameter values at given heights along the channel axis but provide a partial list of the amino acids that contribute to the pore surface. Therefore, minimum diameters and pore lining residues were used to compare PoreWalker and MolAxis results. MolAxis could not identify a channel for 9 of the 19 test protein structures (Table 1), the water, glycerol and ammonia channels and three potassium channels. For the remaining 10 proteins, minimum diameter values derived from the two methods gave poor correlation (R2 = 0.46). The exclusion of the SecYE-beta translocon, incorrectly characterised by PoreWalker, lead to an R2 of 0.69 (corresponding MolAxis-HOLE R2 were 0.60 and 0.57, respectively). Minimum diameters calculated by HOLE and PoreWalker gave a better correlation, with R2 of 0.54 and 0.90, respectively (the overall R2 on the 19 structure set was 0.67). In term of pore-lining residues, MolAxis provides a list of the amino acids responsible for the calculated diameters, i.e. a subset of the amino acids that make the surface pore. MolAxis pore-lining residues were fully included in PoreWalker pore-lining residue list in all the compared proteins but the SecYE-beta translocon. In this case, 23 of the 24 pore-lining residues detected by MolAxis were included in the list generated by PoreWalker, showing that the method can reliably identify amino acids which build a channel despite mis-placements of its pore vector.
Finally, transmembrane pores identified by PoreWalker were found to coincide well with molecules of solute found in the 3D structure. Figure 5C–F shows the SoPIP2;1 plant aquaporin (1z98) and the sodium-potassium channel (2ahy) filled with water molecules and sodium and calcium cations, respectively. In both cases the cavities generated by PoreWalker completely surround and include water molecules and ions, which provide good evidence for the location and shape of the pore. Interestingly, PoreWalker is also able to identify the two main choke points in the water channel of the SoPIP2;1 reported to be in a closed state -the canonical Ar/R constriction site near the top of the pore and a narrower restriction close to the bottom of the channel (Figure 5D). The method can therefore analyse and characterise both “open” and “closed” transmembrane protein channels and transmembrane transporters.
The KcsA potassium channel is a homotetrameric integral membrane protein with high sequence similarity to all the potassium channels, particularly in the pore region. Its channel includes three elements: 1) a narrow entrance, known as the internal pore, starting at the intracellular side of the membrane; 2) an internal cavity, about 10Å in diameter, at the middle of the membrane; 3) a further narrowing, the selectivity filter, which leads to the extracellular environment [30]. The KcsA channel is therefore a good target to assess the ability of PoreWalker to detect constrictions, gates and internal cavities in the 3D-structure of a channel protein.
The 3D structures of the Kcsa potassium channel in the presence of low (3 mM, Figure 7A) and high (200 mM, Figure 7C) K+ concentrations are available at the wwPDB (codes 1k4c [30] and 1bl8 [8], respectively) and their pore features were derived and analysed using PoreWalker (Figures 7–9). The diameter profile of the low-K+ channel (Figure 7B, solid line) shows that PoreWalker can neatly identify the three main features of the channel: first a ∼3Å narrowing corresponding to the internal pore, the internal ∼9.0Å bigger cavity and a second narrower (∼1Å) constriction corresponding to the selectivity filter, highlighted in the Figure in orange, blue and red, respectively. It is interesting to notice here that diameter values calculated at 1Å steps by both HOLE (dotted line) and PoreWalker (dashed line) at the maximum width of the internal cavity (∼4Å) were significantly smaller than those reported in the description of the 3D-structure [30] (∼10Å) and found using the standard PoreWalker protocol at 3Å steps (∼9Å).
The calculated diameters of the internal pore and cavity also strongly agree with the proposed mechanism of ion conductance through the pore. In fact, potassium cations are thought to move through the internal pore and cavity in a hydrated form and to be dehydrated at the selectivity filter. The internal pore detected by PoreWalker is ∼3Å in diameter and could allow through one water molecule per time (the average diameter of a water molecule is usually taken as 2.8Å). Therefore, K+ ion could move through it alternating with water molecules. On the other side, the selectivity filter has a predicted diameter of ∼1Å and could therefore let through only dehydrated K+ cations.
The comparison of the diameter profiles of the channel in presence of low and high quantity of ions (Figure 7D, solid and dotted line, respectively) showed that besides expected differences at the cytoplasmic side of the pore, where a gate mechanism is known to operate, the entrance of the selectivity filter is ∼2.5Å wider at high concentrations of K+. According to PoreWalker, the pore lining residues, which define access to the selectivity filter, are the Thr75s from the four chains making up the pore. The difference in pore diameters at this point seems mainly to be due to different Thr sidechain conformations (Figure 7E–F). A significant difference in the two conformations of the KcsA selectivity filter had been previously highlighted at the level of residues Val76 and Gly77. A deeper analysis of the whole selectivity filter (Figure 8A) showed that the periplasmic side of the filter (at the top of the Figure) varies very slightly, while a major change is hinged at Gly77 and extends through Val76 to Thr75, where a pincher-like shutting mechanism could reasonably be hypothesized (RMSDs of all-atom superpositions were 0.33Å, 0.58Å and 0.99Å for Gly77 (Figure 8B), Val76 (Figure 8C) and Thr75 (Figure 8D), respectively). Besides, the internal cavity accommodates K+ ions as hydrated by eight water molecules. The 3D-structure of the low-K+ channel cavity (Figure 9) shows that the four water molecules facing the filter are aligned to the sidechain oxygens of Thr75s and can make hydrogen bonds with them (inter-oxygen distances are 3.9Å). Moreover, their distances from the corresponding K+ ion are close to optimal (3.4Å versus 2.8Å [31]). Therefore, it might be reasonably thought that the pinching mechanism could be aimed at weakening the water-K+ hydration complex by increasing the distance between the water molecules and the ion to facilitate its way into the pore.
We developed PoreWalker, a novel web-available method for the detection and characterisation of channels in transmembrane proteins from their three-dimensional structure. PoreWalker is fully automated and very user-friendly, requiring as input only the 3D coordinates of a transmembrane protein structure. A key prerequisite of the submitted structure is the presence of a transmembrane helix bundle or beta-barrel creating the pore, which is needed for the geometrical identification of the main protein axis. If this condition is not met, the detection/description cannot be performed with the current version of the software.
In term of outputs, in addition to diameter profiles, PoreWalker describes several specific pore features, in particular the shape and the regularity of the channel cavity, the atoms and corresponding amino acids lining the pore wall, and the position of pore centres along the channel. These features can be very helpful to gain further insights into the functional and structural properties of transmembrane protein channels by triggering specific in silico or experimental analyses, as shown from the recent characterization of the bacterial TolC channel [32].
PoreWalker is based on the assumption that, in a transmembrane channel protein, the pore is made by the longest cavity crossing the protein along the main axis of its transmembrane portion and therefore detects the longest widest cavity in a transmembrane protein structure. However, there are cases, as in the Amt-B and the SecYE-beta translocon, where the longest widest cavity does not correspond to the most likely “true” channel and in such cases the method assigns incorrectly one or both the pore gates. Interestingly, for these examples, calculated optimal cavities partially overlapped with the “true” cavities and most of the pore-lining residues were anyway identified properly.
In summary, PoreWalker provides a robust and automated resource to interpret, coordinate data and derive quantitative descriptors, which help to provide a deeper understanding and classification of membrane protein structures.
|
10.1371/journal.pcbi.1002739 | Evolution of Associative Learning in Chemical Networks | Organisms that can learn about their environment and modify their behaviour appropriately during their lifetime are more likely to survive and reproduce than organisms that do not. While associative learning – the ability to detect correlated features of the environment – has been studied extensively in nervous systems, where the underlying mechanisms are reasonably well understood, mechanisms within single cells that could allow associative learning have received little attention. Here, using in silico evolution of chemical networks, we show that there exists a diversity of remarkably simple and plausible chemical solutions to the associative learning problem, the simplest of which uses only one core chemical reaction. We then asked to what extent a linear combination of chemical concentrations in the network could approximate the ideal Bayesian posterior of an environment given the stimulus history so far? This Bayesian analysis revealed the ‘memory traces’ of the chemical network. The implication of this paper is that there is little reason to believe that a lack of suitable phenotypic variation would prevent associative learning from evolving in cell signalling, metabolic, gene regulatory, or a mixture of these networks in cells.
| Whilst one may have believed that associative learning requires a nervous system, this paper shows that chemical networks can be evolved in silico to undertake a range of associative learning tasks with only a small number of reactions. The mechanisms are surprisingly simple. The networks can be analysed using Bayesian methods to identify the components of the network responsible for learning. The networks evolved were simpler in some ways to hand-designed synthetic biology networks for associative learning. The motifs may be looked for in biochemical networks and the hypothesis that they undertake associative learning, e.g. in single cells or during development may be legitimately entertained.
| Here we evolve chemical networks in simulation to undertake associative learning. We define learning as the process by which information about the world is encoded into internal state (a memory-trace) in order to behave more adaptively in the future. Associative learning is learning of a relation between two types of event. Remarkably, the most frequently found circuits consisted of only one or two core chemical reactions responsible for learning, the other reactions being involved in subsidiary functions such as signal transduction. This is functionally simpler than the previously hand-designed biochemical circuits for classical conditioning that require several chemical reactions to implement Hebbian learning (a term which we use to refer to a mechanism that ensures that event A co-occurring with event B results in a greater probability that event B will occur given future presentations of A alone [1], [2]). Thus, this is a beautiful example of how evolution can find elegant solutions.
Chemical kinetics is Turing complete and therefore any computable mechanism for associative learning is theoretically possible [3], [4], however, this says nothing about which kinds of chemical mechanisms for learning are likely to evolve. Here we use in silico natural selection [5], [6], [7], [8], [9] to evolve chemical networks that are selected on the basis of their ability to carry out various associative learning tasks. Also known as genetic algorithms [10] or evolutionary computation [11], [12], the principle follows that of selective breeding. An initial random population of chemical networks is constructed. Each network is assessed for its quality as defined by a ‘fitness function’ that maps quality to fitness. The next generation is produced by allowing networks to replicate with mutation (and crossover) in proportion to their fitness. This process iterates for many generations, eventually producing higher quality networks that are capable of solving the desired task. The closest work to ours is the evolution of associative learning in continuous recurrent neural networks [13].
Our simulation evolves an abstract chemistry; however unlike many experiments with purely artificial chemistries [14], [15] it was designed to respect conservation of mass and energy, an essential consideration for transferring the insights from in silico models to chemical reality [16], [17], [18], which is our ultimate goal. Each ‘molecule’ consists of ‘0’ and ‘1’ atoms, and only the number of digits (and not their sequence) determines the species' identity. Any interchange of building blocks between molecules was allowed to happen in reactions. With the exception of the implicit decay reactions, all the simulated chemical reactions are reversible., However, some reactions may be effectively irreversible because the reaction rate in the backward direction is very low compared to the reaction rate in the forward direction. For details of the artificial chemistry model refer to the Methods section. Results from a pilot study with simpler chemistry are described in Supporting Information TextS1, part 1.
Traditionally there are two types of associative learning, classical and instrumental conditioning, the former involves passive observation of events, e.g. associating the sound of a bell with the smell of food, and the later involves relating self-generated actions and their consequences, e.g. learning that pressing a lever produces food [19]. We developed tasks that evoke the classical conditioning paradigm in psychology [20]. The network receives input from the environment (in the form of chemical boluses externally introduced into the system) and produces output (defined as the concentration of a particular chemical species measured over a particular test-period). The chemical dynamics of the system (the changes in concentration of chemical species) describe the behaviour of the network according to its sensory input.
In all the learning tasks, the chemical network had to learn to anticipate the injection of a control chemical C, known as the unconditioned stimulus UCS in the classical conditioning literature. Anticipation of C means to act in a manner that shows knowledge that certain events can predict C. Anticipation can be learned or innate. In our tasks it is necessary to learn to anticipate, not just to evolve innate temporal expectations. All tasks involve two possible conditions. In one condition the network should be able to use another chemical S (stimulus pulse), i.e. the conditioned stimulus CS, that reliably precedes C to predict the occurrence of C. Prediction results in the production of an output chemical O - the conditioned response CR - immediately after S is presented but prior to C. If this condition has been properly inferred, output chemical O should then be reliably elicited by the stimulus pulse S alone, after pairing S with C. This describes the “associated” condition. In the other, “non-associated” condition, S cannot theoretically be used to predict C. We therefore no not wish to see a CR (i.e. no output O production) following S. Thus, in all cases the network's fitness depends on whether it has learned the association between S and C by requiring it to produce an output chemical after S only when it is reliably followed by C, but not otherwise. There is no explicit training and testing phase in our experiments. The network's task is to respond appropriately as quickly as possible.
Consider a possible real-world example of how such functionality may be adaptive. Imagine that C (UCS) is a toxin and that S (CS) is a chemical that in some environments (but not others) can predict that toxin. Imagine that a metabolically expensive anti-toxin O (CR) can be synthesised to neutralise the toxin C. Then it would be advantageous to use S to initiate the synthesis of anti-toxin O in lieu of C in the environments in which S was predictive, but not in those environments in which S was not predictive, where instead the no O should occur, i.e. no production of anti-toxin in response to S. All tasks pose variants of this fundamental problem. The fact the network may find itself in either environment within a lifetime means that it could not evolve the simple strategy of sensitization where it always produces output chemical O in response to S. We used five different tasks, designed to provide a systematically more challenging associative learning problem. A summary of the tasks, and the information required for achieving maximal fitness on them (i.e. the simplest discrimination that is sufficient for optimal performance), is given in Table 1. The first two tasks do not require detection of a temporal correlation between S and C, i.e. they can be solved without associative learning, i.e. by sensitization/habituation alone. They demonstrate that in restricted environments, information about associations between things can be equivalent to information about simpler (lower-order) environmental features, such as the frequency of individual event types. However, the later three tasks are designed such that they necessitate discriminations based on observation of associations, e.g. discriminating environments in which S and C are temporally correlated compared to environments in which they occur independently. Thus, the final three tasks are true associative learning tasks that cannot be solved without the capacity to observe associations and modify ones behaviour accordingly.
Classical conditioning involves a wide range of different training and testing regimes, e.g. Pavlovian conditioning [21], Blocking [22], Backwards Blocking [23], overshadowing [19], etc. Typically these paradigms show an unconditioned response to the control (UCS). Above we have used a set of training and testing regimes that do not explicitly require an unconditioned response (UCR) to the UCS (control) molecule alone. In other words, we have assumed that a straightforward chemical reaction exists, independent of the network modelled, that is capable of producing an UCR to the control molecule. An important aspect of classical conditioning is extinction, a reduction in the conditioned response CR when the conditioned stimulus CS (stimulus) is repeatedly presented in the absence of the unconditioned stimulus UCS (control). All the networks presented here show extinction, even though they were not explicitly evolved on an extinction paradigm, see Supporting Information TextS1 Part 2.
The times when the network must respond by producing an output O when stimulus S is associated with chemical C were constrained to regular “clock ticks” to make the task as easy as possible for the networks. Because there is no noise, this is a simple task as the very first input event (which is either S on its own, or S followed by C) provides all the necessary information for maximising fitness (Figure 1). The blue blobs show the time at which the target output is required, i.e. when the target output contributes to fitness. In the associated condition the target output is high (1) and in the unassociated condition the target output is low (0). At all other times it does not matter what the target output is. This was intended to give evolution more leeway by imposing fewer constraints. Even so, many evolved solutions maintained the output concentration at low levels when the target output was not evaluated.
This task is identical to task 1., except that stimulus-control pulse pairs occurred with a low (non-zero) frequency in the unassociated environment and stimulus pulses without control pulses occurred with a low (non-zero) frequency in the associated environment. This produced ambiguity about the hidden state (which environment the network is in) on the basis of observed state variables (S and C pulses). Here, high fitness networks must consider more of the past, since isolated input events are unreliable indicators of the correct output chemical response (Figure 2.). A successful chemical network should update its ‘belief’ in which environment it is in on the basis of several observed associations, not just one; in other words, it must integrate information over time. For example, if we examine Figure 2., we see that the second stimulus pulse is followed by a control pulse even in the unassociated condition, and that the second stimulus pulse is not followed by a control pulse in the associated condition. However, the reader will notice that ‘cheating’ is possible in these two tasks because in the associated condition C occurs more often in total than in the unassociated condition, thus simply learning to respond to S when C is on average higher in concentration is a sufficient strategy. The temporal relation between S and C does not need to be learned here. This simple solution is excluded in the design of the next task.
In this task the timing of stimulus pulse and control pulse input events was unconstrained, and, most importantly, the unassociated and the associated environments received the same number of control pulses, except that in the unassociatied environment they were randomly distributed while in the associated environment they reliably followed stimulus pulses. Therefore this task was harder still, since it involved detecting relational aspects of inputs rather than merely first-order statistics of control pulses like the first two tasks (Figure 3).
Like task 3., this task used unconstrained input timing with noise and required relations between inputs to be detected. The difference is that in the first environment, where the network was required to keep the output chemical concentration low, control pulses reliably preceded stimulus pulses (Figure 4) rather than the other way around. In both cases S and C are associated, but occur in a different temporal order. The network must distinguish between these two kinds of temporal relationship.
The previous tasks described classes of stochastically-generated environment. Hence, any one network could be evaluated only on a sample of the environments typical of the task. By contrast, this task was designed by hand to provide a significant challenge while allowing exhaustive evaluation. The networks performance was measured in four environments (all possible combinations of stimulus-control pulse pairs). Maximal fitness required accumulating relational data over multiple input events; the task was specifically designed to exclude strategies that rely on the first or most recent input event (Figure 5). Unlike the previous experiments, the network must learn 2 bits of information because it must distinguish one of 22 states (not just 21 states).
We were able to evolve highly fit networks for each of the tasks above. Dynamics of the best performing networks on the five different tasks are shown in Figures 6–10 (for details of the chemical networks see Supporting Information TextS1 Part 3). The networks display learning – changes in state which reflect the statistics of their past inputs, and determine their response to input boluses adaptively. Note that the network's performance typically increases over the evaluation period, suggesting that a long-term memory-trace builds up over consecutive stimulus-control pairs.
The differences in task difficulty can also be observed on the graphs. For the simplest, clocked, task one input event was enough for the network to decide about the environment; but for the AB-BA or the 2-bit task a much longer training period was required. Figure 6 shows the performance in the Clocked task. The output chemical O (molecule ‘01’) is shown in black. In the top (unassociated) condition after the first presentation of stimulus S and the absence of a control bolus, its concentration drops and never returns. In the associated environment below, the output chemical shows the opposite dynamics after the first paired input of S and C. Figure 7 shows the evolved performance of a network on the noisy clocked task. The output chemical is again shown in black, and again in the unassociated task its concentration gradually declines (except after a misleading S-C pair shown during the second input event). In the associated environment the black output chemical continues to be produced when the network is stimulated with S. Figure 8 shows performance on the non-clocked task where it is necessary to learn explicitly the temporal correlation between S and C because in both tasks the overall amount of S and C is the same. Again an evolved network is successful in this because the black output chemical is only produced in the associated condition below and not in the unassociated condition above. Figure 9 shows the performance of a network that successfully evolved to solve the AB-BA task. The concentration of the output chemical in the lower condition is higher on average than the output concentration in the upper condition. The performance was only assessed during the second half of the task and this is where the greatest difference in black chemical output is seen. Figure 10 shows successful performance on the 2-bit environment task with the black output chemical only showing high concentration in the third condition.
Having evolved approximately 10 networks capable of solving each task, we ask, how do they work? The evolutionary algorithm permitted increases or decreases in the number of chemical species and the number of chemical reactions, see Methods. The smallest evolved network required only two reactions, but the typical number of reactions in an evolved network was 12 (mean 11.9, median 12). A greedy pruning algorithm applied to the networks revealed that most of these reactions were superfluous; typically only 5 reactions (mean 4.7, median 5) were necessary to achieve a fitness score within 10% of the entire network's fitness. The numbers given are for all tasks in aggregate; statistics for individual tasks are not very different. Although we did not select explicitly for simplicity, smaller networks emerged in the simulations. Figure 11 below shows the core network motifs that were evolved for associative learning, identified after pruning.
The second motif (Figure 11A, bottom) is the most commonly evolved solution. It appeared as a solution to all the above tasks. We analyse that in detail below in a case where it evolved in the best network capable of solving the AB-BA task (the network is described in detail in Supporting Information TextS1 part 3). The task in this case is to produce output (species 11) when control pulses follow stimulus pulses (S→C), but not to produce output chemical O when control pulses precede stimulus pulses (C→S), see Figure 12.
In the S→C environment, a slowly decaying long-term memory chemical LTM (chemical species ‘001’) reacts with the stimulus S to produce output O and a fairly rapidly decaying short term memory chemical STM (0001). Thus, output is produced in response to the stimulus when the memory chemical is present:(1)When the control pulse C occurs, it converts the short-term memory chemical back into the long-term memory molecule, allowing the LTM molecule to be reused again in the next pulse pair.(2)Now consider how the same network must behave quite differently in the C→S condition. Here C occurs before S so there is no STM molecule for C to react with to produce LTM. This means there is no LTM molecule for S to react with to produce output. Instead C readily disintegrates:(3)and the disintegration product reacts with the output molecule thus removing any output that might be produced in response to the stimulus that follows:(4)Whilst reactions (3) and (4) are not shown in Figure 11, we note that such ‘extra’ reactions are typical additions to the core motifs that evolved. Each evolved network contains multiple such extra adaptive reactions that help in various ways to control the dynamics of the system.
This hypothesis for the mechanism of learning was tested by modifying the concentration of the long-term and short-term memory chemicals by manipulating their inflow and decay rates and observing the response to stimulus pulses. We found that, as expected, the LTM and STM molecules determined the magnitude of output produced (Figure 13). Remarkably, this explanation can be re-interpreted in the light of Bayesian posteriors, i.e. ‘beliefs’ that the network has about which environment it is likely to be in, according to the information provided so far by the environment. To do this, we interpreted the internal state of the network as encoding a Bayesian posterior, by fitting a regression model from the chemical concentrations of the network at each point in time to the ideal Bayesian posterior of being in the associated environment given the sensory history encountered so far. If it is possible to fit such a regression model it means that a linear combination of chemical species concentrations encodes in a sense a near-optimal ‘belief’ about which environment the network is in. We found it was indeed possible to fit such a linear model for the above network, see Table 2. Furthermore, the parameters of this model must correspond to each species' role in learning. Positive numbers signify chemical species that are typical for the S→C environment, while negative numbers indicate that these chemicals are more abundant in the C→S environment. As expected, the largest positive posteriors belong to the memory chemicals, and, of course, to the output chemical (reactions 1–2); while large negative numbers indicate the disintegration product and the waste chemical (reactions 3–4).
Many of the evolved networks used the motif described above. There were a few more general features that repeatedly appeared for all tasks. For example, the input (stimulus, control) and output chemicals' concentration typically decreased quickly, either by spontaneous decay or by reactions that converted them to waste products/memory chemicals. A long-term memory chemical could be identified in most networks: this reacted with the stimulus to produce output, and was generated only in the S→C environment.
Apart from these features, the chemical background of learning was diverse and highly specific to the task in question. In the clocked and noisy clocked tasks only the S→C environment contained control pulses, and this was habitually exploited by converting the control directly to the long-term memory chemical (network not shown in Figure 11). In the non-clocked task, many of the networks used the fact that the output needs to be low after the control arrives. The signal in itself was converted to output, while control removed output. This resulted in a dynamics where in the S→C environment, control removed the output of the previous signal; in the case of randomly distributed control pulses, there was no output available when control was added, so, it inhibited the output of the following signal. The AB-BA task was a very special problem and the networks evolved to solve it were even more diverse than usual. In several cases the control was used to inhibit output production, as in the C→S environment it reliably preceded the signal. As the 2-bit task included more environments, it was more difficult for the networks to use “tricks”, and they mostly used the mechanisms depicted on Figure 11. We have evolved a few networks to be able to solve all tasks and the tendency towards simplicity was even clearer in them: they invariably used the most typical mechanism (Figure 11A, bottom) that we have analysed above.
Bayesian statistics provides a valuable framework, not just for statistical analysis of data, but for conceptualising how physical systems can encode models of their environment and update those models. The central concept in Bayesian statistics is that a “belief” can be modelled as a probability distribution; the rational way to modify the belief in response to evidence can then be formally codified. In order to incorporate cumulative evidence rationally into a model of the environment, it is sufficient to apply Bayes' rule repeatedly over time, with the posterior probability after each observation becoming the prior probability for the next observation, see [24] for an overview. This process is known as iterated (or recursive) Bayesian inference.
The typical application of Bayesian statistics would (in effect) be for the experimenter to apply Bayesian inference to their own beliefs, beginning with some probabilistic belief about the system and refining it by the observation of evidence. We turn this on its head by considering, if the system itself were a rational observer, what “beliefs” it should have regarding its environment and how it should update them in response to evidence. A similar approach to ours can be seen in [25]. We found that a Bayesian analysis provides insight into understanding network function. Note that there was no explicit pressure on the networks to perform Bayesian reasoning. However, achieving a high fitness during evolution required the networks to incorporate and integrate information over time. Iterated Bayesian inference is the formal ideal of the process of integrating cumulative evidence; hence, we have a theoretical motivation for interpreting the network dynamics in Bayesian terms.
We attributed “beliefs” to the networks by analytically deriving the Bayesian beliefs (posteriors) of an ideal observer in a given task (over a variety of time steps and environments), and fitting a regression model from the network's state to this ideal belief. (We use a logistic regression model as the natural analogue of a linear model for a range bounded between 0 and 1.) Hence, we determined the maximum extent to which the network's state can be said to encode the correct posterior in a simple form. For comparison purposes, we also performed this procedure on networks that were not evolved on the task in question. This means that the “belief” attributed to a network depended on the task it was being observed on: “belief” in this context really means “most generous attribution of belief given the task”.
The mean correlation between the fitted logistic regression model and the analytic posteriors is extremely high (0.97–0.98) for the highest-fitness evolved networks on both the noisy clocked association task and the AB-BA task (Figure 14). The information required to perform the noisy clocked task is relatively easy to accumulate in a detectable form: for random networks, the mean model/posterior correlation is fairly high (0.82). For the AB-BA task, which requires accumulating more subtle information, the quality of fit of the regression model for random networks is very low (0.06) (Figure 14). Figure 15 shows the dynamics of an evolved network's best “belief” (the output of the regression model) over time for a particular lifetime, compared to the ideal rational belief (the posterior probability). Interestingly, the network evolved on the “2-bit environment” task demonstrated information capture on both the noisy clocked task (rho = 0.83) and the AB-BA task (rho = 0.73). See Supporting Information TextS1, part 4 for other example networks, including networks that were evolved on a different task to the one they are being tested on.
The process of Bayesian inference is characterised by the incorporation of relevant information into a system's internal state. This does not constrain the way in which a Bayesian posterior is encoded into the state of a system; the encoding in principle could be arbitrarily complex. However, our empirical results for the evolved networks indicate that the existence of an encoding can be demonstrated by a simple regression model.
It is worth observing that just because a system's state contains the relevant information to perform a task, this does not necessarily mean that the system uses that information appropriately. For our noisy clocked task, the dynamics of a randomly constituted network usually encode the relevant information for task performance in a nearly linear way, whereas random networks have a poor fitness performance on the task. This is because in the artificial environment for that task, the overall rate of control pulses differs in the two different experimental conditions. To a first approximation, we can regard the two experimental conditions as providing constant driving inputs to the system, but at different rates. Hence, if a system's gross dynamics depend on the rates of control pulse inputs (which will be true for the majority of systems), then observing the system's state after interacting with one or other of our task environments will readily reveal which environment the system was exposed to. We will see below that this issue does not apply to the more complex AB-BA task that requires genuine sensitivity to stimulus pairing (see Table 1 for a comparison of the informational requirements in the noisy clocked task and the AB-BA tasks).
There are important parallels here to liquid state machines [26], [27] and other reservoir machines [28] and to random projections [29] in machine learning: information capture is not necessarily the hardest part of information processing, and randomly constituted systems can often accumulate information in a usable fashion. So, the random networks store information about the rate of control pulses in the environment (although not as much information as a network evolved for the task). That information can be extracted by an observer using a simple regression model, similar to reservoir machine and random projection learning. However, the random networks do not incorporate the machinery to translate the stored information into an appropriate response: a high output following a stimulus pulse when control pulses have occurred at a high rate in the past, and a low output otherwise.
By contrast, we determine empirically that the AB-BA task produces very different information dynamics to the noisy clocked task. In the AB-BA task, the overall rate of control (and stimulus) pulses is identical in the two different task environments. While random networks can be assigned a logistic-model Bayesian interpretation for the first task (i.e. a regression model can be fitted to map from the network state to the current optimal Bayesian posterior), the same is not true for the AB-BA task (see Supporting Information TextS1, part 4), where only the evolved networks have a good logistic-model Bayesian interpretation. Note that to distinguish the AB-BA environments, the network must respond differently to a C pulse followed shortly by a S pulse than a S pulse followed by a C pulse. The information necessary to distinguish the environments optimally is the relative number of CS versus SC pulses.
A nervous system is not necessary for learning. We have shown that associative learning mechanisms implemented by well-mixed chemical reactions can be discovered by simulated evolution. What differences in principle, then, are there between neurons and chemicals? The key difference between learning in neuronal network and learning in our chemical networks is that in neuronal systems generic learning mechanisms exist that are present at each synapse, irrespective of the particular identity of the pre- and post-synaptic neurons. For example, spike-time-dependent plasticity (STDP) can be found between many neurons. This is possible because neurons share the same genome, and this permits each neuron to express the molecular machinery required for plasticity. On top of this, specificity can be achieved through line labelling, i.e. it is the physical pathway from stimulus to neuron A to neuron B etc. that has meaning, and conveys reference. The capacity to associate arbitrary events X and Y arises when a plastic synapse exists between neurons that represent X and neurons that represent Y.
In our chemical networks, however, there is no modular distinction between chemical species that represent events and the chemical reactions that implement learning. The chemical network for associating X and Y by forming memory-trace M cannot work separately to associate P and Q because of two reasons: (i) the reactor is well mixed and the memory-trace M for X and Y will interfere with the memory-trace M for P and Q (ii) the molecule M will react with X and Y but it cannot without modification react with arbitrary P and Q. In the neural system neither of these constraints exists.
This has important consequences on the scaling properties of neural or chemical systems for associative learning. Suppose that the system needs to be able to learn three independent possible associations (say, A→C, B→C and A→D). The weight (strength) of each association needs to be represented independently in the network, and an associative mechanism implemented to update each weight.
In the neural system this is easy; the associative mechanism is a set of molecules that are expressed in each synapse that implements Hebb's rule or some variant of that rule, which states that events that co-occur have a higher probability of co-occurring in the future. In neuronal systems the weights of the associations are the synaptic strengths. Each neural connection contains the molecular capacity to implement Hebb's rule specifically between distinct neurons. In the chemical system, however, each associative mechanism will be a different chemical pathway, and the pathways will need to be functionally similar while involving species whose chemical properties are distinct (since if the species are too similar, there will be crosstalk between the pathways). In essence, it seems plausible that the chemical system will have to re-implement associative learning independently for every possible association.
We have described chemical networks in this paper that can learn to associate one stimulus with another stimulus. An important qualifier here is that they do not display generic associative learning: the two stimuli that can be associated are genetically specified. Of course, more sophisticated cellular systems such as genetic regulatory networks may be able to overcome the problems we have described. Also, the learning is not independent of timing, but instead the ability of an evolved network to undertake associative learning is greatest for environments where the period between successive stimulus-control pairs resembles that period encountered during evolution, see Supporting Information TextS1, part 5).
We used in silico evolution to find small chemical networks capable of carrying out various associative learning tasks. It is often the case that evolution finds solutions that are much more concise, elegant, and parsimonious than would be produced by deliberative cognition. In fact, the simplest chemical network still capable of associative learning consisted of only two chemical reactions. This confirms that there is no reason in principle that associative learning within a lifetime should be confined to multicellular organisms.
So why is the experimental evidence of associative learning in single cells to date equivocal? We are only aware of one experiment that addressed this question [30]. Todd Hennessey showed that aversive classical conditioning occurs in Paramecia. He trained a single paramecium to avoid an electric shock by learning that vibration precedes it. The mechanisms underlying such learning are not known, although it seems possible that voltage gated Calcium channels [31], [32] are involved, perhaps with adenylate cyclase acting as a coincidence detector with cAMP dependent state changes mediating memory as in Aplysia [33], [34]. Similar studies have indicated that other single-celled organisms may have the capacity to learn to associate light and electric shocks [35], [36] although a recent study on individual human immune cells showed habituation but no conditioning [37]. Notice that the task of learning a contingency within a lifetime is entirely different from evolving to respond under an evolutionary regularity that B will regularly follow A in all environments. Whilst there was a recent report that such behaviour is observed in bacteria which anticipate the decrease in oxygen following increase in temperature, these bacteria did not learn to anticipate but rather they evolved to anticipate [38]. Often this critical distinction is not made, resulting in confusion between evolution and learning [39]. To see the difference, note that no bacterium in the above experiment could learn within a lifetime that in some environments increased temperature predicts increased oxygen, whereas in other environments decreased temperature predicts increased oxygen. This association was not learned by a single bacterium, instead, it is an association that was discovered by evolutionary search by populations of bacteria. The very ease with which populations of bacteria and yeast can evolve to anticipate environmental changes in laboratory evolution experiments suggests that it may simply not have been necessary for individual single celled organisms to learn to anticipate within a lifetime [40].
An important implication of our work is that the associative mechanisms we have described may be active during development in cells within a multicellular organism. It will be of interest to use bioinformatics to examine whether the motifs in Figure 11 can be found in regulatory networks involved in development. This paper allows us to re-examine the possible function of simple chemical motifs within an associative learning framework.
In order to enforce conservation of atomic mass in the networks' reactions, we used a combinatorial abstract chemistry for the networks. Each simulated chemical species had a “formula” consisting of a string of digits representing chemical “building blocks”, and reactions were constrained to conserve building blocks. These constraints were modelled using three different abstract combinatorial chemistries: An “aggregate” chemistry, where only the number of digits (and not their sequence) determined the species' identity, somewhat resembling inorganic chemistry with atoms as building blocks. Any interchange of building blocks was allowed to happen in reactions. A “rearrangement” chemistry, where the sequence of digits characterized species, somewhat resembling organic chemistry with atomic groups as building blocks. Any interchange of building blocks was allowed to happen in reactions. A “polymer” chemistry, where only ligation and cleavage reactions could happen among chemical species, resembling polymer reactions with monomers as building blocks.
Simulations of a simple aggregation chemistry provided chemical networks with the highest fitness (Supporting Information TextS1, part 6), thus, results in the main text refer to this particular chemistry. Reactions were modelled reversibly. We incorporated further thermodynamic constraints by assigning “free energy” values to chemical species; these constrained the ratios between forward and reverse reaction rates. Each network received an inflow of one particular chemical type (“food”), and every chemical species exhibited first-order decay, as expected in a flow reactor scenario. Note that although all the parameters of the reaction networks – chemical species, chemical reactions, free energy values, inflow rate of food and species decay rates – were allowed to change during the evolutionary runs, each individual network had its own fixed chemistry that stayed the same during the learning trials. Therefore the difference between chemical networks in the unassociated and associated environments could only be induced by the different history of input boluses; these must have modified the state of the network (the concentration of different chemicals) so that it showed different behaviour when presented with the stimulus chemical.
Networks consisted of a number of chemicals and reactions, the relevant characteristics of which were encoded genetically. See Figure 16 for illustration.
Each abstract chemical species was associated with a number of real-valued parameters: A chemical “potential”, which affected the thermodynamics of the system, an initial concentration, a spontaneous decay rate (conceptualised as decay to inert waste products), an inflow rate if this species was chosen as the network “food” (see below). In addition, chemical species were assigned a binary “formula” string, which constrained how different species could combine (see “chemistry” section).
Reactions were represented as a list of one or two “Left Hand Side” (LHS) species, a list of one or two “Right Hand Side” (RHS) species, and a real- valued “favoured rate constant” (see below). The variation operators used in evolution guaranteed that reactions conserved mass and compositional elements (see below). Note that the intrinsically favoured direction for the reaction was not determined by the reaction's encoding but by the chemical potential values of the species involved. The “favoured rate constant” parameter of the reaction determined the rate constant in the favoured direction; the rate constant in the non-favoured direction was determined by the chemical potential values of the species involved.
The choice of which chemical species the network used as input, output and “food” were under evolutionary control. Part of the network encoding was an ordered list of species: the first species in the list functioned as inputs; the next species as output; the next species as “food”; and the remainder had no special environmental significance, see Table 3.
Network mutations were implemented as follows, based on a mutation rate sigma: All real-valued parameters were mutated by Gaussian noise, with reflection at the upper and lower parameter limits. The standard deviation of the noise was scaled by the product of sigma with the absolute size of the allowable range for that parameter. With probability sigma * 5, the program attempted to add a random new reaction to the network (see “adding new reactions”). With probability sigma * 5, a uniformly chosen reaction was deleted from the network. With probability sigma, two elements of the input-output list for the network were randomly swapped (most of the time, this involved swapping “non-special” elements and had no functional effect).
When a mutation called for adding a new reaction to the network, one of the following three possibilities was chosen uniformly:
In each case, the existing species were chosen uniformly and formulas for the reaction products were generated according to the current chemistry (see “chemistries”). If a formula was generated in this way that did not match a species already in the network, a new species was generated with that formula and added to the network. When a new reaction was added to the network, its “favoured rate constant” parameter was initialised to a low value (uniformly in the range [0, 0.1]) to allow for relatively neutral structural mutations.
Each chemical species in a reaction network was given a binary string “formula” which constrained what products it could form with other species. Reactions were always constrained so that the total number of 0s on the reaction LHS was the same as the total number of 0s on the RHS, and similarly for 1s. In addition, we modelled three different string “chemistries”, each with different compositional rules, see Table 4.
Networks were initialised as follows. A small number of “seed” chemicals (by default, 4) with distinct formulas of length 3 were added to the network. New chemical species, whether generated at initialisation or due to adding a new reaction to the network during initialisation or mutation, were initialised with uniformly random parameters in the following ranges: potential [0–7.5], initial concentration [0–2], food inflow [0–1], decay [0–1]. The function to add a new reaction was called 20 times, thereby adding an unpredictable number of new chemicals to the network. New reactions, whether generated during initialisation or mutation, were initialised with a uniformly random “favoured reaction constant” in the range [0–0.1]. The input-output list for the network was shuffled fairly.
The networks were evolved using a non-generational genetic algorithm (GA) similar to the Microbial GA [41]. A genetic algorithm is the natural selection algorithm run in a computer [10], specifically it is artificial selection in which an explicit fitness function (phenotypic target) is defined, rather than allowing fitness to emerge as the result of ecological interactions. In our case the fitness function rewards chemical networks capable of the kind of associative learning we require. The basic algorithm was as follows:
All reactions were modelled using reversible deterministic mass action kinetics (apart from the implicit decay reactions which are irreversible). It is clearest to explain this scheme by example.
A single reversible reaction can be conceptually split into two parts, so that
is conceptually equivalent to the composition of two reactions
and
The rate at which a reaction takes place, in our simulation, is set equal to the product of the concentrations of those species on its left-hand side, multiplied by its rate constant. The reaction consumes its reactants at this rate and generates its products at this rate. The overall rate of change of a species' concentration due to explicitly-modelled reactions is equal to the sum of the rates at which it is generated (over all reactions) minus the sum of the rates at which it is consumed (over all reactions). Spontaneous decay (at a rate for chemical X) contributes an additional term to this sum, and inflow (at rate for the food chemical F) contributes an additional term to F's rate of change. Hence, a system consisting only of the reaction(with A as the food chemical) has the following differential equations:For computational efficiency, simulations during evolution used Euler integration with a step size of 0.01 time units. Input boluses were modelled as discontinuous jumps in concentration at the given time steps. These simulations were qualitatively validated after evolution using a variable step-size Runge-Kutta ODE solver.
Networks were simulated on chemical protocols, with each protocol consisting of a time series of input boluses, and a time series of target values for the network output. Note that for most time steps, the input bolus values were zero and the target output values were “don't care”. The exact details of the protocol inputs and targets varied from task to task.
For every task, networks were simulated on a number of protocols, and the (instantaneous) concentration of the designated network output chemical compared to the protocol target for every time step. The fitness of a network was set equal to the negative mean square difference between these two quantities averaged over all protocols and all time steps (ignoring time steps where a “don't care” target was specified). In order to provide a reliable fitness comparison, when two networks were chosen for competition during evolution, they were evaluated on the same set of protocols. Additionally, the protocols for different experimental conditions within the same task were deliberately matched to be similar, so that network response to the experimental condition could be measured as directly as possible.
Initial experiments indicated that randomly generating protocols during evolution results in very noisy fitness comparisons, with little fitness gradient for evolution to climb. To avoid this problem, for each task we generated fixed “training data” and saved it to file. Networks were evaluated during evolution on their performance on the training data set. For most tasks, the training data set was a file consisting of 10 randomly generated protocols. A number of tasks were devised requiring the detection of different environmental features by the networks. Some of these tasks were “clocked”, i.e. pulses were constrained to only occur at predetermined regular “clock tick” times, and some were not.
This task constrained B boluses to a regular “clock tick” schedule every 100 time steps and had two experimental conditions. There was only a 0.5 probability of a chemical B bolus on a given clock tick. In the “associated” condition, a chemical B bolus was always followed 20 time steps later by a chemical A bolus. In the “unassociated” condition, chemical A boluses never occurred. A single protocol featured both experimental conditions, with identical B boluses in each condition. The desired behaviour for the network was: upon receiving a pulse of chemical B, output either zero (in the “unassociated” condition) or one (in the “associated” condition) for 20 time steps afterwards.
This was identical to the previously described task except that there was a small (p = 0.1) probability of “noise” occurring at each time step with a chemical B bolus. Noise consisted of a B bolus being followed by an A bolus in the “unassociated” condition or a B bolus followed by no A bolus in the “associated” condition. Within a single protocol, the occurrence of noise was matched between experimental conditions.
This task had two experimental conditions and involved boluses at random intervals. In both conditions, pulses of chemical B occurred at random intervals uniformly in the range [100, 300]. In the first (“associated”) condition, a pulse of chemical B was followed shortly afterwards (20 time steps) by a pulse of chemical A. In the second (“unassociated”) condition, pulses of chemical A occurred independently of B, at random intervals uniformly in the range [100, 300]. Within a single protocol, pulses of chemical B were identical.
This task, featuring two experimental conditions, was specifically designed to involve a non-trivial accumulation of information. Within this task, input “events” occurred randomly at a low rate (0.025 per time step) with a refractory period of 50 time steps between events, over a total period of 2000 time steps. Each event consisted of either a pulse of chemical A followed closely (20 time steps later) by a pulse of chemical B, or vice versa. In the first experimental condition (“A→B”), events were 75% likely to be “A→B” pulses and 25% likely to be “B→A” pulses, and vice versa for the second (“B→A”) experimental condition. The desired output behaviour was to respond to a “B” pulse with a low output in “A→B” environments and a high output in “B→A” environments. Note that this task was both noisier than the other tasks and involved a longer evaluation period (to allow the noise some time to average out).
Unlike the other tasks, every environment in this task was designed by hand. The intention was to construct a range of radically different environments such that both short- and medium- term network memory-traces would be required to attain maximum fitness. The inspiration was loosely drawn from the concept of the “radical envelope of noise” [Jakobi, 1998]. Input pulses (boluses) in this task always occurred in closely-separated pairs, although the second bolus in a pair did not have to contain the same chemical as the first bolus. The pulse pairs occurred at regular intervals of 100 time units each. Each experimental condition was characterised by a “typical” pulse pair (A→A, A→B, B→A or B→B). In addition to the “typical” pulse pair corresponding to the experimental condition, every protocol for this task also had a “noise” pulse pair. There were in total 4 protocols (one for each pulse pair type), each containing 4 experimental conditions, for a total of 16 different input series. A single input series had the following structure: First, a pulse pair corresponding to the protocol's “noise” pair. Next, three “signal” pulse pairs all of the “typical” type for that experimental condition. Next, a “probe” pulse pair (see below). Next, another “noise” pulse pair of the protocol's “noise” type. Last, a final “probe” pulse pair. “Probe” pulse pairs consisted of a pulse of “B” chemical followed by either a pulse of “A” chemical (in the B→A environment) or a pulse of “B” chemical (in other environments). The desired network behaviour was to produce a low output for 10 time steps prior to each “probe” pulse pair, followed by either a high output (in the B→A environment) or a low output (in other environments) for 20 time steps. Errors in the B→A environment were weighted three times as heavily as errors in the three other environments.
We calculate the number of reactions per effective chemical species in a network by first excluding any species which do not take part in reactions (this is possible if all reactions featuring a particular species are lost from a network by structural mutation). We then simply calculate the mean number of distinct reactions each remaining species is involved in.
To investigate the effects of different genetic encoding factors on network connection density, we conducted 10 evolutionary runs on the 2-bit environment problem in each of 4 encoding variations. These were:
For all these runs, we recorded the effect of every mutation on both fitness and also the number of reactions per chemical species.
Our method is as follows. We imagine an ideal Bayesian reasoner, equipped with knowledge of the statistics of the different network task environments. For each input train, at each point in time, we calculate what subjective probability the reasoner should assign to the possibility that the input train up to that point came from an “associated” environment. This establishes what the ideal Bayesian posterior would be at each point in time for each input train. If a network's chemical concentrations somehow encode this time-varying Bayesian posterior in all environments, then it would seem reasonable to attribute a Bayesian interpretation to the network. For the purposes of this paper, we will skirt over the complexities introduced by the non-dissipation of information in smooth continuous dynamical systems. In principle, the state of our simulated networks will usually contain all information about their historical inputs, because information can be stored in arbitrarily small differences in concentrations. However, in practice this information will be destroyed by noise.
Calculation of the ideal posteriors for our environments is straightforward. A random variable X will represent the type of environment: either 1 (“associated”) or 0 (“unassociated”). Another random variable Y(t) will represent the train of input boluses up to time t. The ideal posterior probability of being in the “associated” environment after observing an input train y iswhereThe prior P(X = 1) was set equal to the proportion of “associated” environments in the network training sets, i.e. 0.5. The probabilities P(Y(t) = y|X) were calculated as follows. In the environments we analysed in this way, Input trains were organised into “events”: a bolus of one or other input chemical, followed possibly at a set short interval by another bolus. The time between input events always exceeded the inter-spike interval within an event. The timing of events in an input train provided no information about the type of environment. Hence, events can be extracted from an input train and treated as a discrete process. “Associated” and “unassociated” environments correspond to Bernoulli processes with “associated” (B→A) and “unassociated” (B alone, for the noisy clocked task, or A→B for the AB-BA task) events. Thus, the posterior P(X|Y(t)) can be calculated by counting the total number of associated and unassociated events in Y(t).where n and m are the number of “associated” and “unassociated” events in Y(t), and p1 and p0 are the probabilities of an “associated” event in the “associated” and “unassociated” environments respectively. This giveswhich can be attached to a time series at the appropriate points after events have occurred.
We use a straightforward logistic regression model to match network concentrations to Bayesian posteriors. Given a concentration vector x, a weight vector w and a bias value b, the model iswhere is the scalar product of x and w. Note that the output of this function is bounded between 0 and 1 like a probability value (this would not be the case for a linear regression model). The idea is that we can investigate the degree to which a rational Bayesian belief is encoded transparently in the network's state. We expect that at time t, having observed an input history Y(t),To determine appropriate model parameters, we randomly generate 200 environments (100 “associated” and 100 “unassociated”), and run the evolved network in those environments. Weights w and bias b are set to minimise mean square error over all environments and time steps, using Levenburg-Marquadt optimisation. No attempt was made to regularise the parameters or otherwise avoid overfitting, since the model has relatively few parameters. For comparison, 200 random networks (produced by random initialisation followed by 200 mutations) were tested in the same way.
|
10.1371/journal.pntd.0003269 | Dengue Virus Infections among Haitian and Expatriate Non-governmental Organization Workers — Léogane and Port-au-Prince, Haiti, 2012 | In October 2012, the Haitian Ministry of Health and the US CDC were notified of 25 recent dengue cases, confirmed by rapid diagnostic tests (RDTs), among non-governmental organization (NGO) workers. We conducted a serosurvey among NGO workers in Léogane and Port-au-Prince to determine the extent of and risk factors for dengue virus infection. Of the total 776 staff from targeted NGOs in Léogane and Port-au-Prince, 173 (22%; 52 expatriates and 121 Haitians) participated. Anti-dengue virus (DENV) IgM antibody was detected in 8 (15%) expatriates and 9 (7%) Haitians, and DENV non-structural protein 1 in one expatriate. Anti-DENV IgG antibody was detected in 162 (94%) participants (79% of expatriates; 100% of Haitians), and confirmed by microneutralization testing as DENV-specific in 17/34 (50%) expatriates and 42/42 (100%) Haitians. Of 254 pupae collected from 68 containers, 65% were Aedes aegypti; 27% were Ae. albopictus. Few NGO workers reported undertaking mosquito-avoidance action. Our findings underscore the risk of dengue in expatriate workers in Haiti and Haitians themselves.
| Dengue is the most common mosquito-borne viral disease in the world, and caused an estimated 390 million infections and 96 million cases in the tropics and subtropics in 2010. Over the last decade, the number of cases of dengue and the severity of dengue virus infections have increased in the Americas, including the Caribbean, yet little is still known about dengue in Haiti. Following an outbreak of dengue in mostly expatriate NGO workers, the investigators of this study took blood samples from expatriate and Haitian NGO workers living in two cities in Haiti and tested them for evidence of current, recent, and past dengue virus infection. They also investigated the amount and kinds of mosquitoes at homes and work sites. The study found recent infections among some Haitians and expatriates and widespread past infections among all Haitians and most expatriates. It also found that many people were not doing basic things to avoid mosquito bites, like applying mosquito repellent multiple times a day and wearing long sleeves or pants. These findings highlight the likely endemicity of dengue virus in Haiti, and the need to improve knowledge and awareness of dengue prevention among expatriates visiting Haiti and local Haitians.
| Dengue is the most common mosquito-borne viral disease in the world, and resulted in an estimated 390 million infections and 96 million symptomatic cases throughout the tropics and subtropics in 2010 [1], [2]. Over the last decade, the incidence and the severity of dengue have increased in the Americas, including the Caribbean [3], [4], where the four dengue virus-types (DENV-1–4) that cause dengue and the mosquitoes (i.e., Aedes aegypti and Ae. albopictus) that transmit DENV are endemic [1], [5]–[7]. The risk of acquiring dengue can be greatly reduced by following key mosquito avoidance activities, such as applying mosquito repellent multiple times a day and wearing long sleeves, pants or permethrin-treated clothing [8], [9].
Despite an absence of routine systematic surveillance data, dengue is likely endemic in Haiti, as it is in the Dominican Republic, which shares the island of Hispaniola with Haiti. Both Ae. aegypti and Ae. albopictus have been detected in Haiti, as have all four dengue virus-types [10], [11]. A 2007 study in Port-au-Prince showed that 65% of children <5 years of age had evidence of prior infection with a DENV [12], and a two-year prospective study in an outpatient clinic in Léogane found that 2% of patients presenting with undifferentiated fever tested positive for DENV infection by a rapid diagnostic test (RDT) [13]. Similarly, of 885 patients with acute febrile illness who were admitted to four hospitals in Haiti during 2012–2013, 4% tested positive for DENV infection by RDT [14].
Although dengue has been documented in US military personnel and expatriate relief workers in Haiti in the past two decades [15]–[19], visitors often do not regularly employ mosquito avoidance practices. In a survey conducted among American missionaries returning from Haiti in 2010, only 24% reported using mosquito repellent multiple times a day [15], and in a 1997 study of US military personnel in Haiti only 18% of febrile patients reported always using mosquito repellant [20].
In October 2012, the International Federation of Red Cross and Red Crescent (IFRC) and Red Cross-Haiti alerted the Haitian Ministry of Public Health and Sanitation (French acronym: MSPP) and the US Centers for Disease Control and Prevention (CDC) of 25 recent RDT-positive dengue cases among Haitian and expatriate staff of non-governmental organizations (NGOs) based mostly in Port-au-Prince and Léogane. Seven (28%) of the 25 cases were evacuated from Haiti for advanced medical care. To estimate the incidence of recent and previous DENV infection and identify demographic and behavioral risk factors for infection, we conducted a serologic survey among and administered a questionnaire to Haitian and expatriate NGO workers in Léogane and Port-au-Prince. Additionally, to better understand entomologic risk factors for human infection, we carried out an entomologic investigation around work sites and workers' residences.
Of 776 NGO workers (106 expatriates and 670 Haitians) in Léogane and Port-au-Prince, 181 (23%) participated in the investigation, including 52 expatriates and 129 Haitians. Of those, 173 (96%) provided a blood specimen for diagnostic testing. The majority of participants were male (76%) and Haitian (71%), and the median age was 33 years (Table 1). Most participants worked in administrative or office duties, construction, or community or field work.
Less than a quarter (21%) of expatriates reported being born in a dengue-endemic country. Nearly all expatriates (94%) but less than a quarter (23%) of Haitians reported ever living in (>1 month) or traveling to (>1 week) a dengue-endemic country other than Haiti in their lifetime. Nearly all expatriates (96%), but less than half (39%) of Haitians, reported ever hearing of dengue. Expatriates reported greater knowledge of DENV transmission (89% vs. 29%) and dengue prevention (96% vs. 13%) compared with Haitians. While 6% of expatriates reported a previous dengue diagnosis, no Haitians reported ever being diagnosed with dengue. Overall, 89% and 23% of expatriate staff reported receiving a yellow fever or Japanese encephalitis vaccination, respectively. In contrast, among Haitian staff, only 4% and 0% reported receiving a yellow fever vaccination or Japanese encephalitis vaccination, respectively.
The majority (87%) of expatriates and half (47%) of Haitians reported using mosquito repellent, but less than half (44%) of expatriates and only a small proportion (9%) of Haitians reported using mosquito repellent multiple times a day (Table 2). While most expatriates and Haitians reported using a bed net, only a small percentage of expatriates and Haitians (10% and 2%, respectively) reported using permethrin-treated clothing. Of the 52 expatriate workers, most (87%) said they had made a travel consultation prior to their current trip to Haiti. Of the 45 workers who reported making a travel consultation, approximately half (47%) went to a travel medicine clinic, 71% received mosquito-avoidance information during their consultation, and 39% received information about dengue.
Compared with their Haitian colleagues, more expatriates reported having screens on their windows or doors, and air conditioning at their sleep site. Expatriates also reported more standing water and trash near their work site, off-hours ‘hang-out’ places, and sleep site as compared to Haitians (Table 2).
DENV nucleic acid was not detected in any of the 173 NGO workers who provided blood specimens for dengue diagnostic testing. Both NS1 and anti-DENV IgM antibody were detected in one asymptomatic expatriate. Anti-DENV IgM antibody was detected in 17 (10%) NGO workers (8 [15%] expatriates and 9 [7%] Haitians) (Table 3). Of the 17 participants with evidence of current and/or recent DENV infection, six (35%) participants (five expatriates and one Haitian) reported being ill in the past 90 days, five (29%) reported missing at least one day of work, and three (18%) were hospitalized and subsequently required medical evacuation to the Dominican Republic. Of these three evacuated participants, two had dengue with warning signs: one had menorrhagia and the other had a pleural effusion.
Of 173 specimens tested, 161 (93%) had detectable anti-DENV IgG antibody, including 41 (79%) expatriates and all 121 Haitians. Prior DENV infection was confirmed by microneutralization assay in 17 (50%) of the 34 IgG-positive/IgM-negative specimens from expatriates, and in all 42 randomly selected specimens from the 121 IgG-positive Haitians (95% confidence interval [CI]: 94.5%–100%) (Table 4).
Participants who reported working near “open water sources” had greater odds of having had a current and/or recent DENV infection (odds ratio [OR] = 3.6, 95% CI = 1.3–10.1; Table 3). Participants who reported using mosquito repellent multiple times a day (OR = 3.5, 95% CI = 1.22–10.04), having very good knowledge of infectious disease in Haiti (OR = 3.6, 95% CI 1.16–10.98), and knowing how to prevent mosquito bites (OR = 6.2, 95% CI 1.92–19.72) had greater odds of having had a current and/or recent DENV infection. No other risk factors were found to be statistically significant (Table 5).
One hundred premises were surveyed, including 8 NGO work sites and 28 adjacent buildings, 8 NGO residences and 27 adjacent buildings, and 29 Haitian employee residences. In total, 2,664 containers were inspected for immature mosquitoes. Of these containers, 756 (28%) contained water, of which 198 (26%) contained immature mosquitoes. We collected 254 pupae from 68 water-holding containers; Ae. aegypti was the most abundant mosquito species identified (65%), followed by Ae. albopictus (27%). The remaining 8% of mosquitoes identified were Ae. mediovittatus, Culex species, or were unidentifiable. Vector indices were similar between NGO work sites and residences. All mosquito abundance indices were elevated. For all premises combined, the Premise index was 61%, the Container index was 26%, and the Breteau index was 198 (Table 6). Pupae were found in 46% of tires, 29% of cans, 28% of water drums, 21% of cisterns, and 20% plastic containers that held water.
In our investigation of NGO workers in Léogane and Port-au-Prince, Haiti, we found that a substantial proportion (15% of expatriates and 7% of Haitians) had recently been infected with a DENV. Six of the infected workers reported being ill, and three required evacuation from Haiti for medical care. This rate of recent DENV infection is similar to findings from two previous studies in Haiti that reported rates of infection as high as 25% in expatriates and 29% in military personnel (12,15). These findings demonstrate the risk of dengue for visitors to and residents of Haiti, and also illustrate the potential economic consequences of dengue through missed work days, hospitalization, and medical evacuation [27], [28].
While there is no vaccine to prevent dengue, people at risk for DENV infection, such as the NGO workers in our investigation, can reduce their chance of getting infected through a number of preventive measures like applying mosquito repellent multiple times a day and wearing permethrin-treated clothing [8], [9]. The NGO workers in our investigation variably employed these preventive measures: less than half of expatriates reported using mosquito repellent multiple times a day, and only 10% of expatriates used permethrin-treated clothing. The majority of expatriates in our investigation had a pre-travel health consultation. This rate is higher than previous reports of pre-travel health consultations among US citizens who traveled to countries with elevated public health risks [29], [30]. However, in our investigation, less than half of expatriates received information about dengue during their pre-travel consultation. Improving pre-deployment education of expatriate NGO workers could increase the likelihood that they will employ preventive measures once they are in the field.
All Haitian NGO workers had evidence of prior DENV infection, providing further evidence of dengue endemicity in Haiti. While our investigation and previous studies [11], [12], [14], [31] collectively provide strong evidence for dengue endemicity in Haiti, questions remain about the clinical course of dengue among Haitians. Some studies have hypothesized that Haitians and persons of African descent are less likely to experience severe dengue [11], [32]. In fact, in our investigation, some Haitian NGO staff said that dengue was not a health threat to Haitians and therefore declined to participate. In our investigation, only one Haitian (11%) with a recent DENV infection reported dengue-like symptoms, and no Haitians reported symptoms of severe dengue. However, hospital and clinic-based surveillance conducted over the last two years in Haiti has shown that dengue is associated with both clinic visits and hospital admissions among Haitians [13], [14]. Broader surveillance should be undertaken in Haiti to better understand the burden and clinical course of dengue in Haitians.
Among all the sites we inspected in Léogane, the Premise and Breteau indices were 61% and 198, respectively, reflecting an increased risk for DENV transmission [2], [33]. These findings, according to WHO guidelines, indicate a need to prioritize vector control [26]. The density of immature vectors found in Léogane was greater than what was reported in a survey conducted in Port-au-Prince in May 2011 [34]. In our risk factor analysis, we found that NGO staff who worked near open water had an increased risk of DENV infection. Although this question did not clearly define open water with examples, anecdotally respondents interpreted this question to mean open containers filled with water. Other studies, including a recent study conducted in Saudi Arabia [35], have identified proximity to standing water as a risk factor for DENV infection. Efforts should be made by NGOs and individuals to eliminate mosquito-breeding habitats by systematically reducing standing water in containers around worksites and residences. While vector control has had mixed results with regard to decreasing DENV infections [36], it is still effective at reducing DENV-transmitting mosquitoes by eliminating container habitats [37], [38]. However, dengue risk perceptions need to be addressed within these communities to make these efforts sustainable [39], [40].
We found some unexpected results in our risk factor analysis. The use of mosquito repellent multiple times a day was associated with DENV infection. This finding is most likely due to sampling bias; four (24%) of the recently infected participants in this investigation had received a diagnosis of dengue in the three months prior to our investigation and subsequently received education about dengue. Their responses to these questions likely reflected a change in risk perceptions and an awareness of dengue that they acquired after receiving their diagnosis [41], [42]. Because the questionnaire did not distinguish whether information acquired after receiving a diagnosis of dengue had an effect on knowledge or practice, our findings related to this risk factor are likely spurious. Avoidance of mosquito bites by use of mosquito repellent is a widely supported measure for dengue prevention [8], [9], [43].
Our investigation was subject to several limitations. Because we used a convenience sample and less than half of all NGO workers at the Léogane-based NGOs and IFRC in Port-au-Prince participated, our results may not be representative of all workers at these NGOs. Also, we were not able to systematically evaluate whether there were demographic differences between participants and non-participants. While anecdotally some NGO workers declined to participate because of doubts about the relevance of dengue in Haiti, others were unavailable at the time of the survey. In addition, the investigation was conducted only among NGO workers in the two cities, and therefore our results may be not generalizable to other parts of the expatriate and Haitian population. Our analysis of risk factors for DENV infection was limited by a relatively small sample size. Finally, we were not able to link the results from the serosurvey with our findings from the entomologic investigation.
Our findings underscore the risk of dengue in expatriate workers in Haiti. Expatriate NGO staff should be briefed on dengue risk and prevention measures prior to their arrival in Haiti, and NGOs should systematically employ vector control measures at their work sites and residences to reduce mosquito populations. We found evidence of acute dengue virus infections in Haitians and we found a high rate of previous infection among Haitians. Surveillance and research should be undertaken to better understand clinical dengue in Haitians.
|
10.1371/journal.pbio.2003174 | A damped oscillator imposes temporal order on posterior gap gene expression in Drosophila | Insects determine their body segments in two different ways. Short-germband insects, such as the flour beetle Tribolium castaneum, use a molecular clock to establish segments sequentially. In contrast, long-germband insects, such as the vinegar fly Drosophila melanogaster, determine all segments simultaneously through a hierarchical cascade of gene regulation. Gap genes constitute the first layer of the Drosophila segmentation gene hierarchy, downstream of maternal gradients such as that of Caudal (Cad). We use data-driven mathematical modelling and phase space analysis to show that shifting gap domains in the posterior half of the Drosophila embryo are an emergent property of a robust damped oscillator mechanism, suggesting that the regulatory dynamics underlying long- and short-germband segmentation are much more similar than previously thought. In Tribolium, Cad has been proposed to modulate the frequency of the segmentation oscillator. Surprisingly, our simulations and experiments show that the shift rate of posterior gap domains is independent of maternal Cad levels in Drosophila. Our results suggest a novel evolutionary scenario for the short- to long-germband transition and help explain why this transition occurred convergently multiple times during the radiation of the holometabolan insects.
| Different insect species exhibit one of two distinct modes of determining their body segments (known as segmentation) during development: they either use a molecular oscillator to position segments sequentially, or they generate segments simultaneously through a hierarchical gene-regulatory cascade. The sequential mode is ancestral, while the simultaneous mode has been derived from it independently several times during evolution. In this paper, we present evidence suggesting that simultaneous segmentation also involves an oscillator in the posterior end of the embryo of the vinegar fly, Drosophila melanogaster. This surprising result indicates that both modes of segment determination are much more similar than previously thought. Such similarity provides an important step towards our understanding of the frequent evolutionary transitions observed between sequential and simultaneous segmentation.
| The segmented body plan of insects is established by two seemingly very different modes of development [1–4]. Long-germband insects, such as the vinegar fly D. melanogaster, determine their segments more or less simultaneously during the blastoderm stage, before the onset of gastrulation [5, 6]. The segmental pattern is set up by subdivision of the embryo into different territories, prior to any growth or tissue rearrangements. Short-germband insects, such as the flour beetle T. castaneum, determine most of their segments after gastrulation, with segments being patterned sequentially from a posterior segment addition zone. This process involves tissue growth or rearrangements as well as dynamic travelling waves of gene expression, which result from periodic oscillations that are driven by a molecular clock mechanism [7–10] (technical terms in bold are explained in the glossary, in S1 Text). The available evidence strongly suggests that the short-germband mode of segment determination is ancestral, while the long-germband mode is evolutionarily derived [1, 2, 11].
Although the ancestor of holometabolan (metamorphosing) insects may have exhibited some features of long-germband segment determination [12], it is clear that convergent transitions between the two modes have occurred frequently during evolution [2, 11, 13]. Long-germband segment determination can be found scattered over all four major holometabolous insect orders (Hymenoptera, Coleoptera, Lepidoptera, and Diptera). Furthermore, there has been at least one reversion from long- to short-germband segment determination in polyembryonic wasps [14]. This suggests that, despite the apparent differences between the two segmentation modes, it seems relatively easy to evolve one from the other. Why this is so, and how the transition is achieved, remains unknown.
In this paper, we provide evidence suggesting that the patterning dynamics of long- and short-germband segmentation are much more similar than previously thought. Specifically, we demonstrate that shifting domains of segmentation gene expression in the posterior of the D. melanogaster embryo can be explained by a damped oscillator mechanism, dynamically very similar to the clocklike mechanism underlying periodically oscillating gene expression during short-germband segment determination. We achieve this through analysis of a quantitative, data-driven gene circuit model of the gap network in D. melanogaster. The gap gene system constitutes the topmost hierarchical layer of the segmentation gene cascade [6]. Gap genes hunchback (hb), Krüppel (Kr), giant (gt), and knirps (kni) are activated through morphogen gradients formed by the products of maternal coordinate genes bicoid (bcd) and caudal (cad). Gap genes are transiently expressed during the blastoderm stage in broad overlapping domains along the anteroposterior (A–P) axis of the embryo (Fig 1A). They play an important role regulating spatially periodic pair-rule gene expression. Pair-rule genes, in turn, establish the precise pre-pattern of the segment-polarity genes, whose activities govern the morphological formation of body segments later in development, after gastrulation has occurred.
Our aim is to go beyond the static reconstruction of network structure to explicitly understand the regulatory dynamics of the patterning process [15, 16]. To achieve this, we use the powerful tools of dynamical systems theory—especially the geometrical analysis of phase (or state) space [17]—to characterize the patterning capacity of the gap gene network. We study the complex regulatory mechanisms underlying gap gene expression in terms of the number, type, and arrangement of attractors and their associated basins of attraction, which define the phase portrait. The geometry of the phase portrait in turn determines the flow of the system. This flow consists of individual trajectories that describe how the system state changes over time given some specific initial conditions. In our gap gene circuit model, initial conditions are given by the maternal Hb gradient, boundary conditions by the maternal Bcd and Cad gradients, and the state variables consists of the concentrations of regulators Hb, Kr, Kni, and Gt. Different configurations of phase space give rise to differently shaped trajectories and, thus, to different gap gene regulatory dynamics.
The power of analogy between phase space and its features, and developmental mechanisms, has long been recognized and exploited. In their original "clock-and-wavefront" model, Cooke and Zeeman [18] characterize cells involved in somitogenesis in the pre-somitic mesoderm as "oscillators with respect to an unknown clock or limit cycle in the embryo." More recently, geometrical analysis of phase space has been successfully used to study developmental processes such as vertebrate somitogenesis [19], vulval development in nematodes [20], A–P patterning by Hox genes [21], and—particularly relevant in our context—the robust (canalized) patterning dynamics of gap genes [22–25]. To make the problem tractable, these analyses are often performed in a simplified framework. For example, in previous studies of Drosophila segmentation, models were used with a static Bcd gradient and Cad dynamics frozen after a particular time point during the late blastoderm stage [22, 23, 25–27]. This rendered the system autonomous, meaning that model parameters—and therefore phase space geometry—remain constant over time.
However, the maternal gradients of Bcd and Cad change and decay on the same timescale as gap gene expression [28]. Taking this time dependence of maternal regulatory inputs into account leads to a nonautonomous dynamical system, in which model parameters are allowed to change over time (see [29] and S1 Text for a detailed model comparison). This causes the geometry of phase space to become time-variable: the number, type, and arrangement of attractors and their basins change from one time point to the next. Bifurcations may occur over time, and trajectories may cross from one basin of attraction to another. All of this makes nonautonomous analysis highly nontrivial. We have developed a novel methodology to characterize transient dynamics in nonautonomous models [30]. It uses instantaneous phase portraits [29, 31] to capture the time-variable geometry of phase space and its influence on system trajectories.
By fitting dynamical models to quantitative spatiotemporal gap gene expression data, we have obtained a diffusion-less, fully nonautonomous gap gene circuit featuring realistic temporal dynamics of both Bcd and Cad (Fig 1A) [29, 32] (see Materials and methods and S1 Text for details). The model has been extensively validated against experimental data [22, 23, 26, 27, 29, 32] and represents a regulatory network structure that is consistent with genetic and molecular evidence [6].
We have performed a detailed and systematic phase space analysis of this nonautonomous gap gene circuit along the segmented trunk region of the embryo, explicitly excluding head and terminal patterning systems [29] (see Materials and methods for details). At every A–P position between 35% and 73%, we calculated the number and type of steady states in the associated phase portrait [29]. This allowed us to characterize the different dynamical regimes driving gap gene expression along the embryo trunk and to explicitly identify the time-dependent aspects of gap gene regulation [29]. In the anterior trunk region of the embryo, where boundary positions remain stationary over time, gap gene expression dynamics are governed by a multi-stable dynamical regime (Fig 1B) [29]. This is consistent with earlier work [23], indicating that modelling results are robust across analyses. Here, we focus on the regulatory mechanism underlying patterning dynamics in the posterior of the embryo, which differs between autonomous and nonautonomous analyses.
Posterior gap domains shift anteriorly over time [26, 28]. Autonomous analyses suggested that these shifts are driven by a feature of phase space called an unstable manifold [23], while our nonautonomous analysis reveals that they are governed by a mono-stable spiral sink (Fig 1B). The presence of a spiral sink indicates that a damped oscillator mechanism is driving gap domain shifts in our model [17]. Here, we present a detailed mathematical and biological analysis of this damped oscillator mechanism in the posterior of the embryo, between 53% and 73% A–P position, and discuss its implications for pattern formation and the evolution of the gap gene system. Our results suggest that long-germband and short-germband modes of segmentation both use oscillatory regimes (damped and limit cycle oscillators, respectively) in the posterior region of the embryo to generate posterior to anterior waves of gene expression. Characterizing and understanding these unexpected similarities provides a necessary first step towards a mechanistic explanation for the surprisingly frequent occurrence of convergent transitions between the two modes of segment determination during holometabolan insect evolution.
The gap gene circuit model used for our analysis consists of a one-dimensional row of nuclei along the A–P axis [32, 33]. Continuous dynamics during interphase alternate with discrete nuclear divisions. Our full model includes the entire segmented trunk region of the embryo between 35% and 92% A–P position. It covers the last two cleavage cycles of the blastoderm stage (starting at the end of cleavage cycle 12, C12, at t = 0, including C13 and C14A) up to the onset of gastrulation; C14A is subdivided into 8 equally spaced time classes (T1–T8). Division occurs at the end of C13.
The state variables of the system represent the concentrations of proteins encoded by gap genes hb, Kr, gt, and kni. The concentration of protein a in nucleus i at time t is given by gia(t). Change in protein concentration over time occurs according to the following system of ordinary differential equations:
dgia(t)dt=Raϕ(ua)−λagia(t)
(1)
where Ra and λa are rates of protein production and decay, respectively. ϕ is a sigmoid regulation-expression function used to represent the cooperative, saturating, coarse-grained kinetics of transcriptional regulation. It incorporates nonlinearities into the model that enable it to exhibit complex behavior, such as multi-stability and damped or sustained oscillations. It is defined as
ϕ(ua)=12(ua(ua)2+1+1)
(2)
where
ua=∑b∈GWbagib(t)+∑m∈MEmagim(t)+ha
(3)
The set of trunk gap genes is given by G = {hb, Kr, gt, kni} and the set of external regulatory inputs by the products of maternal coordinate and terminal gap genes M = {Bcd, Cad, Tailless(Tll), Huckebein(Hkb)}. Concentrations of external regulators gim are interpolated from quantified spatiotemporal protein expression data [28, 32, 34]. Changing maternal protein concentrations means that parameter term ∑m∈MEmagim(t) is time dependent, which renders the model nonautonomous.
Interconnectivity matrices W and E represent regulatory interactions between gap genes and from external inputs, respectively. Matrix elements wba and ema are regulatory weights. They summarize the effect of regulator b or m on target gene a and can be positive (representing an activation), negative (repression), or near zero (no interaction). ha is a threshold parameter representing the basal activity of gene a, which includes the effects of regulatory inputs from spatially uniform regulators in the early embryo. The system of equations (Eq 1) governs regulatory dynamics during interphase; Ra is set to zero during mitosis. Additional information about our model formalism can be found in S1 Text.
We obtained values for parameters Ra, λa, W, E, and ha by fitting the model to data over a full spatial range covering the segmented trunk region between 35% and 92% A–P position (see S1 Data) [26, 32, 35, 36]. Signs of parameters in the genetic interconnectivity matrices W and E were constrained during the fit to allow direct comparison with previously published models [23, 32]. A detailed account of how we fit the model and selected solutions for analysis has been published previously [29]; we provide a summary in S1 Text. Briefly, model equations (Eq 1) are solved numerically, and the resulting model output is compared to a quantitative data set of spatiotemporal gap protein profiles. The difference between model output and data is minimized using parallel Lam Simulated Annealing (pLSA). Model fitting was performed on the Mare Nostrum cluster at the Barcelona Supercomputing Centre (http://www.bsc.es). The best-fitting solution was selected for further analysis, as described in S1 Text (model parameters are shown in S1 Table). The resulting diffusion-less, nonautonomous gene circuit has a residual error (measured by its root mean square score) of 14.53 (see S1 Text). It reproduces gap gene expression with high accuracy, showing only minor defects in the shape of expression domain boundaries (Fig 1A).
The modelling and optimization code to reverse-engineer the gap gene network is implemented in C, using MPI for parallelization and the GNU Scientific Library (GSL, http://www.gnu.org/software/gsl) for data interpolation. It is available for download online at https://subversion.assembla.com/svn/flysa.
Embryos derived from cad mutant germ-line clones were generated and collected as previously described [39, 40], and females were then mated to wild-type males. The resulting embryos all lack maternal cad activity but carry one paternal copy of the cad gene. mRNA expression patterns of the gap genes gt or kni, and the pair-rule gene even-skipped (eve) were visualized using an established enzymatic (colorimetric) in situ hybridization protocol [36]. Images were taken and processed using FlyGUI (https://subversion.assembla.com/svn/flygui) to extract the position of expression domain boundaries, as described in [41]. The image data and extracted boundary positions are available from figshare at https://figshare.com/s/839791c208e42b7e61fe (DOI: 10.6084/m9.figshare.5809653).
Gap domain boundaries posterior to 52% A–P position shift anteriorly over time (Fig 1A and Fig 2A) [26, 28]. These domain shifts cannot be explained by nuclear movements [42], nor do they require diffusion or transport of gap gene products between nuclei [22, 23, 26, 29] (see also S1 Text). Instead, gap domain shifts are kinematic, caused by an ordered temporal succession of gene expression in each nucleus, which produces apparent wavelike movements in space [23, 26]. This is illustrated in Fig 2A for nuclei between 55% and 73% A–P position (see Materials and methods). Each nucleus starts with a different initial concentration of maternal Hb, which leads to the expression of different zygotic gap genes: Kr in the central region of the embryo or kni further posterior. Nuclei then proceed through a stereotypical temporal progression, in which Kr expression is followed by kni (e.g., nucleus at 59%), kni by gt (nucleus at 69%), and, finally, gt by hb (nuclei posterior of 75%; not shown). No nucleus goes through the expression of all four trunk gap genes over the course of the blastoderm stage and each nucleus goes through a different partial sequence within this progression, according to its initial conditions. This coordinated dynamic behavior is what we need to explain in order to understand the regulatory mechanism underlying gap domain shifts.
To do this, we carried out a systematic characterization of the dynamical regimes driving A–P gap gene patterning in a nonautonomous gap gene circuit model [29]. For every nucleus along the trunk region of the embryo, we visualized the dynamics of gap gene expression in the context of the instantaneous phase portraits that underlie them. That is, we calculated the positions and types of steady states present at every time class and plotted them (color coded for time) with the simulated expression dynamics for that nucleus. This yielded a full nonautonomous phase portrait associated with each nucleus. In this way, we can understand each trajectory's shape in terms of the changing geometry of the flow (see Materials and methods for details).
Our analysis revealed that phase portraits of nuclei between 53% and 73% A–P position are mono-stable throughout the blastoderm stage (see, for example, Fig 2B). Given enough time, all trajectories would approach the only attractor present, which, at the end of the blastoderm stage (time class T8), is located close to the origin (Fig 2B, yellow cylinder). Due to the nonautonomy of the system, this attractor moves across phase space over developmental time. However, this movement of the attractor is not the most important factor determining the shape of trajectories. Due to the limited duration of the blastoderm stage, the system always remains far from steady state, and posterior gap gene expression dynamics are determined by the geometry of transient trajectories relatively independently of the precise position of the attractor. Because the moving attractor positions are similar for all posterior nuclei, we were able to plot the trajectories of the different nuclei onto the same projection of phase space (Fig 2C). Over time, posterior nuclei transit through buildup of Kr, then Kni, then Gt proteins. Their initial conditions are given by Hb and this determines where in the sequence they start. The plots in Fig 2B and 2C show that the ordered succession of gap gene expression is a consequence of the rotational (spiral-shaped) geometry of the trajectories.
Eigenvalue analysis revealed that the mono-stable steady state of posterior nuclei is a special type of point attractor: a spiral sink, or focus [17, 29]. Trajectories do not approach such a sink in a straight line but spiral inward, instead. This contributes to the curved rotational geometry of the trajectories shown in Fig 2B and 2C. From the theory of dynamical systems, we know that spiral sinks are the hallmark of damped oscillators [17]. Given that spiral sinks are the only steady states present in the mono-stable phase portraits of posterior nuclei, we concluded that, in our model, posterior gap gene expression dynamics are driven by a damped oscillator mechanism. This damped oscillator mechanism imposes the observed temporal order of gap gene expression (Fig 2A). Temporal order is a natural consequence of oscillatory mechanisms, one obvious example being the stereotypical succession of cyclin gene expression driven by the cell cycle oscillator [43, 44]. In contrast, the imposition of temporal order is not a general property of unstable manifolds (found to drive gap domain shifts in previous autonomous analyses [23–25]). For this reason, our damped oscillator mechanism provides a revised understanding of gap domain shifts, which is more general and therefore constitutes an important conceptual advance over previous characterizations.
Each nucleus runs through a different range of phases within a given time period (see color wheel diagrams in Fig 2A), as determined by the damped oscillator. Arranged properly across space, phase-shifted partial trajectories create the observed kinematic waves of gene expression. In this sense, the dynamics of the shifting gap domains in the D. melangoaster blastoderm and those of the travelling waves of gene expression in short-germband embryos are equivalent, because they are both an emergent property of the temporal order imposed by an underlying oscillatory regulatory mechanism.
In principle, domain shifts are not strictly necessary for subdividing an embryo into separate gene expression territories. Wolpert's French Flag paradigm for positional information, for example, works without any dynamic patterning downstream of the morphogen gradient [45, 46]. This raises the question of why such shifts occur and what, if anything, they contribute to pattern formation. One suggestion is that feedback-driven shifts lead to more robust patterning than a strictly feed-forward regulatory mechanism, such as the French Flag [47, 48]. This is supported by the fact that the unstable manifold found in autonomous analyses [23] has canalizing properties: as time progresses, it attracts trajectories coming from different initial conditions into an increasingly small and localized subvolume of phase space. This desensitizes the system to variation in maternal gradient concentrations [22]. Based on these insights, we asked whether our damped oscillator mechanism exhibits similar canalizing behavior, ensuring robust gap gene patterning.
A closer examination of the spiral trajectories in Fig 2C reveals that they are largely confined to two specific sub-planes in phase space (see S1 and S2 Movies). Specifically, they tend to avoid regions of simultaneously high levels of Gt and Kr, allowing us to "unfold" the three-dimensional volume of Kr-Kni-Gt space into two juxtaposed planes representing Kr-Kni and Kni-Gt concentrations (Fig 2D). This projection highlights how trajectories spend variable amounts of time on the Kr-Kni plane before they transition onto the Kni-Gt plane.
In order to investigate the canalizing properties of our damped oscillator mechanism, we performed a numerical experiment, shown in Fig 3A and 3B. We chose a set of regularly distributed initial conditions for our model that lies within the Kr-Gt plane (Fig 3A) and used this set of initial conditions to simulate the nucleus at 59% A–P position, with a fixed level of Kni (Fig 3B). These simulations illustrate how system trajectories converge to the Kr-Kni or Kni-Gt plane, avoiding regions of simultaneously high Kr and Gt concentrations. Convergence occurs rapidly and is already far advanced in early cleavage cycle 14A (Fig 3B, time class T1), demonstrating that the subvolume of phase space in which trajectories are found becomes restricted long before a steady state is reached. At later stages, convergence slows down but continues confining trajectories to an increasingly restricted subvolume of phase space (up to late cleavage cycle 14A, Fig 3B, time class T8). This phenomenon can be seen as the equivalent of trajectories becoming restricted to valleys in Waddington's original landscape metaphor, which motivated the definition of the term "canalization" [49]. The canalizing behavior is robust with regard to varying levels of Kni (S1 Fig).
It is straightforward to interpret the exclusion of trajectories from regions of simultaneous high Kr and high Gt in terms of regulatory interactions. There is strong bidirectional repression between gt and Kr, which is crucial for the mutually exclusive expression patterns of these genes [6, 27, 36]. In the context of our damped oscillator mechanism, this mutual repression implies that the system must first transition from high Kr to high Kni/low Kr before it can initiate gt expression. This is exactly what we observe (Fig 2A), confirming that the damped oscillator in the posterior of the D. melanogaster embryo has canalizing properties due to mutually exclusive gap genes.
How do spiral trajectories switch from one plane in phase space to another? To answer this question, we examined the flow of the system. We unfolded the Kr-Kni and Kni-Gt planes and projected trajectories and states of posterior nuclei onto this unfolded flow (Fig 3C and S2 Fig). These plots reveal drastic differences in flow velocity (magnitude) in different regions of phase space at different points in time. At early stages, close to the origin, we observe a fast initial increase in Kr and Kni concentrations, indicated by red arrows at low Kr and Kni concentrations in Fig 3C (C13 and T2). Nuclei whose trajectories remain on the Kr-Kni plane then show a dramatic slowdown. They either continue to gradually increase levels of Kr or exhibit slow buildup of Kni, combined with consequent decrease of Kr due to repression by Kni (Fig 3C, T4 and T6). As trajectories of different nuclei approach the border between the Kr-Kni and Kni-Gt planes, the Gt component of the flow on the Kr-Kni plane becomes positive (trajectories marked by stars in Fig 3C and S2 Fig). This "lifts" the trajectory out of the Kr-Kni and into the Kni-Gt plane. In the border zone between the two planes, the flow in the direction of Gt is high throughout the blastoderm stage (Fig 3C), ensuring that the switch between planes occurs rapidly. Nuclei then again enter a zone of slower dynamics with a gradual buildup of Gt, combined with consequent decrease of Kni due to repression by Gt (Fig 3C, T4 and T6).
Thus, the flow of our model combines relatively slow straight stretches within a plane of phase space with rapid turns at the border between planes. Similar alternating fast-slow dynamics have been observed in autonomous models [24]. These dynamics are important for gap gene patterning because they influence the width of gap domains (through relatively stable periods of expressing a specific gap gene) and the sharpness of domain boundaries (through abrupt changes in gene expression at borders between planes). Such fast-slow dynamics are characteristic of relaxation oscillations [17]. A relaxation oscillator combines phases of gradual buildup in some of its state variables with rapid releases and changes of state, resulting from an irregularly shaped limit cycle. Although there seem to be no limit cycles present in our phase portraits, the irregular geometries of spiralling transient trajectories in our model can be understood as relaxation-like (fast-slow) dynamics, which, driven by a damped oscillator, govern the shape and the shift rate of posterior gap domains.
In the short-germband beetle T. castaneum, an oscillator mechanism governs travelling waves of pair-rule gene expression [7, 8]. The frequency of these repeating waves is positively correlated with the level of Cad in the posterior of the embryo: the more Cad present, the faster the oscillations [9]. In addition, a recent publication proposes that waves of gap gene expression observed in the T. castaneum blastoderm and elongating germ band may be caused by a succession of temporal gene expression switches whose rate and timing is also under control of the posterior gradient of Cad [50]. These authors speculate that Cad may control gap gene expression in D. melanogaster in an equivalent way. In D. melanogaster, changing concentrations of maternal morphogens do indeed influence posterior gap domain shifts [29, 39]. Therefore, we asked how altered levels of Cad affect the damped oscillator mechanism regulating gap genes in D. melanogaster.
We assessed the regulatory role of Cad by multiplying its concentration profile with different constant scaling factors—reducing Cad levels in space and time without affecting overall profile shape—and by measuring the dynamics and extent of gap domain shifts in the resulting simulations (Fig 4). In particular, we focus on how lowered levels of Cad affect the position of the Kr-Gt interface over time (Fig 4A and 4B). Our model makes three specific predictions. First, the initial position of the Kr-Gt border interface does not change when Cad levels are decreased (Fig 4B, C13). Second, between C13 and C14A-T1, gap domains simulated with lowered concentrations of Cad start to lag behind those simulated with wild-type levels (Fig 4B, C13 and T1). Third, from T1 onwards, shift rates become independent of Cad concentration, and boundary positions move in parallel in different simulations for the remainder of the blastoderm stage (Fig 4B, T1–T8). This last prediction is incompatible with a mechanism in which the rate of successive bifurcation-driven switches is under the direct control of Cad, which requires the shift rate to be sensitive to Cad concentration [50].
A comparison of the flow in models with reduced and wild-type levels of Cad revealed that this maternal factor affects the timing of gap domain shifts by modulating the fast-slow dynamics of the gap gene damped oscillator. While the direction of the flow remains largely constant across different concentrations of Cad, its magnitude changes significantly (Fig 4C–4E and S3 Fig). The magnitude of the flow is most sensitive in the area of the Kr-Kni plane around the origin, where it is strongly reduced at early stages in simulations with lowered levels of Cad (Fig 4C–4E, time class C12). This implies a slower initial buildup of Kr and Kni protein at low Cad and hence the delayed onset of domain shifts. At later stages, when wild-type Cad levels decrease, differences in the magnitude of the flow are very subtle (Fig 4C–4E, time class T8, and S3 Fig, from time class C14A-T3 onwards). As a result of the altered early flow, the curvature of trajectories is decreased with lower Cad concentration, leading to tighter spirals. This demonstrates that the early difference in Cad levels continues to influence the behavior of the gap system into the late blastoderm stage (S4 Fig). Progress along these tightened spirals is much slower than along the wider ones followed in wild type, due to the weaker flow in regions near the origin (compare S2 Fig and S4 Fig). This slowed progress compensates for the tightened geometry of the spiral trajectories, preserving the rate of change in the "phase" of gap gene expression. In this way, the relative rate of the shifts remains unperturbed by changing the concentration levels of Cad, leading to the parallel trajectories after C14A-T1 depicted in Fig 4B.
To experimentally test the predictions from our model, we need to carefully manipulate the levels of Cad protein in blastoderm embryos without disturbing the spatial pattern too much. This is difficult to achieve due to the lack of well-characterized hypomorphic mutants of cad in D. melanogaster and the overlapping but distinct spatiotemporal profiles of the maternal and zygotic expression contributions [51, 80]. In the absence of more precise genetic tools, we quantified boundary shifts of Gt and Kni domains in mutant embryos derived from cad germ-line clones, which lack the maternal contribution to Cad expression. These mutants are viable as long as one paternal copy of cad is present, and exhibit reduced levels of (zygotic) Cad protein, with a spatial expression profile that is comparable to the wild type at the late blastoderm stage [51]. As predicted by our simulations, these mutants show delayed shifts of the posterior Gt (Fig 5, and S5 Fig) and the abdominal Kni domains [39].
Here, we focus on the anterior boundary of the posterior Gt domain (Fig 5A, arrowhead), which corresponds to the Kr-Gt interface measured in Fig 4. It satisfies all three model predictions. First, its position at the onset of Gt expression in C13 is the same in mutant and wild-type embryos. This corroborates earlier analyses suggesting that maternal Hb (and not Cad) is the main morphogen in the posterior of the embryo [6, 23, 29, 52]. Second, between C13 and C14A-T1, it lags behind its wild-type position, exhibiting a subtle but clearly detectable posterior displacement by T1 (Fig 5A). Gap domain shifts are only initiated around late C13, when enough gap protein has accumulated to initiate cross-regulatory interactions [6, 53]. The slower accumulation of gap protein in the posterior of the embryo therefore causes a delay in the onset of the shifts in the mutant. Third, from T1 onwards, shift rates in wild type and mutants remain more or less the same, indicating that they are robust towards changes in levels of Cad (Fig 5, after C14A-T1). Even though the conditions of model simulations and mutants may not match perfectly, this provides clear evidence that gap domain shifts are relatively insensitive to the precise level of Cad concentration.
Taken together, our experimental and modelling evidence suggest that Cad regulates the timing but not the positioning of gap gene expression in early blastoderm stage embryos of D. melanogaster. At later stages, gap domain shift rates are robust towards changes in Cad concentration. This is not entirely surprising, because the shifts result from gap–gap cross-regulatory interactions rather than depending directly on maternal input [6, 26, 32, 36]. Analysis of our model shows that this robustness is entirely consistent with a damped oscillator mechanism, while a mechanism based on temporal switching under the control of Cad [50] would be much more sensitive to altered levels of the maternal gradient.
In this paper, we have shown that a damped oscillator mechanism—with relaxation-like behavior—can explain robust segmentation gene patterning of the long-germband insect D. melanogaster. Even though they may not be periodic, the kinematic shifts of gap gene expression domains in our model are an emergent property of temporally regulated gene expression driven by a damped oscillator. In this sense, they are dynamically equivalent to the travelling waves of gene expression involved in vertebrate somitogenesis [19, 54] and short-germband arthropod segmentation [7–9, 55, 56], both of which also emerge from temporal order imposed by oscillatory mechanisms. This lends support to the notion that the regulatory dynamics of segmentation gene expression in long- and short-germband insects are much more similar than is evident at first sight [57, 58].
The mechanism described in this paper differs from an earlier proposal that gap domain shifts are driven by an unstable manifold [23]. Can these two mechanisms be distinguished experimentally? We think they can, because the two models make different predictions for embryos misexpressing hb in the posterior region of the embryo. According to the model put forward by Manu and colleagues [23], nuclei exposed to high maternal Hb concentrations will rapidly converge to an attractor with high zygotic Hb concentration by the end of the blastoderm stage. In contrast, our model predicts these nuclei will express high levels of Kr in addition to hb (S6 Fig). Because real embryos misexpressing hb under a heat-shock promoter show high levels of Kr in the posterior embryo trunk region [59, 60], our model is better supported by the available experimental evidence.
In addition to these empirical considerations, the proposed damped oscillator provides a more general explanation of the developmental and evolutionary dynamics of gap gene expression than the unstable manifold reported previously [23]. The spiral geometry of this manifold is contingent. It happens to traverse all the relevant expression states (from Kr to kni to gt to hb), but such a succession of states is not a general characteristic of unstable manifolds. In contrast, cycling through successive states is not just typical for our proposed damped oscillator; it is the hallmark of gene expression oscillators in general.
A succession of gene expression states could also be generated by a timed series of bifurcation-based switches, as suggested by Tufcea and François [61]. This relies on a precise mechanism for the temporal regulation of the switches. Zhu and colleagues [50] have recently proposed that Cad controls such a cascade of gap gene switches in both T. castaneum and D. melanogaster. The evidence presented here renders this scenario unlikely, at least in the case of D. melanogaster. One problem with the timed-switch mechanism is that it remains unclear how it could be implemented by the known interactions among gap genes [6]. Another problem is that it operates at criticality throughout the embryo—undergoing a rapid series of bifurcations. This leaves it extremely sensitive to changes in Cad concentration, unlike the robust oscillator reported here. Interestingly, there is some indication for such widespread criticality in the gap gene system from a recent study using quantitative co-expression measurements and a simplified set of gene regulatory models [62]. We could not find any evidence for this type of criticality in our model, which is based on a detailed and experimentally validated regulatory structure of the gap gene network [6, 23, 26, 29, 32].
Shifting gap domains play a central role in segmental patterning in D. melanogaster by directly regulating stripes of pair-rule gene expression. Posterior pair-rule stripes also exhibit anterior shifts in this species. They are produced by and closely reflect the expression dynamics of the gap genes [28]. In fact, dynamic shifts in gap domain positions are strictly required for the correct spatiotemporal expression of pair-rule genes in D. melanogaster [58]. In contrast, gap genes play a much less prominent role in patterning posterior segments in short-germband arthropods. Instead, periodic kinematic waves of pair-rule gene expression are thought to be generated by negative feedback between the pair-rule genes themselves (in T. castaneum [63]) or by an intercellular oscillator driven by Notch/Delta signalling (in cockroaches [64] and centipedes [55, 56]).
The evolutionary transition from short- to long-germband segmentation has long been thought to have involved the recruitment of gap genes for pair-rule gene regulation, to replace the ancestral oscillatory mechanism [6, 12, 13, 65, 66]. The mechanistic details of how this occurred remain unclear. Gap gene–driven and segmentation clock–driven modes of patterning have been assumed to be mutually exclusive in any given region of the embryo. In contrast, our results suggest that during the replacement process, gap and pair-rule oscillators might have temporarily coexisted, which would greatly facilitate the transition. In this scenario, gap genes gradually take over pair-rule–driven oscillatory patterning in the posterior and later convert to a more switch-like static patterning mode, as observed in the anterior of the D. melanogaster embryo [23, 27–29]. This is tentatively supported by the fact that the spatial extent of the posterior region, which is patterned by shifting gap domains, differs between dipteran species [39, 67]. This scenario suggests that posterior gap domains shift as a result of the dynamic regulatory context into which they have been recruited during evolution. In addition, it provides an explanation for why gap domain shifts are essential for the correct placement of pair-rule stripes in D. melanogaster [58].
Seen from another angle, our results imply that equivalent regulatory dynamics (in this case, domain shifts and travelling waves of gene expression) can be produced by different oscillatory mechanisms. The use of divergent regulatory mechanisms to independently pattern identical expression domains appears to be very common (see, for example, [68–71]). Indeed, the relative contribution of different mechanisms may evolve over time, with little effect on downstream patterning [72]. This type of compensatory evolution is called developmental system drift [73–77]. It has recently been shown to occur extensively in the evolution of the dipteran gap gene system [39, 78]. System drift provides the necessary conditions that enable the facilitated gradual transition between the different regulatory mechanisms described above.
Even though the core mechanisms that generate both behaviors differ, some aspects of segmentation gene regulation are strikingly similar between long- and short-germband insects. In different species of dipteran insects, as well as in T. castaneum, travelling kinematic waves of gene expression are involved in segment determination [9, 26, 39, 50, 67]. Cad is always involved in the initial activation of these patterns [9, 39, 50, 79–82]. It also appears to control aspects of pair-rule gene regulation in centipedes [55, 56]. From this, we conclude that the activating role of Cad in initiating these dynamics is highly conserved. In contrast, our evidence argues against a proposed universal role of Cad in regulating the rate and dynamics of travelling waves of segmentation gene expression [50]. In D. melanogaster, Cad exerts its effect primarily through regulating levels of gap gene expression; it has no direct role in the positioning of gap gene expression domains [29].
Travelling waves of gene expression that narrow and slow down over time are involved in both arthropod segmentation and vertebrate somitogenesis. It has long been recognized that these expression dynamics imply differential regulation of the rate of an oscillatory process along the A–P axis [54]. However, mechanistic explanations for this phenomenon remain elusive. A number of recent models simply assume that the concentration of some posterior morphogen determines the period of cellular oscillators, without investigating how this might arise (see, for example, [9, 83, 84]). Experimental evidence from vertebrates suggests alteration of protein stability or translational time delays as a possible mechanism [85, 86]. In contrast, our dynamical analysis illustrates how slowing (damped) oscillations can emerge directly from the intrinsic regulatory dynamics of a transcriptional network, without altering rates of protein synthesis or turnover, or even the need for external regulation by morphogens. A similar mechanism based on intrinsic oscillatory dynamics of a gene network was recently proposed for vertebrate somitogenesis [87]. It will be interesting to investigate which specific regulatory interactions mediate the effect of Cad on the T. castaneum pair-rule gene oscillator.
Patterning by the gap gene system also shows interesting parallels to the developmental system governing the dorsoventral subdivision of the vertebrate neural tube. In both cases, the target domains of the respective morphogen gradients move away from their initial position over time due to downstream gene interactions, and in both cases, this involves a temporal succession of target gene expression [88]. Previous analyses suggest that this temporal succession of gene expression in the vertebrate neural tube may be caused by a succession of bistable switching events [61, 89]. However, the possibility of damped oscillations was never explicitly investigated in any of these analyses. In light of the results presented here, it would be interesting to check for their presence in this patterning system.
In summary, we argue that oscillatory mechanisms of segmentation gene regulation are not exclusive to short-germband segmentation or somitogenesis. Our analysis provides evidence that the spatial pattern of gap gene expression in the posterior region of the D. melanogaster embryo also emerges from a temporal sequence of gap gene expression driven by an oscillatory mechanism: a regulatory damped oscillator. This results in the observed anterior shifts of posterior gap domains. We suggest that the dynamic nature of posterior gap gene patterning is a consequence of the context in which it evolved and that two different oscillatory mechanisms may have coexisted during the transition from short- to long-germband segmentation. Studies using genetics and data-driven modelling in non-model organisms will reveal the regulatory circuits responsible for driving the different dynamics involved in segmentation processes, as well as the precise nature of the regulatory changes involved in transitions between them [39, 78, 90]. Given the insights gained through its application to gap gene patterning in D. melanogaster, phase space analysis will provide a suitable dynamic regulatory context in which to interpret and analyze these results.
|
10.1371/journal.pcbi.1000061 | Distinct Timing Mechanisms Produce Discrete and Continuous Movements | The differentiation of discrete and continuous movement is one of the pillars of motor behavior classification. Discrete movements have a definite beginning and end, whereas continuous movements do not have such discriminable end points. In the past decade there has been vigorous debate whether this classification implies different control processes. This debate up until the present has been empirically based. Here, we present an unambiguous non-empirical classification based on theorems in dynamical system theory that sets discrete and continuous movements apart. Through computational simulations of representative modes of each class and topological analysis of the flow in state space, we show that distinct control mechanisms underwrite discrete and fast rhythmic movements. In particular, we demonstrate that discrete movements require a time keeper while fast rhythmic movements do not. We validate our computational findings experimentally using a behavioral paradigm in which human participants performed finger flexion-extension movements at various movement paces and under different instructions. Our results demonstrate that the human motor system employs different timing control mechanisms (presumably via differential recruitment of neural subsystems) to accomplish varying behavioral functions such as speed constraints.
| A fundamental question in motor control research is whether distinct movement classes exist. Candidate classes are discrete and continuous movement. Discrete movements have a definite beginning and end, whereas continuous movements do not have such discriminable end points. In the past decade there has been vigorous, predominantly empirically based debate whether this classification implies different control processes. We present a non-empirical classification based on mathematical theorems that unambiguously sets discrete and continuous rhythmic movements apart through their topological structure in phase space. By computational simulations of representative modes of each class we show that discrete movements can only be executed repetitively at paces lower than approximately 2.0 Hz. In addition, we performed an experiment in which human participants performed finger flexion-extension movements at various movement paces and under different instructions. Through a topological analysis of the flow in state space, we show that distinct control mechanisms underwrite human discrete and fast rhythmic movements: discrete movements require a time keeper, while fast rhythmic movements do not. Our results demonstrate that the human motor system employs different timing control mechanisms (presumably via differential recruitment of neural subsystems) to accomplish varying behavioral functions such as speed constraints.
| Discrete movements constitute singularly occurring events preceded and followed by a period without motion (i.e., with zero velocity) for a reasonable amount of time, such as a single finger flexion or flexion-extension cycle [1],[2]. Continuous movements lack such recognizable endpoints, and normally are considered rhythmic if they constitute repetitions of particular events, in which case they often look quite sinusoidal. While it is trivial that discrete movements can be repeated periodically, the question whether motor behavior is fundamentally discrete or rhythmic is not. Is motor behavior fundamentally discrete, reducing rhythmic movement to mere concatenations of discrete movements [3],[4]? Or is motor control fundamentally rhythmic, in which case discrete movements are merely ‘aborted’ cycles of rhythmic movements [5]–[7]? Alternatively, both types of movements may belong to distinct classes that are irreducible to each other [8]–[10], hence implying the utilization of different movement generating mechanisms.
Proponents of the ‘discrete perspective’ have sought evidence for discrete movement control through the identification of movement segments in movement trajectories. However, segmented motion need not imply segmented control [11]. In fact, the possibility to settle the dispute (solely) on the basis of kinematic features of movement (movement time, peak velocity, symmetry of velocity profiles, etc.) has recently been questioned [12]. Other researchers have aimed to identify the neural structures associated with discrete and rhythmic movements. For instance, Schaal and colleagues [9] showed that the brain areas that were associated with rhythmic movements were approximately a subset of those that were active during discrete movement execution. Differential involvement of neural subsystems does not provide a classification principle, however. Unambiguous classification requires the identification of invariance that is unique to each class so that the intersection of these two sets of characteristics is empty. Such a result will provide unambiguous evidence that two classes indeed are distinct. Dynamic systems theory offers such a classification principle based on phase flow topologies, which identify all behavioral possibilities within a class. Its significance lies in the fact that the classification is model-independent; every behavior within a class can be mapped upon others, whereas maps between classes do not exist. We use this principled approach to address the controversy whether discrete and rhythmic movements are fundamentally different. To that aim, we introduce the notion of phase flow topologies, identify the invariance separating two movement classes, and present an experimental study testifying to the existence of (at least) two different movement classes.
Deterministic, time-continuous and autonomous systems can be unambiguously described through their flow in state (or phase) space, defined as the space spanned by the system's position x and velocity (under the commonly adopted assumption that the deterministic component of movement trajectories can be fully described by two state variables). Whereas the phase flow quantitatively describes the system's evolution as a function of its current state (x, ); the system's qualitative behavior is solely determined by its phase flow topology. From the Poincaré-Bendixson theorem [13],[14] it follows that the only possible topologies in two dimensional systems are composed of elements referred to as fixed points, limit cycles, and separatrices. A fixed point of the system identifies a rest state (i.e., rate of change is zero, ), and, if stable, all trajectories in phase space eventually converge to it (Figure 1A). A system located at a fixed point can only depart from it in the presence of an external stimulation. A separatrix is a subset of points in the phase space that divides locally distinct phase flows (Figure 1A and 1B). In most cases for two-dimensional phase spaces, a separatrix is a line from which the flow points away in approximately opposite directions. Even simpler, for one-dimensional phase spaces any unstable fixed point is a separatrix. Limit cycles (Figure 1C) are closed loops in a two-dimensional phase space. If a limit cycle is stable, then all trajectories converge to it. A system on a limit cycle will repetitively traverse the same trajectory in phase space and sustain a periodic motion. Since these elements, fixed points and limit cycles, compose all phase flows in two dimensions, we associate discrete and rhythmic movements with these. The Hartman-Grobman theorem [13],[14] states that the flow in the local neighborhood of a fixed point is topologically equivalent to that of its linearization, which implies that a continuous invertible mapping (a homeomorphism) between both local phase spaces exists. From these theorems it follows that dynamical systems belong to the same class if, and only if, they are topologically equivalent. Therefore, movements that can be shown to be governed by fixed point dynamics versus movements governed by limit cycle dynamics are not reducible to each other, and as such we can make the strong claim that they are from different equivalence classes.
In consideration of the notion of topological equivalence, Jirsa and Kelso [15] recently formulated a generic model construct that allows for a stable fixed point and a separatrix (referred to as the mono-stable regime) or a stable limit cycle regime (Figure 1) in its corresponding phase space (see Text S1). These topologies correspond to single (i.e., discrete) flexion-extension movements and rhythmic movement, respectively. This perspective has three crucial features. First, the qualitative behavior in each regime is model independent. Second, each single movement execution in the mono-stable regime depends on an external triggering (mathematically speaking, the system is non-autonomous). In contrast, in the (autonomous) limit cycle regime no external stimulation is required and movement is self-sustaining. Third, the phase flow underlying movement is invariant on the time scale of the movement in both cases. Here, we examine this perspective by directly investigating numerically generated phase flows as well as those generated by humans and show that discrete and continuous movements belong to distinct dynamical classes.
We computationally examined the generic model under a large parameter and frequency range in order to examine the robustness and limits of its behavior in both dynamical regimes (see Materials and Methods). In the limit cycle regime, the timing requirement (i.e., the computationally implemented movement frequency) was met under all movement paces (i.e., frequencies). In contrast, in the mono-stable regime the actual timing deviated from the required timing due to a period-doubling when the movement pace exceeded approximately 2.0 Hz. (Figure 2A), which occurs due to the arrival of stimulus n before movement n−1 has finished. These observations were robust under all parameter settings within each dynamical regime (see Text S1 and Figures S1, S2, and S3), although the frequency at which the period doubling occurred showed a small variation as a function of one of the model parameters. In fact, while the exact frequency at which stimulus – movement interference occurs will show little variation as a function of the specific model realization (i.e., through function g1 and g2; see Equation 1 in Text S1), its occurrence with increasing frequency of stimulation is unavoidable. By implication, every discrete movement system has an upper (frequency) limit in generating sequential movements.
In the behavioral experiment human participants (n = 8) performed an auditory-paced unimanual finger flexion-extension timing task under similar movement paces (from 0.5 Hz to 3.5 Hz; step size 0.5 Hz) that were presented in ascending or descending order (see Materials and Methods). The participants were instructed to synchronize their full flexion with the metronome under three instruction conditions: to move as fast as possible (with staccato like movements being initiated to end/start a cycle), as smooth as possible (move so that the finger is continuously moving during the movement period interval) or without any specific instruction. We refer to these conditions as ‘discrete’, ‘smooth’, and ‘natural’, respectively (Figure 2B). Please note that, notwithstanding the repetitiveness of the movements, these instructions may elicit movements generated by distinct control mechanisms but do not prescribe the latter.
We reconstruct the vector fields underlying the phase flow (see Figure 3 and Materials and Methods) using a novel technique [16],[17] that has been successfully tested on simulated data from dynamical systems [18],[19] and applied in fields like (among others) physics [16],[17], engineering [20], economics [21], and which was recently introduced in the study of human movement [19],[22],[23]. In addition, we investigate the phase spaces in terms of two-dimensional probability distributions and performed more ‘traditional’ kinematic analysis commonly utilized in the (human) movement sciences (see Text S1 and Figures S4, S5, S6, S7, S8, and S9). Figure 3 represents the vector fields (Figure 3A, 3B, 3D, 3E) from five trials of a single participant and the corresponding angle diagrams (Figure 3C and 3F, respectively), and clearly indicates the existence of a fixed point (Figure 3A–3C) and a limit cycle (Figure 3D–3E). Figure 4A–4C (upper row for each subfigure) shows the angle diagrams averaged across all participants for each frequency and instruction condition. Obviously, the averaging across participants, to some extent, smears out the representation of the topological structures, as indicated by the standard deviations across participants of the angle reconstructions in the lower rows of Figure 4A to 4C. Regardless, the existence of a single fixed point at slow movement paces in the discrete condition, indicating the utilization of the mono-stable regime dynamics, can be appreciated from Figure 4A (upper row). In the natural and smooth condition the vector fields are less structured at slow paces, especially at 0.5 Hz (Figure 4A–4C). Scattered (to some degree) vector fields and the existence of either one or two fixed points appear at 0.5 Hz in the smooth and the natural condition. The fixed point(s) appears clearer at 1.0 Hz to 2.0 Hz in both conditions. Under all instruction conditions, however, the fixed point(s) vanishes at high movement paces and invariantly gives way to limit cycle dynamics (Figure 4A–4C). These results indicate that humans utilize distinct timing mechanisms in a movement pace-dependent manner.
What are the implications of these finding? First and foremost, our results lay the foundation of a motor behavior classification scheme based on mathematical theorems. We demonstrated that discrete and fast rhythmic movements constitute distinct classes; their genesis is, by implication, underwritten by different mechanisms. Fast rhythmic movements are autonomous and their timing emerges from the movement dynamics. In contrast, discrete movements are non-autonomous: Their timed execution cannot originate from their dynamics and hence requires external time keeping, most likely arising from a neural structure or network that is not implicated in the implementation of the dynamics. In that regard, the discrete movements studied here constituted full, repetitive (flexion-extension) cycles. Similar movements are sometimes referred to as continuous movements in the presence of temporal events [24],[25]. We refer to them as ‘discrete’ as they are governed by fixed point dynamics. Regardless, please note that even though in many cases the exact timing of a discrete movement is hardly of importance, every discrete movement initiation (be it embedded in a regular or irregular sequence of movements or not) requires ‘external’ stimulation, which is ultimately timed. This also holds for an additional class of discrete movements, namely, point-to-point movements (cf. [9]), in which two stable fixed points exist simultaneously (see Supporting Information, and [15]). While our findings are by and large in line with the more ‘traditional’ and purely behaviorally-defined classification [2] as well as recent versions thereof in terms of movement continuity [24],[25], they also identify their limitation; continuous movements do not constitute a single class. This limitation indeed strengthens our call for a classification of movement rooted in mathematical theory that bears directly on the mechanisms underlying movement genesis.
The movements at a slow pace, in particular at 0.5 Hz, under the ‘smooth’ instruction (and for some participants under the ‘natural’ instruction) were invariantly characterized by (relatively) irregular phase flows (see Figure 4C). The Poincaré-Bendixson theorem [13],[14] rules out topological structures other than fixed points (and separatrices) and limit cycles in two-dimensional phase space. The (relatively) irregular phase flows (with indices of multiple fixed points) may (by hypothesis) represent movements whose phase flow changes on a similar time scale as the movement. Such flows can be predicted for equilibrium point models [4]–[6] that, from a dynamical perspective, can be interpreted in terms of (the relocation of) a fixed point [26]. In fact, phase flow changes on the time scale of the movement also underwrite an alternative dynamical model [7]. Accordingly, discrete movements are accounted for by the destabilization and subsequent stabilization of fixed points interspersed by a time interval in which a limit cycle exists that effectively generates the (discrete) movement. The destabilization is accounted for by an external impact relative to the dynamics (‘behavioral information’). In other words, discrete movement generation is non-autonomous according to this account also.
The notion of time keepers versus timing resulting from movement dynamics are not new. On the contrary, these notions are central to two distinct theoretical camps (the information processing perspective and dynamical system approach, respectively) that have little interaction ([27]; and see e.g., the special issue of Brain & Cognition 48, 2002). The notion of a time keeper (or central timer) became firmly established by the well-known two-level timing model [28],[29]. Accordingly, the behavioral expression in tapping movements – the often observed negative correlation between consecutive tapping intervals – is the resultant of the repetitive movement initiation by a central time keeper and the impact of the motor delays preceding and following each particular tap (which are all random variables). Notwithstanding the various elaborations of (‘cognitive’) timing models ever since [30]–[33], the notion of time keeping is inherently connected with abstract mental representations. In contrast, eschewing representational concepts, dynamicists view timing and coordination as properties arising from (self-organized) pattern formation processes [34]–[37]. Here, we elaborated on two distinct dynamical organizations and report evidence that humans ‘implement’ either of these depending on movement rate. In the non-autonomous scenario movement initiation (and thus timing) depends on a mechanism external to the dynamics. While we framed this in terms of time keeping, this should not be taken to imply that we adhere to a representational account thereof (cf. [36]). In other words, the non-autonomous case should not be simply equated with a dynamical version of a two-level model (notwithstanding the – to some extent superficial – similarity in terms of a distinction between ‘clock’ and ‘motor’ components).
The implication of external timekeeper during discrete movements begs the question what neural structure(s) could fulfill this function? Spencer and colleagues [25] showed that patients with cerebellar lesions have deficits in producing discontinuous but not continuous movements, which supports the idea that the cerebellum is implicated in timing in the non-autonomous but not autonomous case (see also [38]–[40]). However, Schaal and colleagues [9], using fMRI, reported contralateral activity in several non-primary motor areas and the cerebellum during discrete wrist movements that was absent during their rhythmic counterparts. This result favors the suggestion that timing is a property originating from a distributed neural network [41],[42]. Indeed, the neural basis underlying timing remains yet to be elucidated. Implementing the present paradigm in the context of brain imaging may help establishing that aim.
Finally, it has been repeatedly suggested that motor control is simplified through the use of ‘motor primitives’, the motor system's elements thought of as its ‘building blocks’. The modular organization of the vertebrae spinal motor system and the reproducibility of specifically coordinated muscle activity upon stimulation of certain modules (neural circuits) instigated the idea that motor behavior is organized along such hard-wired structures [43]–[45]. On a more abstract level, the two timing architectures we identified here qualify as candidate building blocks in human motor control.
We numerically investigate the equationin which a and b, and γ, represent parameters, ω represents the system's eigenfrequency, τ represent a time constant, and I the external stimulation. For all simulations we use τ = 1, and if applicable, a stimulus duration corresponding to 80 ms and magnitude of 3.5.
For the mono-stable regime, the following parameter settings are implemented: γ = 1; ω = 1; a = [1.01, 1.09] with steps of 0.02; b = [−0.1, 0.8] with steps of 0.1; and I = [0.25 Hz, 4.00 Hz] with steps of 0.25 Hz. For the limit cycle regime, the implemented parameters are: a = 0; b = [−0.2, 0.3] with steps of 0.1; and ω = [0.25 Hz, 4.00 Hz] with steps of 0.25 Hz. For each frequency ω, γ is chosen to as to ensure that the system oscillates with the appropriate frequency. All simulations are performed using a fourth-order Runge-Kutta method. Gaussian white noise ξ(t) is added to the evolution equations of the y-variable, where 〈ξ(t)〉 = 0, 〈ξ(t)ξ(t)〉 = Q2δ(t−τ), Q = 0.01. The triangular brackets 〈·〉 denote time averages.
Eight participants (mean age = 27.9 years) took part in the experiment. Seven participants were (self-reported) right-handed, one participant was left-handed. Participants reported an average of 2.75 years of musical experience with a minimum of 0 years and a maximum of 8 years. The protocol was approved by the Purdue University Committee on the Usage of Human Research Participants and was in agreement with the Declaration of Helsinki. Informed consent was obtained from all participants.
Data were collected using a Polhemus Liberty-8 receiver (23×13×11 mm, 4 gm) that was affixed to the participant's index finger with adhesive tape. This receiver was controlled by Matlab using an AuSIM-AuTrakMatlab USB driver and collection interface via library C++ calls. Three dimensional position data were collected at 240 Hz. The motion in the medio-lateral direction was used for further analysis.
The flexion-extension movements were performed in the transverse plane involving no physical contact with any object. During the performance, the participants were seated at a 77-cm high table, and each participant rested the medial portion of his or her hand on a padded wooden block and Velcro held their hand in place. Ten trials were performed under three instruction conditions. Under each instruction condition, the participant was instructed to time the full finger flexion with the occurrence of the metronome tone. The instruction for the ‘natural’ condition was to do so in a manner that felt most natural. The instruction for the ‘smooth’ condition was to execute the movements as smooth (sinusoidal) as possible so as to be moving always ‘at an even pace’. For the ‘discrete’ condition the instruction was to execute each complete flexion and extension movement as quickly as possible. In each condition five trials were performed with increasing metronome pace (from 0.5 Hz to 3.5 Hz; step size 0.5 Hz) and five trials with decreasing pace. Every frequency plateau lasted for 15 tones. Participants were instructed to quickly and smoothly adjust to changes in pace. A 30 second rest interval was provided between trials. Feedback was given after a trial if the participant's average cycle duration for any of the seven metronome paces had deviated more than 15 percent of the goal interval duration. The order of increasing or decreasing set of trials was performed in a blocked design. All participants performed the first condition (‘natural’) on day one. The order of the other two conditions was balanced for all participants. Each session lasted approximately one and a half hour.
Human movement is inherently stochastic; its dynamics constitutes a deterministic and a stochastic (i.e., random) component [19],[34],[35]. The future state of a stochastic process is conditional upon the probability for its state to be at a given time instant at a specific point in phase space, which can be described by probability distributions [34],[46]. The computation of probability distributions allows one to disentangle the deterministic and stochastic dynamical components underlying stochastic processes [16]–[19]. Here, we extract these components to focus on the deterministic dynamics. Thereto, for each trial, we computed the movement velocity and normalized all position (x) and velocity (y) time-series to the interval [−1, 1]. Next, using a grid size of 31, we computed for all trials the conditional probability matrix, P(x,y,t|x0,y0,t0), that is, the probability to find the systems at state (x,y) at a time t given its state (x0,y0) an earlier time step t0. Subsequently, we computed the Kramers-Moyal coefficients [16]–[20] representing the drift coefficient according to
The coefficients Dx and Dy were averaged across the five trial repetitions for each participant, instruction condition and movement frequency. From the first two coefficients (that represent the x-, and y-component of the corresponding velocity vector), we computed for each bin the angle θ between its corresponding velocity vector and that of each of its neighbors (provided their existence) according toin which u and v represent two neighboring vectors defined by Dx(x,y) and Dy(x,y) at position x and y in phase space. Next, we extracted the maximal value of θ in phase space. The existence of locally opposing vectors (i.e., with an angle of approximately 180°) indicate the existence of a fixed point. We then computed for each instruction condition×movement frequency condition the mean and standard deviation of the maximal angle across participants and frequency order. |
10.1371/journal.ppat.1002649 | In Vivo Suppression of HIV by Antigen Specific T Cells Derived from Engineered Hematopoietic Stem Cells | The HIV-specific cytotoxic T lymphocyte (CTL) response is a critical component in controlling viral replication in vivo, but ultimately fails in its ability to eradicate the virus. Our intent in these studies is to develop ways to enhance and restore the HIV-specific CTL response to allow long-term viral suppression or viral clearance. In our approach, we sought to genetically manipulate human hematopoietic stem cells (HSCs) such that they differentiate into mature CTL that will kill HIV infected cells. To perform this, we molecularly cloned an HIV-specific T cell receptor (TCR) from CD8+ T cells that specifically targets an epitope of the HIV-1 Gag protein. This TCR was then used to genetically transduce HSCs. These HSCs were then introduced into a humanized mouse containing human fetal liver, fetal thymus, and hematopoietic progenitor cells, and were allowed to differentiate into mature human CD8+ CTL. We found human, HIV-specific CTL in multiple tissues in the mouse. Thus, genetic modification of human HSCs with a cloned TCR allows proper differentiation of the cells to occur in vivo, and these cells migrate to multiple anatomic sites, mimicking what is seen in humans. To determine if the presence of the transgenic, HIV-specific TCR has an effect on suppressing HIV replication, we infected with HIV-1 mice expressing the transgenic HIV-specific TCR and, separately, mice expressing a non-specific control TCR. We observed significant suppression of HIV replication in multiple organs in the mice expressing the HIV-specific TCR as compared to control, indicating that the presence of genetically modified HIV-specific CTL can form a functional antiviral response in vivo. These results strongly suggest that stem cell based gene therapy may be a feasible approach in the treatment of chronic viral infections and provide a foundation towards the development of this type of strategy.
| There is a desperate need for the development of new therapeutic strategies to eradicate HIV infection. HIV actively subverts the potent natural immune responses against it, particularly cellular cytotoxic T lymphocyte (CTL) responses. The development of a therapy that allows long-lived immune self-containment of HIV and restoration of these CTL responses by the host would be ideal. Through genetic manipulation of human blood-forming stem cells, we introduced a molecule– an HIV-targeting T cell receptor (TCR)–that allowed the generation of functional HIV-specific CTLs following differentiation within human tissues in a humanized mouse model. To assess if these newly developed, HIV-specific CTLs can allow active suppression of HIV replication, we infected these mice with HIV. We found that the development of genetically modified, HIV-specific CTLs in these mice results in the presence of a functional antiviral CTL response in vivo that significantly lowers viral replication following HIV infection. These results have strong implications for the use of this technology to engineer the human immune response to combat viral infections and suggest that genetic engineering via HSCs may allow tailoring of the immune response to target and eradicate HIV.
| Human hematopoietic stem cells (HSCs), through development in the thymus, are capable of producing progeny T cells that generally display one of a vast repertoire of T cell receptors (TCRs). In the case of many non-persistent viral infections, T cells bearing TCRs specific to viral antigens mediate a potent antiviral response that results in the clearance of the virus from the body. Even in the presence of most persistent viral infections, a potent T cell response is mounted; however it often fails to clear the virus from the body. A critical component of the T cell antiviral response is the CD8+ cytotoxic T lymphocyte (CTL), whose primary function is to recognize viral antigens (in the context of human leukocyte antigen class I (HLA I)) and kill virus-infected cells. In HIV infection, the potent antiviral CTL response is critical for establishing relative control of viral replication during the acute and chronic infection stages of the disease [1]–[6]. However, unlike what is observed in most non-persistent viral infections, the CTL response fails to clear HIV from the body. The magnitude, breadth, functional quality, and kinetics of the antiviral CTL response all are critical in controlling ongoing viral replication; however, the reasons for the failure to rid the body of virus are not completely understood [7], [8]. Ongoing viral replication and viral evolution in the infected host is one important, although highly confounding, factor in the persistence of HIV in chronic infection [4], [5]. Even under effective antiretroviral therapy (ART), the virus is not cleared from the body and the level of HIV specific CTLs declines, likely due to lower levels of antigen to stimulate the persistence/generation of these cells [9], [10]. Due to the importance of T cell responses in controlling and eliminating viral infection there exists a great need to explore ways to enhance antiviral T cell immune responses.
Recently, much of attention in HIV research has focused on ways to enhance or correct the defects in HIV-specific CTL responses. Gene therapy-based approaches that augment immunity towards viral antigens represent unique, yet largely unexplored, strategies towards the treatment of HIV disease. We have previously examined the feasibility of a stem cell-based gene therapy approach to enhance cell-mediated immunity towards chronic HIV infection. In these studies, we demonstrated that human HSCs genetically modified with genes encoding a human HIV-specific TCR can produce mature, fully functional T cells in human thymus implants in severe combined immunodeficient (SCID) mice. The resulting genetically directed CD8+ T cells are capable of killing HIV antigen-expressing cells ex vivo [11]. Further, we showed that the appropriate restricting human leukocyte antigen (HLA) class I molecule is required for proper development of transgenic TCR-containing CTLs. In all, our earlier studies demonstrated that TCR-modified human HSCs can be directed to develop into mature CTLs in a human thymus environment in the context of the proper HLA type. However, as the SCID-hu mouse model demonstrated poor peripheral reconstitution and function of human immune cells, these studies did not address the ability of these cells to suppress HIV replication in vivo.
In the present studies, we examined the ability of genetically modified T cells derived from HSC transduced with a single HIV-specific TCR to suppress viral replication in vivo. We utilized a modified version of a newly established humanized mouse model, the non-obese diabetic (NOD)-SCID, common gamma chain knockout (γc−/−), humanized bone marrow, fetal liver and thymus (the NSG-BLT) mouse model, which allows the generation of peripheral human immune responses, and serves as an effective model for HIV infection and pathogenesis [12]–[14] (see Figure 1A). These humanized mice display multilineage human hematopoiesis and systemic engraftment of peripheral organs with human blood cell types including T lineage cells, B lineage cells, myeloid lineage cells, NK cells, as well as cells from other lineages [12] (and see Figure 1B). We modified human hematopoietic stem cells in this model with molecularly cloned genes corresponding to a TCR specific to the HIV-1 Gag 77–85 SLYNTVATL (SL9) epitope to allow the production of mature HIV-specific CTLs in multiple organs of these reconstituted mice. We determined that human T cells that expressed the HIV-specific TCR were capable of suppressing HIV replication in vivo and preventing or slowing viral damage to the engrafted human immune system. These studies establish a system to examine “genetic vaccination” approaches that target chronic viral infection and to more closely examine mechanisms of human antiviral immunity in vivo.
We have previously demonstrated that human hematopoietic stem cells can be genetically modified by delivering a gene for an HIV-specific TCR, and develop into mature T cells in an HLA-restricted fashion in the human thymus of SCID-hu mice [11]. These newly produced, SL9 gag antigen-specific, naive T cells were determined to be capable of producing IFN-γ in response to peptide stimulation and were found to be lytic to SL9 peptide loaded target cells. However, it was not known whether these genetically modified HIV-specific CTLs could traffic to relevant organs in the mice and whether they were capable of killing HIV infected cells in vivo. To address this question, we established an improved model, based on the NSG-BLT model previously shown to allow HIV replication [15], [16], as a surrogate system to assess the antiviral efficacy of engineered, HIV-specific T cells in vivo. NSG mice were implanted with human fetal liver-derived CD34+ HSCs that had been modified with a lentiviral vector containing the genes for a TCR targeting the HIV Gag SL9 epitope, or as a control, with HSCs modified with a lentiviral vector containing a non-HIV-specific TCR with unknown specificity. In addition, these mice received implantation of human fetal Thymus and Liver under the kidney capsule to facilitate human T cell development. Hence, we term this the NSG-CTL model (Figure 1A).
As genetic manipulation of HSCs is required in this model, we initially determined the effects on this type of lentiviral transduction on multilineage hematopoietic potential of HSCs in the humanized mice. Phenotypic markers of human hematopoiesis were examined by flow cytometry in mice within 6 weeks following implantation of human tissues. One hundred percent of the mice receiving human tissue had human cells in the peripheral blood, including myeloid, natural killer (NK), T cell, and B cell lineages (Figure 1B). In these mice, the average percentage of human CD45+ cells in the peripheral blood was 53% of the total cells (with a standard deviation of 29% and a range of 19%–80%, n = 12). We more closely examined the bone marrow in these mice for the presence of human cell engraftment, particularly human HSC engraftment. We found a significant population of human CD34+ HSCs in the bone marrow (Figure 1C). The majority of these cells coexpressed the CD45 molecule, which is indicative of cells with lymphopoietic potential [17]. In addition, there were significant populations of both CD3 expressing T cells and CD19 expressing B cells in the bone marrow of these mice. This indicated that multilineage human hematopoiesis occurs in these mice and provides evidence that, in addition to T cells, other components of the human immune system are present. These data demonstrated that our modification of the NSG-BLT humanized mouse utilizing genetically modified human hematopoietic stem cells does not negatively affect human hematopoiesis.
We then examined the animals for the presence of cells expressing the transgenic, HIV specific TCR by MHC tetramer staining. We found CD3+ T cells expressing the transgenic TCR in all organs assessed, including the bone marrow, thymus, spleen, liver, and peripheral blood of the mice receiving transduced human hematopoietic stem cells (Figure 1D). Thus, we have observed long-term, multilineage human immune reconstitution and the development of mature T cells that express the transgenic, HIV-specific TCR in multiple organs in the NSG-CTL mouse.
To assess if peripheral cells resultant from human hematopoietic stem cells that expressed the recombinant SL9-specific TCR were capable of suppressing HIV replication in vivo, NSG-CTL mice containing the HIV specific TCR or a control TCR were infected with HIV-1NL-HSA-HA. HIV-1NL-HSA-HA is an engineered variant of HIV-1NL4-3 that contains the murine heat stable antigen (HSA) reporter gene modified to contain an Influenza hemagglutinin (HA) antibody epitope, which is cloned into the open reading frame of the vpr gene to allow detection of HIV-infected cells by cell surface detection of HA expression using flow cytometry [18]. Peripheral blood was assessed for the level of productively infected cells two and six weeks post infection. Within 2 weeks post infection, we observed a reduced level of productively infected cells in mice containing the HIV-specific TCR versus mice containing the control TCR (Figure 2A). In addition, there was less initial CD4 depletion in mice containing the HIV-specific TCR versus mice containing the control TCR. Within six weeks post infection, while there was an overall increase in virus-expressing cells from the earlier time point, we observed a marked reduction in productively infected cells in mice containing the HIV specific TCR versus the control TCR, indicating suppression of viral replication over time (Figure 2B). At this time point, mice containing cells expressing the HIV-specific TCR had a greater preservation of CD4+ T cells and higher CD4 to CD8 T cell ratios when compared to mice expressing the control TCR. Amongst all mice in the experiment, there was no statistically significant difference 2 weeks following infection with either CD4 cell count or with the percentage of cells expressing HIV, however there was a trend towards better preservation in CD4+ cell numbers as well as lower levels of virus-expressing cells in mice containing the HIV-specific TCR (Figure 3). However, by 6 weeks post-infection, there was a statistically significant difference in CD4 cell numbers and levels of infected cells between mice with cells expressing the HIV-specific TCR and mice with cells expressing the control TCR. Thus, genetic modification of HSCs with a single HIV-specific TCR produces peripheral T cells capable of suppressing cellular HIV expression and CD4 depletion in vivo.
We next sought to determine if cells modified with an HIV-specific TCR could suppress virus levels in peripheral blood plasma. However, quantitating plasma viremia in the mouse model is difficult due to the amount of plasma obtained per blood draw (typically ∼50 microliters), the limit of detection obtainable with this amount of blood, and the high cost associated with commercial assays. Therefore to measure viremia in this system, we developed a novel quantitative PCR-based technique for HIV in mouse plasma. Based on the recently elucidated secondary structure of the HIV genome [19], primers were designed to specifically target relatively “open” regions of the RNA genome that contain minimal secondary structure to attempt to allow increased sensitivity to detect viral RNA. Utilizing this technique, which has a reliable sensitivity of 5 copies of HIV RNA per sample, we determined that the viral load 2 weeks and 6 weeks post infection was significantly lowered in mice receiving the HIV-specific TCR versus mice receiving cells transduced with the control TCR (Figure 4A). This suggested systemic suppression of HIV replication in vivo. Surprisingly, analysis of the viral RNA for mutations in the SL9 epitope did not reveal the presence of any mutations in this epitope in the majority quasispecies, which was identical in comparison to the sequence of the input virus and the virus of infected mice containing the non-specific TCR control (Figure 4B). This suggested that in this period of time, viral escape to the selective pressure of the SL9 specific TCR had not occurred in the blood of these mice, possibly due to limited viral replication in this model. Thus, there was significant suppression of viral replication in vivo in mice expressing the HIV-specific TCR versus the control TCR and this suppression did not result in significant viral escape within 6 weeks following infection.
As illustrated in Figure 1, T cells expressing transgenic HIV-specific TCRs were found in multiple organs in mice receiving genetically modified HSCs. Based on this, we next addressed suppression of HIV in multiple organs in the lymphoid compartment in mice containing cells expressing the HIV-specific TCR. NSG-CTL mice that had received HSCs transduced with the HIV SL9-specific TCR or, separately, the non-specific control TCR were infected with HIV-1NL-HSA-HA. Sets of infected animals were then assessed 2 weeks and 6 weeks post infection for the quantity of HIV proviral DNA sequences in human cells in the spleen, bone marrow, and human thymus implant (Figure 5). We observed significant suppression of HIV replication in human cells in these organs as early as 2 weeks post infection (in the bone marrow) in mice receiving HSC containing the HIV-specific TCR. 6 weeks post-infection, HIV levels were significantly lower in the spleen, bone marrow, and human thymus implant in animals receiving the HIV-specific TCR as compared to mice receiving the control TCR. In addition, analysis for proviral DNA in human cells in the pooled peripheral blood cells (n = 3 mice per treatment group), revealed a similar trend, with 37 copies and 529 copies of HIV per 10,000 human cells at 2 weeks and 6 weeks post infection respectively, in mice containing the HIV-specific TCR, and 356 copies and 792 copies of HIV per 10,000 human cells at 2 weeks and 6 weeks, respectively, in mice containing the control TCR. Thus, these data indicate that there is significant suppression of HIV in multiple lymphoid tissues in animals receiving HSCs genetically modified to produce cells that specifically target HIV infected cells.
We assessed the antiviral effector function of CTLs expressing HIV-specific transgenic TCRs in mice receiving genetically modified HSCs. In an additional series of experiments, mice containing the SL-9 specific TCR were infected with HIV or left uninfected and cells from the peripheral blood were assessed for phenotypic changes that would suggest differentiation. HIV infection resulted in phenotypic differentiation of HIV specific cells, as determined by SL9 MHC tetramer staining, into cells possessing an effector phenotype [20], [21](CD8+SL9Tetramer+CD45RA-CCR7-)(Figure 6A). This was similar to the phenotypic changes we observed in previous studies following ex vivo peptide stimulation of SL9-specific, TCR transgenic thymocytes [11] and in vivo responses to the MART-1 tumor antigen by MART-specific CD8 cells [22]. This increased loss of CD45RA and CCR7 expression that we observed in HIV- specific TCR-expressing cells in infected mice versus uninfected mice is indicative of antigen-specific induction of cellular differentiation. We then more closely analyzed the differences we observed viral suppression by and expansion of HIV-specific CTLs in vivo in infected mice. We found a significant correlation between the highest levels of reconstitution of HIV-specific TCR-expressing cells prior to infection and more effective suppression of viral loads in the serum six weeks following infection (Figure 6B). Interestingly, we noted that at six weeks following infection, mice that had greater levels of HIV-specific TCR-expressing cells in the peripheral blood had higher viral loads at this time point (Figure 6C). In addition, we saw significant antigen-driven expansion of HIV-specific TCR-expressing CTLs in infected animals compared to controls, with the greatest levels of expansion seen in animals with the lowest initial (week -2) transgenic TCR reconstitution (Figure 6D). Taken together, these results suggest that greater initial reconstitution of transgenic HIV-specific cells is more effective at controlling early viral replication. Furthermore these data suggest that the higher resultant viral loads in animals with initially low human immune reconstitution drive greater antigen-specific cell expansion over time. Thus, CTLs expressing the HIV-specific TCR undergo antigen-driven phenotypic differentiation and expansion in this model, which correlates with control of viral replication.
The CTL response has a pivotal role in controlling HIV replication in infected individuals. While HIV generates a potent natural immune response during the acute stage of infection, this response does not result in the control of viral replication or clearance of the virus from the body [4]–[6]. There are critical defects in the CTL response that result during chronic viral infection. These defects include the inadequate generation of a functional response due to low antigen-specific precursor frequency, expression of functional inhibitory molecules such as programmed death-1 (PD-1) and T-cell immunoglobulin domain and mucin domain 3 (TIM-3), and Cytotoxic T-Lymphocyte Antigen 4 (CTLA-4), and activation of suppressor cell activity [23]–[26]. In addition, HIV can directly or indirectly perturb viral antigen presentation, immunoregulatory cytokine production, T cell differentiation and effector/memory generation, and can infect CTLs themselves [27]–[33]. The maintenance of a potent antiviral CTL response is critical in all stages of infection and there are strong associations between the preservation of CTL responses specific for more conserved HIV epitopes, greater control of viral replication, and slower disease progression [5], [6].
In the present study, we demonstrate the feasibility of engineering human hematopoietic stem cells to become peripheral T cells capable of targeting HIV replication in vivo. Our previous studies provided evidence that the genetic modification of human hematopoietic stem cells with a lentiviral vector containing an antigen-specific TCR (specific to the SL9 Gag epitope) allowed the development of functional human T cells in human thymus implants in SCID-hu mice [11]. While this study demonstrated that transgenic TCR-containing T cells are capable of developing in the human thymus, the ability of these cells to target and kill HIV infected cells in vivo was not known. In the present study, we use an improved chimeric mouse model exhibiting a high degree of human immune cell reconstitution to significantly extend these observations to demonstrate that mature T cells expressing an antigen-specific human TCR are capable of developing and migrating to peripheral organs in vivo. In contrast to the SCID-hu Thy/Liv model, which is an excellent model for studies examining human thymopoiesis but limited in examining peripheral immune responses [34], we utilized a variation of the humanized BLT mouse model utilizing the NSG strain that allows multilineage hematopoiesis and human cell repopulation in peripheral organs [35], [36]. The generation of natural immune responses to HIV in these systems appears to be relatively limited, particularly the ability of these mice to elicit HIV specific human T cell responses which is likely due to incomplete human immune cell reconstitution, particularly antigen presenting cell reconstitution, to the levels seen in humans [12], [36], [37]. In addition, lower antigen-specific cell precursor frequency and the lack of or lower levels of human-specific cellular support immune components (such as costimulatory or immunoregulatory molecules, adhesion molecules, and cytokines) likely contribute to the lower levels of antiviral immune responses generated in humanized mice. The incomplete and varied immune reconstitution in the current humanized mouse systems results in differences in immune responses and kinetics of viral pathogenesis compared to natural HIV infection in humans. The reasons for this are unclear and vary between the different types of humanized mouse models; however, there are many similarities and parallels between HIV infection in humanized mice and humans which makes these surrogate models very powerful in their ability to allow the close examination of many aspects of HIV infection, transmission, pathogenesis, immunity, and therapeutic intervention [36]. While natural antiviral T cell immune responses are limited in current humanized mouse models, our studies suggest that the genetic “programming” of HSCs to produce T cells specific for HIV can overcome this limitation in this system and produce measurable T cell responses that have a significant antiviral effect in vivo. Further, we found it startling that the use of a single HIV-specific TCR can result in significant HIV suppression while natural suppressive antiviral CTL responses are polyclonal. These observations can provide the platform for future studies that allow the closer examination of the generation of human antiviral immune responses and the identification of factors involved in the persistence and potential eradication of HIV infection.
Previous attempts utilizing a gene therapy approach towards enhancing antigen specific cellular immune responses have focused on “redirecting” mature T cells towards viral or cellular antigens [38]–[46]. In these cases, genes for HIV-specific T cell receptors (TCRs) or chimeric antigen specific receptors were utilized to modify mature T cells to specifically target virus infected cells or malignancies. In some cases of the latter, tumor regression has occurred in treated individuals [45]–[47], which suggests that the genetic modification of T cells towards a specific antigen is feasible in vivo in humans and alludes to the potential for the further development of these strategies to target other diseases. However, the modification of mature T cells has several limitations, including the possibility of endogenous TCR miss-pairing with the newly introduced TCR, the development of intrinsic functional defects and/or the alteration of cellular effector/memory maturation pathways in the cells following heavy ex-vivo manipulation [47], and the maintenance of long-lived fully functional cells. A stem cell-based approach where HSCs are modified with an antigen specific receptor, however, may abrogate these complications by allowing the long term, continual natural development of mature T cells that bear the transgenic antigen-specific molecule. Normal development of these cells in the bone marrow and selection in the thymus would reduce the possibility of producing cells that are autoreactive through TCR miss-pairing and functionally altered through ex vivo manipulation, major drawbacks of mature T cell modification. We have recently shown that genetic modification of human HSC with a TCR specific for human melanoma allows the generation of melanoma-specific human T cells capable of clearing tumors in BLT mice [48]. Our current studies extend this type of approach to demonstrate the in vivo efficacy of TCR-modified stem cells to generate antigen-specific T cells that target a rapidly replicating viral infection in vivo. Our results document the ability of the resulting HIV-specific CTLs to dramatically reduce viral replication and consequent CD4 cell loss in a relevant model of HIV pathogenesis.
Recent stem cell-based attempts at protecting cells from direct infection by HIV through the modification of HSCs with antiviral genes or genes that knock down viral coreceptors [16], [49], [50] require high percentages of HSCs to be genetically modified to be protected from infection. Our results suggest that the approach of genetically vaccinating cells to target HIV infection would require much lower levels of genetic modification. Modification of human HSCs with a transduced TCR results in significantly increased naïve, antigen specific precursors. This level of transduction is sufficient to result in decreased viral replication and increased immune protection. Correspondingly in humans, uninfected HLA-A*0201+ individuals have an estimated natural SL9 epitope-specific, naïve CTL precursor frequency of approximately one in 3.3×106 cells in the peripheral blood, which is similar to the precursor frequency of naïve cells specific to a variety of other viral antigens [51]. In our studies, the TCR-transduced population typically accounted for 0.75–5.5% of the CD8+ T cell population in a given organ in the mouse following their differentiation from HSCs (the illustration in Figure 1D represents a single mouse from a single experiment). The frequency of transgenic cell reconstitution did not correlate with transduction efficiencies of the vector into the stem cells, rather it appears to be due to individual engraftment rates of CD34+ cells into each mouse. However, even at low frequencies of transgenic TCR expressing cells, this represents a significant increase in the naïve cell precursor frequency for cells specific to the SL9 Gag epitope, as mice harboring the control non-specific TCR and untransduced mice had undetectable levels or very low levels of natural SL9-specific CTLs as determined by MHC tetramer staining. Utilizing TCR gene transductions to yield increases of HIV-specific precursor frequency to conserved antigenic epitopes could potentially reconstitute innate defects in the ability of peripheral T cells to clear infected cells. While the human thymus involutes over time, thus producing fewer T cells in adults than in children, it does retain some activity throughout life [52], [53]. A recent study involving introduction of an antiviral gene into the autologous HSC of HIV infected adults illustrated that naïve T cells bearing the transgene were detected in the peripheral blood of these subjects, indicating that genetically engineered T cells can develop from HSC in adult HIV infected subjects [54]. Through this type of therapeutic intervention, our results suggest the feasibility of supplying newly developed, naïve antigen-reactive cells, that could allow the overall T cell response to overcome limits in the magnitude of the response that inhibit effective viral clearance.
This type of gene therapy-based approach could further diversify the breadth of the responses by naïve, antigen specific cells by utilizing TCRs specific to other epitopes of HIV. The use TCR gene transduction as a therapeutic approach would have to be tailored to the HLA type of the individual receiving treatment in order to produce cells that survive T cell selection processes. Immune epitope escape from the transduced TCR, which did not occur in the time frame of our experiments, is likely to occur in vivo in a clinical setting. One potential caveat of the humanized mouse model is the lower level of human immune cell reconstitution than is seen in humans; which significantly, yet incompletely, recapitulated the human immune system in the mouse. While HIV replication rates and viral loads persist detectably over weeks, they do not achieve the levels observed in natural infection in humans. This lower level of viral replication is one potential reason that viral escape mutants to the SL9-specific TCR may be slower to develop. The potential for viral immune escape necessitates the use of multiple TCRs in a therapeutic setting targeted to the antigen or antigens of interest. Careful selection of multiple TCRs targeted to relatively conserved antigenic epitopes within defined HLA types could reduce the possibility of viral epitope evolution and immune escape, perhaps driving the evolution of the virus into a less fit state [55]. The evidence that immune escape and viral evolution against many specific epitopes occurs relatively slowly suggests that an engineered immune response and the immune pressure created by these antigen-specific cells may be therapeutically beneficial by lowing viral replication, decreasing levels of infected cells, and impairing the fitness state of the virus [55], [56]. In sum, our results demonstrate the feasibility of a therapeutic approach that involves the modification of human HSCs by delivering genes for antigen-specific TCR to produce peripheral, naïve, antigen-specific T cells that are capable of reducing HIV replication in vivo. These studies provide a foundation and a model system that would allow the closer examination of human antiviral T cell responses and the development of therapeutic strategies that target chronic viral infection.
Peripheral blood mononulear cells was obtained at the University of California, Los Angeles in accordance with UCLA Institutional Review Board (IRB) approved protocols under written informed consent using an IRB-approved written consent form by the UCLA Center for AIDS Research Virology Laboratory and Dr. Yang and was distributed for this study without personal identifying information. Human fetal tissue was purchased from Advanced Biosciences Resources or from StemExpress and was obtained without identifying information and did not require IRB approval for its use. Animal research described in this manuscript was performed under the written approval of the UCLA Animal Research Committee (ARC) in accordance to all federal, state, and local guidelines. Specifically, these studies were carried out under strict accordance to the guidelines in The Guide for the Care and Use of Laboratory Animals of the National Institutes of Health and the accreditation and guidelines of the Association for the Assessment and Accreditation of Laboratory Animal Care (AALAC) International under UCLA ARC Protocol Number 2010-038-02B. All surgeries were performed under ketamine/xylazine and isofluorane anesthesia and all efforts were made to minimize animal pain and discomfort.
The following antibodies were used in flow cytometry: CD3, CD4, CD11c, CD8, CD45, CD45RA, CD34, HLA-DR (Coulter), CD19, CD56, CCR7, HSA, and IgG controls (eBioscience), hemagglutinin (HA) sequence YPYDVPDYA (Roche), and HLA-A*02 (Serotech). HLA-A*0201 tetramer containing the HIV Gag SL9 SLYNTVATL (SL9) peptide was purchased from Coulter. Cell surface marker expression was analyzed utilizing antibodies conjugated to either fluorescein isothiocyanate (FITC), Peridinin Chlorophyll Protein (PerCP)-Cy5.5, phycoerythrin (PE), electron coupled dye (ECD), PE-Cy5, PE-Cy7, allophycocyanin (APC), APC-Alexa750, APC-eFluor780, Alexa700, eFluor405, Pacific Blue, or Pacific Orange in appropriate combinations. Cells were acquired on a LSR II flow cytometer (BD Biosciences) using FACSDiva software. FlowJo software was used for analysis.
Lentiviral production from the plasmid containing the HIV SL9 specific TCR (pCCL.PPT.hPGK.1.9.IRES.eGFP) or a control TCR with an unknown specificity (pCCL.PPT.hPGK.α4.IRES.eGFP) was produced using the Invitrogen ViraPower Lentiviral Expression System using the pCMV.ΔR8.2.Δvpr packaging plasmid and the pCMV-VSV-G envelope protein plasmid as previously described [11].
NSG mice were initially purchased from Jackson Laboratories and bred at the UCLA Division of Laboratory Animal Medicine. To construct NSG-CTL mice, fresh human HLA-A*0201+ fetal liver and thymus pairs from the same donor were obtained from Advanced Biosciences Resources or from StemExpress. Fetal liver was then homogenized and CD34+ cells were isolated as described [11]. Briefly, fetal liver is diced into small (∼3 mm) pieces, homogenized and digested with collagenase type IV (1 mg/ml), hyaluronidase (1 mg/ml), DNase I (2 U/ml)(Sigma). CD34+ cell were purified using magnetic activated cell sorting (Miltenyi). The negative fraction of cells, which contains fetal liver stromal cells (CD34− cells) is saved. CD34+ cells were then genetically transduced following resuspension in Yssel's medium containing 2% human serum albumin and placed in a tissue culture plate coated with 20 µg/ml retronectin (Takara Bio, Inc.) along with lentiviral vector at a multiplicity of infection of 5 overnight at 37°. Fetal liver stromal cells and the matched fetal thymus, cut into small pieces (2 mm), were cultured at 37° overnight in RPMI-1640 containing 10% fetal calf serum (FCS) and 0.44 mg/ml Piptazo. The next day, tissue and cells were washed with PBS and a fraction of the transduced CD34+ cells were then viably frozen. The remaining CD34+ cells were combined with fetal liver stromal cells in cold Matrigel (BD Biosciences) in a 1∶9 ratio (CD34+ cells:fetal liver stromal cells, typically 500,000 transduced CD34+ cells:4,500,00 fetal liver stromal cells) and combined with a 2 mm fetal thymus piece in a trocar and placed under the kidney capsule of NSG mice. Transduction efficiency was determined following culturing in IMDM containing 20%FCS, 50 ng/ml of IL-3, IL-6, and SCF for 3 days, and subsequent assessment of GFP fluorescence by flow cytometry. Transduction efficiency of CD34+ cells occurred at a mean rate of 12.7% (Standard deviation = 12.6%, range 1.63%–38.5%, n = 14). Three weeks following implantation, mice were irradiated with 3 Gy using a cobalt-60 source to clear a niche for human CD34+ cell engraftment of the bone marrow. The frozen CD34+ cells were then thawed and then injected intraveneously into the mice. Mice were then checked for human cell engraftment 6–10 weeks post-injection. Multiple experiments were performed with a minimum of 3 mice per experimental group to yield statistical significance. Each experiment utilized humanized mice that were made from human tissue from same donor and the donor tissue was unique experiment to experiment.
A virus variant of HIV-1NL4-3-HSA-HA containing the mouse heat stable antigen (HSA) which has been modified to contain the influenza virus hemaggluttin YPYDVPDYA sequence (HA) cloned into the vpr open reading frame, and which also contains the SL9 Gag epitope, has been previously described [18]. Virus was grown in CEMx174 from plasmid-derived virus stock. Viral infectivity was determined by limiting dilution titration on CEMx174 cells. Mice were infected by intraperitoneal injection 10–12 weeks following CD34+ cell injection with 50–100 ng of previously frozen virus stock.
Mouse blood peripheral blood was drawn by retro-orbital bleeding into glass capillary tubes coated with 330 mM EDTA (Gibco), and 3% sterile human serum albumin (Baxter Healthcare). Viral RNA was extracted from plasma with the High Pure Viral RNA Kit (Roche). The kit is designed to extract 200 µl of plasma. Since there is generally less plasma than this, the volume was estimated by weight and brought up to 200 µl with phosphate buffered saline (Gibco). DNA standards and the template for in vitro transcribed RNA for quantitative PCR was derived from pNL101 linearized with EcoRI, checked with electrophoresis, and quantitated by spectrophotometry (A260). A section of the gag gene in pNL101 was amplified with the primers NG1CF, position 366–398 (5′-GGAGAATTAGATAAATGGGAAAAAATTCGGTTA-3′) and NG1CR position 679–648 (5′-GCCTTTTTCTTACTTTTGTTTTGCTCTTCCTC-3′), and cloned into pCR4TOPO. The product containing the cloned gag was then digested with SpeI and BsrGI and gel purified. This fragment was then translated to RNA with T7 RNA polymerase (Promega Riboprobe Transcription Kit) and quantitated by spectrophotometry (A260). This RNA was serially diluted in The RNA Storage Buffer (Ambion) with 0.4 U/µl Rnasin and 5 ng/µl Lambda DNA/HindIII (as carrier), to make a stock of 500,000 copies/µl. Before each RT run, a fresh vial of RNA was serially diluted in the RNA Storage Buffer (Ambion) to make standards of 100,000 to 10 copies. Quantitative reverse transcriptase-PCR (RT-PCR) was performed using the following primers/probe specific for gag sequences: NG1F (position 453–480) 28 bp (5′-GAGCTAGAACGATTCGCAGTTAATCCTG-3′), NG1R (position 570–534) 37 bp (5′-ATAATGATCTAAGTTCTTCTGATCCTGTCTGAAGGGA-3), NG1Z probe (position 482–520) 39 bp (FAM-5′ -CCTTTTAGAGACATCAGAAGGCTGTAGACAAATACTGGG-3-BHQ).
The final reaction concentration consists of 2.5 µM NG1F, 7.5 µM NG1R, and 2.5 µM NG1Z. These primers were based on sequences identified to be in relatively “open” regions of HIV RNA not impeded by secondary structure interference as determined by [19].
Reverse transcription was performed using the SuperScript III kit (Invitrogen). The annealing step consisted of 5 µl of template RNA plus 3 µl of a mixture consisting of 1.5 parts 20 µM NG1R, 0.5 parts Rnasin plus (Promega), 16 parts 5× RT buffer, and 12 parts water. The resulting 8 µl was heated to 70° for 2 minutes, then at 60° for 5 minutes, then cooled to room temperature. The RT step was run by adding 2 µl of a mixture of 8 parts water, 4 parts 5× buffer, 5 parts DTT, 2 parts 25 mM dNTPs (Invitrogen), and 1 part SuperScript III. This was heated to 55° for 30 minutes, 85° for 5 minutes, then cooled to room temperature.
For quantitative DNA PCR following the reverse transcription step, 15 µl of the PCR mix consisting of 38.5 parts water, 44 parts 25 mM MgCl2, 50 parts NG-FRZ oligos, 5 parts 500 mM Tris buffer pH 8.3, 8.5 parts 1 M KCl, 2.5 parts 25 mM dNTPs, and 1.25 parts Platinum Taq was added to all wells that underwent the reverse transcription reaction and mixed. Real-time, quantitative PCR was performed with 5 minutes activation at 95°, and followed by 45 cycles of 95° for 15 seconds and 60° for one minute on a BioRad CFX96 thermocycler. An additional set of DNA standards, serially diluted from 2×105 copies to 20 copies, of linearized pNL101 was run in parallel to control for the efficiency of the RT step. Results from samples were interpolated within the quantitation derived from the RNA standards.
Statistical support was provided through the UCLA Center for AIDS Research (CFAR) Biostatistical Core. Experiments were analyzed utilizing the Student's t test, the Spearman rank correlation test (SRCT), or the Wilcoxon Rank Sum Test (WRST)(when n = 3), as indicated.
Evolution/mutation of the dominant version of the introduced Gag-SL9 epitope sequences from the plasma of mice infected with HIV-1NL4-3-HSA-HA was determined by bulk sequencing of the segment of the Gag coding region. Plasma viral RNA was isolated as described above and cDNA was synthesized utilizing the Superscript cDNA synthesis kit (Applied Biosystems). Alternatively, proviral DNA from lymphocytes on infected mice was isolated as described above. Utilizing these DNAs, the region of the HIV-1 Gag flanking the SL9 epitope (a.a. 77–85) was PCR amplified using the 737-Forward primer (5′-GCGACTGGTGAGTACGCC-3′) and the 1255-Reverse primer (5′-ACCCATGCATTTAAAGTTC-3′) and purified by Gel-purification (Qiagen Inc., USA). This purified bulk PCR product was then directly used for dye-terminator sequencing with both 737-Forward and 1255-Reverese primers in parallel. The data was then analyzed by the ABI-3130 genetic analyzer (Applied Biosystems, USA).
|
10.1371/journal.pgen.1004461 | Cis and Trans Effects of Human Genomic Variants on Gene Expression | Gene expression is a heritable cellular phenotype that defines the function of a cell and can lead to diseases in case of misregulation. In order to detect genetic variations affecting gene expression, we performed association analysis of single nucleotide polymorphisms (SNPs) and copy number variants (CNVs) with gene expression measured in 869 lymphoblastoid cell lines of the Avon Longitudinal Study of Parents and Children (ALSPAC) cohort in cis and in trans. We discovered that 3,534 genes (false discovery rate (FDR) = 5%) are affected by an expression quantitative trait locus (eQTL) in cis and 48 genes are affected in trans. We observed that CNVs are more likely to be eQTLs than SNPs. In addition, we found that variants associated to complex traits and diseases are enriched for trans-eQTLs and that trans-eQTLs are enriched for cis-eQTLs. As a variant affecting both a gene in cis and in trans suggests that the cis gene is functionally linked to the trans gene expression, we looked specifically for trans effects of cis-eQTLs. We discovered that 26 cis-eQTLs are associated to 92 genes in trans with the cis-eQTLs of the transcriptions factors BATF3 and HMX2 affecting the most genes. We then explored if the variation of the level of expression of the cis genes were causally affecting the level of expression of the trans genes and discovered several causal relationships between variation in the level of expression of the cis gene and variation of the level of expression of the trans gene. This analysis shows that a large sample size allows the discovery of secondary effects of human variations on gene expression that can be used to construct short directed gene regulatory networks.
| Humans differ in their genetic sequences at millions of positions but only a subset of these differences have a functional effect. In order to detect functional genetic differences, we assessed the impact of common genetic variants on gene expression in 869 individuals and discovered that the expression of many genes is affected by common variants in cis or in trans. We show that the effect of some variants on gene expression cannot be detected in other tissues, highlighting the tissue specificity of gene regulation. In addition, we show that variants associated to common diseases are more likely to affect gene expression in cis and in trans. Finally, we show that variants affecting gene expression in cis often affect gene expression in trans, which suggests that the trans effects are due to the cis genes expression. We tested this hypothesis and discovered several cases of genes regulated in trans by a cis regulated gene in a causal manner. This shows that a population-based strategy with a large number of individuals has the potential to detect secondary effects of common variants that can be used to construct short directed regulatory networks.
| Genome-wide association studies (GWAS) have discovered a large number of loci implicated in many complex traits and diseases [1]. The vast majority of variants discovered are found in non-coding regions (88%), which challenges the interpretation of their functional effect [1]. One way to overcome this challenge is to look for associations between variants and an intermediate cellular phenotype, such as gene expression. Expression quantitative trait loci (eQTL) analysis have been successful in mapping variants to gene expression in several cell types providing a better understanding of the genetics of gene expression, and revealing functional impacts of variants associated with complex traits and diseases [2]–[10].
Most studies so far were conducted on relatively small sample sizes [11], limiting the power to detect variants affecting gene expression in cis and to a greater extent in trans, as trans-eQTLs typically have weaker effect sizes than cis-eQTLs [12]. Detecting eQTLs with small effect sizes is indeed important, as a variant can have a weak effect in the tissue sampled but a strong effect in the tissue relevant for a specific disease. Furthermore, small effects in cis can be important if the associated gene plays a substantial role in a cellular process. A large sample size also allows the discovery of variants that are both cis and trans-eQTLs, suggestive of a regulatory relationship between the cis regulated gene and the trans regulated gene.
Here we measured gene expression in Lymphoblastoid cell lines (LCLs) from 869 genotyped individuals of the Avon Longitudinal Study of Parents and Children (ALSPAC) cohort in order to map single nucleotide polymorphisms (SNPs) and copy number variants (CNV) with minor allele frequency >1% to gene expression in cis and in trans. We then investigated trans effects of cis-eQTLs and used causal models to investigate the mechanism by which a variant can affect the expression of a gene in cis and in trans.
In order to better understand how genetic variations affect gene expression in LCLs, we associated 2'290'057 imputed single-nucleotide polymorphisms (SNPs) from the HapMap2 reference set and 3329 copy number variants (CNVs) to gene expression in 869 individuals from the ALSPAC cohort. We first looked for cis-eQTLs, defined as variants associated with gene expression in a 2 MB window surrounding the transcriptional start site (TSS). We used spearman rank correlation to test for association between 20'745 probes on autosomes, measuring the expression of 14'835 genes, and variants present in the windows. A gene-based significance threshold was determined by permuting all expression phenotypes 1000 times (Methods). We discovered that 3534 genes had a cis-eQTL at a false discovery rate (FDR) of 5% (3498 due to SNPs, 36 due to CNVs) (Table S1, Table S2). As shown previously [5], [9], we found that cis-eQTLs cluster around the TSS (Figure S1). CNVs were more often cis-eQTLs than expected by chance (i.e: if SNPs and CNVs had the same probability to be a cis-eQTL, we would expect the same fraction of CNVs and SNPs in our cis-eQTL results than in the whole data set) (odd ratio: 7.1, Fisher's exact test pvalue<2.2e-16), suggesting that CNVs are more likely to affect gene expression than SNPs. We observed a strong correlation (rho = 0.98, pvalue = 1e-13) between sample size and the number of genes with a cis-eQTL discovered (Figure S2). Although, it appears that the rate of discovery is decreasing for large sample sizes, it is likely that further increases in sample size would allow the discovery of more genes with at least one cis-eQTL.
In order to detect genes affected by more than one cis-eQTL, we performed conditional regression by removing the effect of the cis-eQTL(s) on gene expression and repeating the association analysis on the residuals until no significant associations could be detected (Methods) [13], [14]. We found that 694 genes (19.6% of the genes with a cis-eQTL) have at least two independent cis-eQTLs (Figure S3). For 374 genes (53.9%), independent cis-eQTLs were located in the same recombination interval, showing that variants in linkage disequilibrium can have different functional effects on gene expression. In order to evaluate the importance of independent cis-eQTLs on the heritability of gene expression, we obtained heritability estimates from the MUTHER cohort in LCLs [2]. On average, the heritability of genes with multiple cis-eQTLs was greater than for genes with only one cis-eQTL detected (mean heritability = 0.24 for genes with one cis-eQTL, mean heritability = 0.38 for genes with multiple eQTLs, Mann-Whitney U test pvalue<2.2e-16). For each gene with multiple eQTLs, we compared the heritability explained by the best eQTL to the heritability explained by all independent eQTLs using a linear model where the standard normal expression of the gene is explained by the best eQTL or all independent eQTLs (Methods). We observed that the best eQTL explains on average 46% of the heritability of the traits while all independent eQTLs explain on average 57% of the heritability of gene expression (Figure 1). These results show that independent cis-eQTLs are detected preferentially in genes with a relatively high genetic component of their expression and that independent cis-eQTLs explain 11% more of the heritability of the gene expression on average than using only the best cis-eQTL.
Several studies have shown that the effect of variants on gene expression is tissue dependent [5], [9], [10], [15]. Indeed, some eQTLs can have different effect sizes in cells of different developmental origin [9] or can be detected only in specific tissues [5], [9], [10], [15]. However, the estimated tissue sharing of eQTLs has steadily increased in function of the cohort sample sizes, ranging from 20%–31% [5] with a small cohort to 56–83% in a larger cohort [2], questioning the relevance of interrogating different tissues. In order to address this question, we took advantage of the large sample size of the ALSPAC cohort to investigate the effect of sample size on tissue sharing. We obtained cis-eQTLs detected in LCLs, skin and adipose tissues from the MUTHER cohort [2], one of the largest studies investigating eQTL tissue specificity. We assessed tissue sharing as a function of sample size in a continuous manner by matching cis-eQTLs detected by the MUTHER study (1%FDR) with the pvalues detected for the same associations in different subsets of individuals of the ALSPAC cohort and computed the π1 statistic (estimate of the proportion of true positives in a pvalue distribution) for each sample size [16]. We observed little tissue sharing for small sample sizes (30.6% with adipose tissue, 34.8% with skin, and replicated 46.9% in LCLs for 50 individuals) (Figure 2). In contrast, using the entire ALSPAC cohort, we replicated 79.2% of the eQTLs in LCLs, detected by the MUTHER project, and estimated tissue sharing to be 61.6% for adipose tissue and 61.7% for skin cells. We did not observe an increase in the π1s for LCL and skin cells after 600 individuals, while the sharing of MUTHER adipose eQTLs with ALSPAC LCLs continued to increase slightly for larger sample sizes (Figure 2). We found stronger concordance in the directionality of eQTLs replicating within LCLs (1.8% with opposite directionality at 5% FDR) compared to eQTLs shared across tissues (10.4% with opposite directionality in skin and 10.5% in adipose tissue at 5%FDR). These results indicate that a relatively large proportion of cis-eQTLs detected in one tissue cannot be detected in other tissues and support the idea that one should perform eQTL analysis in different tissues in order to map all regulatory variants in the genome.
We next investigated whether we could detect variants affecting gene expression in trans. We defined trans-eQTLs as variants affecting gene expression at a distance greater than 5 MB from the TSS or on another chromosome. We used spearman rank correlation to test for association between 21'634 probes, measuring the expression of 14'441 genes on autosomes and chromosome X, and all variants further than 5 MB from their TSS. A genome-wide significance threshold was determined by permuting a subset of the expression phenotypes 1000 times (Methods). We discarded trans associations of CNVs when the gene associated was on the same chromosome or on a chromosome with SNPs correlated with the CNV (r2>0.1) as some CNVs appeared to be mismapped (Text S1). In addition, we discarded all significant associations of CNVs with an imbalance in copy number between males and females, as this resulted in the false trans associations of the CNVs with genes differentially expressed between males and females. After filtering, we discovered trans-eQTLs for 48 genes (FDR = 5%) (45 due to SNPs and 3 to CNVs) (Table 1, Table S2, Table S3). We assessed the replication of the trans-eQTLs using the MUTHER cohort [2], a twin cohort, which we separated in two sets of unrelated individuals (group 1: 340 individuals, group 2: 338 individuals). We replicated 22 trans-eQTLs (of 40 tested) with a pvalue<0.05 in the first group and 19 in the second group (union = 23). In order to validate the array-based trans-eQTLs with an independent technology, we used the GEUVADIS [17] cohort (373 individuals with RNA-seq in LCLs) and replicated 11 trans-eQTLs (of 32 tested). As for cis-eQTLs, CNVs were more often trans-eQTLs than expected by chance (i.e: if SNPs and CNVs had the same probability to be a trans-eQTL, we would expect the same fraction of CNVs and SNPs in our trans-eQTL results than in the whole data set) (odd ratio: 45.8, Fisher's exact test pvalue = 5e-5). Two variants, rs1156058 and rs705170, were associated with a total of 14 and 7 genes in trans respectively (Figure S4). We also found that rs4781011, located on chromosome 16 within 5 kb of the TSS of the gene CIITA (class II, major histocompatibility complex transactivator), a gene known to activate in trans the HLA locus on chromosome 6, was a trans-eQTLs of CD74 on chromosome 5, a protein that regulates antigen presentation. This analysis shows that it is much more difficult to detect trans-eQTLs than cis-eQTLs at the same false discovery rate. Although our replication cohorts had a sample size representing only roughly 40% of the discovery cohort, we replicated approximately 50% of the trans-eQTLs attempted. This encouraging result suggests that more trans-eQTLs could be replicated with a bigger replication cohort and that our trans-eQTLs detection methodology is efficient.
Genome-wide association studies (GWAS) found many SNPs associated with diverse phenotypes but the mechanistic link between the GWAS-SNP and the phenotype remains unclear for the vast majority of the associated SNPs. One possibility is that a GWAS-SNP affects gene expression, which then leads to the phenotype. It was previously shown that trait associated SNPs were more likely to be cis-eQTLs [18]. However, since the publication of this result, many more GWAS were performed, increasing dramatically the number of variants associated with complex traits and a much larger number of eQTLs were discovered in this study. In order to confirm that GWAS identified variants are more likely to be cis-eQTLs and to investigate if a similar relationship exists for trans-eQTLs, we accessed the catalog of published genome-wide association studies (http://www.genome.gov/gwastudies/) on 19 March 2012. 5381 SNPs reported in the catalog at that date were genotyped in our study. We looked for GWAS-SNPs overlapping eQTLs and found that 850/3 (15.8%/0.06% of the GWAS-SNPs) GWAS-SNP co-localized with variants significantly associated in cis/trans (Table S5) (Table S6). This is significantly more than using SNPs matched to the GWAS-SNPs for distance to closest gene and minor allele frequency (for cis-eQTLs, median = 585, pvalue<0.001) (for trans-eQTLs, median = 0, pvalue<0.01) (Figure S5). This confirms that many GWAS-SNPs are probably playing a role on disease susceptibility by affecting gene expression in cis and that trait associated SNPs are also more likely to be trans-eQTLs [19].
We next sought to determine whether trans-eQTLs were also cis-eQTLs, as this may indicate that the genes regulated in cis play a role in the regulation of the trans genes. We examined the overlap between trans-eQTLs and cis-eQTLs and found that 5 (18.5%) of the unique trans-eQTLs were also associated with gene expression in cis. This overlap is significantly greater than the overlap obtained using variants matched to the trans-eQTLs for distance to closest gene and minor allele frequency (1000 permutations, median = 0, pvalue<0.001) (Figure S6). We find that the cis-eQTLs of two transcription factors, BATF3 and HMX2, are associated to the most genes in trans. The cis-eQTL of BATF3, a gene involved in the differentiation of CD8α+ dendritic cells and IL17-producing T helper cells [20], [21], is a trans-eQTL of 14 genes, distributed on 8 chromosomes. The cis-eQTL of HMX2 is a trans-eQTL of 7 genes distributed on 4 chromosomes. HMX2 is a transcription factor directing development of inner ear and hypothalamus in mice [22] and deletion of the chromosomal region containing HMX2 in human is associated to inner ear malformations, vestibular dysfunction and hearing loss [23]. Other genes with a cis-eQTL that is also a trans-eQTL are: GNA15, a G protein, S1PR4, a G protein coupled receptor, PIDD, an effector of p53 apoptosis in mice and CRIPAK, an inhibitor of the PAK1 transcription factor. These results show that we can detect potential new functional targets of important genes in LCLs by combining cis-eQTLs and trans-eQTLs.
In order to detect more possible functional relationships between genes regulated in cis and in trans by the same variants, we looked for the trans effects of the subset of variants that were found to be cis-eQTLs. Since we discovered 3475 variants associated to 3534 genes in cis, a trans-analysis of this subset of variants has the advantage of reducing the number of tests performed and therefore allows us to discover more trans effects of cis-eQTLs. Before investigating which cis-eQTLs are affecting which genes in trans, we first aimed to assess how many trans effects of cis-eQTLs could be detected if we had a much larger sample size. We used spearman rank correlation to test for associations between 23'935 probes, measuring the expression of 16'505 genes and all unique cis-eQTLs further than 5 MB from the TSS. We obtained approximately 23'935 trans association pvalues per cis-eQTL and computed the π1 statistics (estimate of the proportion of true positives in a pvalue distribution) on each set of pvalues, resulting in 3475 π1 estimates [16]. These estimates represent the proportion of probes that are affected in trans by the 3475 variants that are cis-eQTLs, without being able to pinpoint all individually significant effects. We observed that a large number of cis-eQTLs are affecting a large number of probes in trans (52% of the cis-eQTLs have a π1 >0) ranging from a few probes affected to up to 37.2% of the probes (median = 0.004603 corresponding to 110 probes) (Figure 3A). Interestingly, the variant with the most trans effects (37.2% of the probes), rs482519, is the cis-eQTL of WHSC1 (Wolf-Hirschhorn Syndrome candidate1), a histone methyltransferase. A potential explanation for this result it that variation of the level of expression of this histone methyltransferase could affect the expression of many genes by modifying chromatin accessibility. The second variant with the most trans effects (33.5% of the probes), rs2978387, is the cis-eQTL of ZNF16 (Zinc Finger Protein 16), a protein that may act as a transcription factor. The third variant with the most trans effects (32.1% of the probes), rs12196956, is the cis-eQTL of TBC1D22B (TBC1 domain family member 22B), a protein that may act as a GTPase-activating protein for Rab family protein. Furthermore, we observed a negative correlation between the strength of the cis-eQTLs and the number of probes affected in trans (spearman rho = −0.1, pvalue = 4.2e-10), suggesting that strong cis-eQTLs may be selected against in the population for genes modulating the expression of many genes. These results show that cis-eQTLs can have trans effects on many genes, which have direct consequences on regulatory network perturbations.
Although we estimated that a large number of cis-eQTLs are affecting many genes in trans, we would need a very large sample size to detect all of them at a reasonable false discovery rate. In order to assess which cis-eQTL is affecting which genes in trans, a genome-wide significance threshold was determined by permuting all expression phenotypes 1000 times (Methods). 92 genes had significant trans-effects due to cis-eQTLs (FDR = 5%) (Table S4). We replicated 31 associations (of 79 tested) in the first set of twins of the MUTHER cohort, 22 in the second set (union = 34) and 27 in GEUVADIS (of 75 tested). We discovered substantially more trans-effects of the cis-eQTLs of BATF3 and HMX2 with 39 and 18 genes regulated in trans respectively (Figure 3B). Other examples of cis-eQTLs with several significant trans associations include the cis-eQTL of PSMG1 (proteasome assembly chaperone 1) affecting 3 genes in trans and the cis-eQTL of BRWD1 (bromodomain and WD repeat-containing protein 1), which may be a transcriptional activator [24], also affecting 3 genes in trans. We did not find significant effects of the cis-eQTL of WHSC1, indicating that the large number of effects on gene expression have too small effect sizes to be discovered individually given our sample size. In total we found that 26 variants are cis-eQTLs of 27 genes and trans-eQTLs of 92 genes. 4 genes associated in cis to a cis/trans-eQTLs also had independent cis-eQTLs. We regressed out the effect of the main eQTLs on the trans genes expression and found that in 95% of the cases the independent eQTLs had the same allelic effect as the main eQTLs, i.e. the high expressing allele of the main eQTL has the same effect - high or low – in the trans gene as the high expressing allele of the second independent eQTL in 95% of the cases. This concordance further highlights the biological relevance of these trans eQTLs since their downstream biological effects, mediated by the modulation of the cis genes, are replicated by independent variants. 1 independents cis-eQTL (associated to HMX2) was also significantly associated to 1 gene in trans (5% FDR) and had the same allelic effect as the main eQTL. The strong concordance in allelic effects between main cis-eQTLs and independent cis-eQTLs indicate that for those 4 genes, most of the trans effects are due to variations in the level of expression of the cis gene.
We then explored whether the trans associations of the cis-eQTLs were causally due to the variation in the expression level of the cis genes. We assessed the likelihood of three different models using two methods: Bayesian networks and a causal inference test (CIT) (Methods) [25]. The first model (SCT) states that the variant is affecting the expression level of the cis gene, which then leads to variation in the level of expression of the trans gene. The second model (INDEP) states that the trans effect and the cis effect are independent and the third (STC) unlikely model states that the variant is affecting the level of expression of the trans gene, which then affects the level of expression of the cis gene. We observed that 100%/100%/94% of the SCT/STC/INDEP models detected by the CIT method are also detected as the best model by Bayesian networks. Conversely, 86%/33%/100% of the SCT/STC/INDEP models called by Bayesian networks were also detected as the best model by CIT. By taking the overlap of the two methods, we obtain 19 SCT, 2 STC and 49 INDEP relationships. We found causal effects (SCT) of CRIPAK on AVP, CCL5 on NPSR1, BATF3 on three genes and HMX2 on 14 genes (Table 2). The large representation of INDEP relationships is due to several factors. First, false positives will be called INDEP because their association is not due to the cis gene expression. As we found 92 trans associations of cis-eQTLs at a 5% FDR, we expect ∼5 INDEP relationships due to false positives. In addition, we expect 1 INDEP relationship because one cis-eQTLs is associated to two genes in cis and 2 genes in trans in total. It is unlikely that both of the cis associated genes would have causal effect on one trans associated gene leading to INDEP calls for 1 relationship. Finally, we observed that 34 INDEP associations are due to the cis-eQTL of BATF3. The INDEP relationships show that the trans gene associations are not due to the cis gene expression. However, they could be due to change in the structure of the protein if other functional variants, such as non-synonymous SNPs or splice variants, are in linkage disequilibrium with the cis-eQTLs. Alternatively, the cis-eQTLs could be affecting the expression of non-coding RNA in the vicinity of the cis genes that could drive the trans associations.
As we hypothesised that the trans-effects of some cis-eQTLs could be due to changes in the protein structure, we investigated the trans effects of 11564 non-synonymous SNPs discovered by the 1000 genome project and genotyped in the ALSPAC cohort. We used spearman rank correlation to test for associations between the 23'935 probes, measuring the expression of 16'505 genes and all non-synonymous SNPs further than 5 MB from the TSS. We first looked at large effects that could be detected given our sample size and found that 9 genes were affected in trans by non-synonymous SNPs (5% FDR) (Table S7). We replicated 4 associations (of 6 tested) in the first set of twin of the MUTHER cohort, 4 in the second set (union = 4) and 1 (of 7 tested) in GEUVADIS. We then looked at the global effects of non-synonymous SNPs on gene expression in trans by looking at the proportion of true positive in the distribution of the trans association pvalues for each variant using the π1 statistic [16]. We found that non-synonymous SNPs have significantly less trans effects on gene expression than cis-eQTLs (Mann-Whitney U test, pvalue = 1.8e-9) (Figure S7), as observed previously [4]. This result is compatible with the observation that common regulatory SNPs have more effects on complex traits and common diseases than common non-synonymous SNPs.
Finally, we explored the trans effects of the 5381 SNPs associated with complex traits and diseases in order to detect potential effects of these variants on gene expression and discovered that 66 of them are significantly associated to 10 genes in trans (5% FDR) (Table S8). We replicated 6 associations (of 7 tested) in the first set of twin of the MUTHER cohort, 6 in the second set (union = 6) and none in GEUVADIS (of 3 tested). For example, we found that rs11171739, which is associated to type 1 diabetes, is a trans-eQTL of DCAF16, as previously shown in monocytes [26] and is also a trans-eQTL of BEND4. We also found that rs4781011, which is associated to ulcerative colitis, is a trans-eQTL of CD74, a protein involved in immune response. rs2227139, which is associated to hematological parameters was associated in trans to ERG, which regulates hematopoiesis and the function of adult hematopoietic stem cells [27]. These results show that we can detect downstream effects of disease-associated variants, an important step to understand the relevant biological pathways in common diseases.
The large sample size of the ALSPAC cohort allowed us to discover that 3534 genes are affected by genetic variants in cis and 48 in trans. We found that CNVs are enriched in the best associations per gene in cis and to an even greater extent in trans. This enrichment is not surprising as CNVs are more likely to disrupt regulatory elements than SNPs due to their size [8]. This result indicates that CNVs are more likely to be causal than SNPs in genetic diseases resulting from the misregulation of gene expression. Several examples of genetic disorders, such as aniridia, sex-reversal and holo-prosencephaly are already known to be caused by duplications or deletions of CNVs located in non-coding regions of developmental genes [28]. We found that SNPs associated with complex traits and common diseases are more likely to be cis and trans-eQTLs than matched variants. Although some of these overlaps might be coincidental [3], these results further confirm that a significant fraction of trait associated SNPs are acting at the gene expression level.
We observed that many eQTLs detected in skin and adipose tissues could not be detected in LCLs irrespectively of the sample size, showing that a significant fraction of eQTLs is tissue specific. Therefore, eQTL studies in many different tissues are needed in order to map all regulatory variants in the human genome and understand their precise tissue specific effect, a necessary step to understand why a specific tissue becomes the “disease” tissue and not other tissues.
We estimated that 52% of the cis-eQTLs have trans-effects on gene expression ranging from a few probes to up to 37.2% of the probes. The large number of trans-effects of cis-eQTLs is concordant with the fact that on average 65% of the heritability of gene expression is trans to the gene in LCLs [2]. As we can detect only a minority of these effects at a reasonable false discovery rate with our relatively large sample size, it indicates that most of the trans-effects of the cis-eQTLs are of small effect sizes. If complex traits and common diseases have the same underlying architecture as gene expression, a substantial part of the missing heritability will then be due to many common variants of very small effect sizes.
Using Bayesian networks and causal inference tests [25], we could detect 19 cases where a variant affects the expression of a gene in cis that is causally affecting the expression of a gene in trans (14 due to the cis-eQTL of HMX2). For example, the level of expression of CRIPAK, a protein implicated in cytoskeleton remodelling and influencing PAK1 mediated estrogen transactivation activity [29], is causally affecting the level of expression of AVP (arginine vasopressin), a hormone with anti-diuretic effects on the kidney and affecting social behaviour [30]. Taken together, these results show that population based strategies allow to detect important relationships between genes. Ultimately, this type of approach performed with larger sample sizes will allow us to uncover the cascade of events that lead a disease associated variants to the disease phenotype.
Ethical approval for the study was obtained from the ALSPAC Ethics and Law Committee and the Local Research Ethics Committees.
ALSPAC is a prospective birth cohort which recruited pregnant women with expected delivery dates between April 1991 and December 1992 from Bristol UK. 14,541 pregnant women were initially enrolled with 14,062 children born. Detailed information on health and development of children and their parents were collected from regular clinic visits and completion of questionnaires. A detailed description of the cohort is available on our website (http://www.bristol.ac.uk/alspac/researchers/) and has been published previously [31]. Please note that the study website contains details of all the data that is available through a fully searchable data dictionary (http://www.bris.ac.uk/alspac/researchers/data-access/data-dictionary/).
DNA has been extracted as described previously from blood samples collected from cord blood at research clinics [32]. Lymphoblastoid cell lines were established by transforming lymphocytes from blood samples taken when the study participants were 9 years old, with Epstein Barr Virus.
ALSPAC individuals were genotyped using the Illumina HumanHap550 quad genome-wide SNP genotyping platform by 23andMe subcontracting the Wellcome Trust Sanger Institute, Cambridge, UK and the Laboratory Corporation of America, Burlington, NC, USA. Markers with <1% MAF, >5% missing genotypes or which failed an exact test of Hardy-Weinberg equilibrium (P<5×10−7) were excluded from further analysis. Any individuals who did not cluster with the CEU individuals in multidimensional scaling analysis, who had >3% missing data, minimal or excessive heterozygosity (>33% or <31%), evidence of cryptic relatedness (>10% IBD) or incorrect gender assignments were excluded from further analysis. After data cleaning 315,807 SNPs were left. Imputation was carried out using MACH 1.0.16, Markov Chain Haplotyping [33], [34], using CEPH individuals from phase 2 of the HapMap project as a reference set. Imputed markers with imputation quality r2<0.8, with MAF<1% or which failed an exact test of Hardy-Weinberg equilibrium (P<5×10−7) were excluded resulting in a total of 2'290'057 high quality SNPs. The CNVs were genotyped using a targeted Agilent 105K CGH array. The design of the array and the methodology for analyzing the array data was previously described in details [35].
LCL's from unrelated individuals were grown under identical conditions and cells frozen in RNAlater. RNA was extracted using an RNeasy extraction kit (Qiagen) and was amplified using the Illumina TotalPrep-96 RNA Amplification kit (Ambion). Expression profiling of the samples, each with two technical replicates, were performed using the Illumina Human HT-12 V3 BeadChips (Illumina Inc) including 48,804 probes where 200 ng of total RNA was processed according to the protocol supplied by Illumina. Raw data was imported to the Illumina Beadstudio software and probes with less than three beads present were excluded. Log2 - transformed expression signals were then normalized with quantile normalization of the replicates of each individual followed by quantile normalization across all individuals. We restricted our analysis to 23'935 probes tagging genes annotated in Ensembl. Principal component analysis was performed on 931 individuals. 62 individuals with principal component 1 or 2 greater than one standard deviation of the population were excluded from further analysis. Raw expression data are available upon request at http://www.bristol.ac.uk/alspac/researchers/data-access/policy/.
All eQTL analysis were performed at the single variant level and assumed an additive model. We used spearman rank correlation to test for association between probe expression and genotype. For the cis-analysis, we limited the variants tested to variants present in a 2 MB window surrounding the transcription start site of the gene and we filtered out probes containing SNPs with minor allele frequency >1% according to the 1000 genomes project dataset [36]. To assess significance, we permuted all expression probes 1000 times and kept the best pvalue per gene after each permutation. For each gene, we ranked the permutation pvalues and assessed whether a variant in the non-permuted data had a lower association pvalues than the permutation threshold considered. We then computed the false discovery rate associated with the permutation threshold and subsequently selected the permutation threshold that provides a 5% false discovery rate.
For the trans analysis, we tested all variants except variants present in a 5 MB window surrounding the transcription start site. In order to remove false positives, we excluded probes mapping to multiple locations according to ReMOAT [37]. To assess significance, we permuted 1000 times 288 random probes, each corresponding to one gene. As each probe is tested by approximately the same number of SNPs and as we used spearman rank correlation, which is robust to outliers, we treated our permutations as if we had permuted one probe 288'000 times. We combined all pvalues obtained from the permutations (288*1000), ranked them and increased the genome-wide pvalue threshold until we reached a 5% false discovery rate (corresponding to a pvalue of 9.5e-11).
For the trans analysis of cis-eQTLs, we tested all unique cis-eQTLs except variants present in a 5 MB window surrounding the TSS. In order to remove false positives, we excluded probes mapping to multiple locations according to ReMOAT [37]. To assess significance, we permuted all expression probes 1000 times. As for the trans analysis of all variants, we combined all pvalues obtained, ranked them and increased the genome-wide pvalue threshold until we reached a 5% false discovery rate (corresponding to a pvalue of 7.6e-8).
For the trans analysis of non-synonymous SNPs and SNPs associated to complex traits and diseases, we tested all SNPs except variants present in a 5 MB window surrounding the TSS. In order to remove false positives, we excluded probes mapping to multiple locations according to ReMOAT [37]. To assess significance, we permuted 1000 random probes, corresponding to 1000 genes, 10000 times. As for the other trans analysis, we combined all pvalues obtained, ranked them and increased the genome-wide pvalue threshold until we reached a 5% false discovery rate (corresponding to a pvalue of 2.0e-9 for non-synonymous SNPs and 5.4e-10 for SNPs associated to complex traits and diseases).
For each gene with an eQTL, we performed linear regression of the best variant on the standard normalized probe expression and kept the residuals. We repeated the association analysis on the residuals using spearman rank correlation and kept any SNPs passing the gene-based permutation threshold obtained during the initial association analysis. We repeated this procedure regressing out all previous best associations until no variants were significant.
For each gene with cis-eQTL(s), we computed the variance explained (r2) by the best cis-eQTLs or all independent cis-eQTLs on the standard normalized probe expression using the lm() function in R. We then obtained the heritability explained by dividing the heritability of the probe with the variance explained by the cis-eQTL(s). If the variance explained by the cis-eQTL(s) was greater than the heritability estimate of the probe, the heritability explained was set to 1.
We matched each significant variant (cis-eQTLs, trans-eQTLs or GWAS SNPs) with a variant with the same minor allele frequency in our data set (±1%) and distance to the closest gene (±2 kb).
Bayesian networks (BN) are directed acyclic graphs where nodes represent random variables and edges represent the conditional dependences among nodes. The direction of the edges between two nodes can be interpreted as causal relationships and allowed to infer causality in genetics studies previously [38]–[40].
Likelihood methods are commonly used to compare different BN and estimate the most likely—that is, the set of causal relationships among the different variables that better agrees with the data. In a BN, every node is associated with a probability distribution and, together with the conditional dependencies represented by the edges, forms the join probability distribution of the network. BN satisfy the local Markov property—that is, each variable is conditionally independent of its non-descendants given its parent variables. The Markov property allows the decomposition of the joint probability distribution of the network into a set of local distributions, which allows to easily calculate the likelihood of a given BN.
We used the R package bnlearn [41] to calculate the maximum likelihood of three different networks that we defined using eQTLs as anchors. In the first network (SCT), we fixed the first node as the eQTL genotype with a forward directional edge to the second node (standard normalized cis gene expression) and a second forward directional edge starting from the second node to the third node (standard normalized trans gene expression). For the second network (STC), we fixed the first node as the eQTL genotype with a forward directional edge to the second node (standard normalized trans gene expression) and a second forward directional edge starting from the second node to the third node (standard normalized cis gene expression). For the third network (INDEP), we fixed the first node as the eQTL genotype with a forward directional edge to a node representing the standard normalized cis gene expression and a second forward directional edge starting from the first node to a node representing the standard normalized trans gene expression.
Different networks often have different complexities and it is common to use a score that takes into account the network complexity instead of the raw likelihood to compare different networks. We used the Akaike Information Criterion (AIC) score (AIC = 2k-2ln(L), where k is the number of parameters (5 for all models in our case) and L is the maximum likelihood) to compare our networks. To compare how good is a network compared to another, we used the relative likelihood of one network against the other. If we have two networks, N1 and N2 and AIC(N1)≤AIC(N2), then the relative likelihood of N2 with respect to N1 is defined as: exp((AIC(N1)–AIC(N2))/2). We kept only networks where the best model was at least ten times more likely than the second best model. In order to have high confidences in our calls, we required that the Causal Inference Test (CIT), described previously [25], also calls the same model as the most likely. The CIT is a semi-parametric method that tests a series of conditions and then provides p-values for the SCT and STC models. If none of them has a pvalue<0.05, it makes a call for the INDEP model, and if both of them are significant it makes no call. In order to take into account multiple testing with the CIT method and to reduce the number of networks resulting in a “no call” by the CIT, we used Bonferroni corrected pvalues for model calling instead of the nominal pvalue of 0.05.
|
10.1371/journal.pgen.1000779 | Epigenetic Control of Virulence Gene Expression in Pseudomonas aeruginosa by a LysR-Type Transcription Regulator | Phenotypic variation within an isogenic bacterial population is thought to ensure the survival of a subset of cells in adverse conditions. The opportunistic pathogen Pseudomonas aeruginosa variably expresses several phenotypes, including antibiotic resistance, biofilm formation, and the production of CupA fimbriae. Here we describe a previously unidentified bistable switch in P. aeruginosa. This switch controls the expression of a diverse set of genes, including aprA, which encodes the secreted virulence factor alkaline protease. We present evidence that bistable expression of PA2432, herein named bexR (bistable expression regulator), which encodes a LysR-type transcription regulator, controls this switch. In particular, using DNA microarrays, quantitative RT–PCR analysis, chromatin immunoprecipitation, and reporter gene fusions, we identify genes directly under the control of BexR and show that these genes are bistably expressed. Furthermore, we show that bexR is itself bistably expressed and positively autoregulated. Finally, using single-cell analyses of a GFP reporter fusion, we present evidence that positive autoregulation of bexR is necessary for bistable expression of the BexR regulon. Our findings suggest that a positive feedback loop involving a LysR-type transcription regulator serves as the basis for an epigenetic switch that controls virulence gene expression in P. aeruginosa.
| Bistable switches allow the expression of a gene, or set of genes, to switch from one stable expression state to another and can generate cells with different phenotypes in an isogenic population. In this work we uncover a previously unidentified bistable switch that controls virulence gene expression in the opportunistic pathogen P. aeruginosa. This switch is controlled by a LysR-type transcription regulator that we call BexR. As well as identifying specific genes that are regulated by BexR, we show that bexR is itself bistably expressed and positively autoregulated. Furthermore, we present evidence that positive autoregulation of bexR is necessary for bistable expression of the BexR regulon. Our findings support a model for BexR-mediated bistability in which positive feedback amplifies bexR expression in a stochastically determined subset of cells, giving rise to heterogeneous expression of BexR target genes within a cell population. By generating diversity in an isogenic population of P. aeruginosa this bistable switch may ensure the survival of a subset of cells in adverse conditions, such as those encountered in the host. Our study defines an epigenetic mechanism for phenotypic variation in P. aeruginosa.
| The Gram-negative bacterium Pseudomonas aeruginosa is an opportunistic pathogen of humans. It can cause infection in a wide variety of tissues in the immunocompromised host, and is the leading cause of morbidity and mortality in cystic fibrosis (CF) patients [1]. This breadth of infectious capacity is thought to result from differential gene expression, as genomic variability between clinical and environmental isolates is low and the genome of P. aeruginosa encodes a high proportion of transcription regulators [2],[3]. Studying the mechanisms and outcomes of transcription regulation in P. aeruginosa may offer some insight into how cohorts of virulence factors are coordinately expressed to influence pathogenesis in a range of pseudomonal infections.
Bacteria are traditionally thought to use transcription regulation to adapt to changing environmental conditions, such as the presence of a new carbon or energy source, a change in temperature or pH, or introduction to a host environment. However, in harsh environmental conditions that exert a sudden selective pressure on a population of cells, the time needed to respond using a genetic regulatory network may prove fatal. The ability of isogenic populations of bacteria to exhibit phenotypic variation allows them to cope with such situations by pre-adapting a subset of the population to the sudden introduction of harsh conditions. Several examples of phenotypic variation in P. aeruginosa have been identified, such as the phase-variable expression of the cupA fimbrial gene cluster under anaerobic conditions and the transient formation of antibiotic resistant, hyperadherent rough small-colony variants under antibiotic exposure [4]–[7]. These phenotypes may contribute to the ability of infecting bacteria to withstand chemical or mechanical insults encountered during colonization of the CF lung. Examples such as these suggest that phenotypic variation by P. aeruginosa allows the organism to thrive in a complex environment. However, the mechanisms by which these phenotypes are variably expressed are unknown.
Phenotypic variation in bacteria can arise from a variety of mechanisms, both genetic and epigenetic in nature. Classical phase-variation is thought to be genetically mediated, such as the variable expression of the flagellum in Salmonella enterica serovar Typhimurium, which is mediated by specifically catalyzed changes in promoter DNA orientation [8]. Phase-variation can also be mediated by epigenetic mechanisms, such as the one involving DNA methylation that controls the phase-variable expression of pyelonephritis-associated pili genes in uropathogenic Escherichia coli [9],[10]. Phenotypic heterogeneity can arise in the absence of DNA sequence variation or DNA modification in bistable systems (i.e. systems that can exist in one of two alternative expression states, and reversibly switch between them), such as in the case of the lysogenic switch of bacteriophage λ [11],[12]. Bistability can arise when there exists a mechanism for amplifying differences in protein levels between individual cells and stably propagating these differences to daughter cells (reviewed in [13]). The bistable expression of genes can be achieved using a positive regulatory feedback loop, as is the case in the development of competence under nutrient limitation in Bacillus subtilis; positive feedback of ComK, the master regulator of competence, is required for bistable development of competence in B. subtilis [14],[15]. Thus, the architecture of a particular gene regulatory circuit can enable stochastic, reversible differentiation of subsets of bacterial populations into distinct cell types.
Here we uncover a previously unidentified bistable switch in P. aeruginosa controlled by BexR, a LysR-type transcription regulator. We demonstrate that bexR is itself bistably expressed in a BexR-dependent manner and that BexR positively regulates the expression of its own gene. Using DNA microarrays and quantitative real-time RT-PCR (qRT-PCR), we define the bistable regulon of BexR, which contains a diverse set of genes and includes aprA, which encodes the virulence factor alkaline protease. We show further that BexR acts directly at the promoters of many of its regulatory targets, including that of its own gene. Finally, we describe a series of single-cell population analyses that suggest that this bistable switch requires bexR autoregulation. We propose a model for the BexR switch in which positive feedback amplifies bexR expression in a stochastically determined subset of cells, giving rise to bistable expression of BexR target genes in an isogenic population.
In the course of unrelated microarray experiments, we observed a small set of genes that exhibited variable expression between replicates of wild-type cultures of P. aeruginosa strain PAO1 (data not shown). This set includes PA1202, which encodes a hypothetical protein with homology to a predicted hydrolase of Escherichia coli, and PA2432 (herein named bexR for bistable expression regulator), which is predicted to encode a member of the LysR family of transcription regulators. To confirm that PA1202 is expressed in a variable manner, we constructed a strain of PAO1 in which lacZ was placed downstream of the PA1202 gene (Figure 1A). This strain exhibits reversible bistable expression of the lacZ reporter. Specifically, wild-type cells of this reporter strain give rise to both blue (“ON”) and white (“OFF”) colonies on LB agar plates containing X-Gal (Figure 1B). When re-streaked on LB agar with X-Gal, ON colonies give rise to both ON and OFF colonies, and OFF colonies give rise to both OFF and ON colonies. Because our initial microarray analyses suggested that bexR, which encodes a putative transcription activator, co-varied with PA1202, we hypothesized that BexR may positively regulate expression of PA1202 and that bistable expression of bexR may be responsible for the observed bistability in PA1202 expression. To begin to test this hypothesis, we constructed an unmarked in-frame deletion of bexR in PAO1 PA1202 lacZ. Compared to the wild-type reporter strain, the ΔbexR mutant exhibits constitutively low-level expression of PA1202 (Figure 1B). Ectopic expression of bexR in the ΔbexR mutant resulted in increased PA1202 expression (Figure 1C), suggesting that BexR positively regulates expression of PA1202. However, bistable expression of PA1202 is lost when bexR is expressed ectopically; PAO1 ΔbexR PA1202 lacZ grows only as ON colonies on LB agar with X-Gal when carrying a plasmid containing bexR (data not shown), suggesting that native regulation of bexR is necessary for bistable PA1202 expression. Quantification of the frequency at which this switch in expression state occurs reveals a relatively infrequent switch with a bias in favor of the OFF to ON transition (Table 1).
To determine whether bexR, like PA1202, is expressed in a bistable manner, we constructed a reporter strain in which the putative bexR promoter region was placed upstream of a GFP-lacZ reporter in single copy at the ΦCTX attachment site in the PAO1 chromosome (Figure 1D) [16],[17]. Individual cells of wild-type PAO1 carrying this PbexR-GFP-lacZ reporter either express the GFP reporter, or do not, leading to heterogeneity in the cell population (Figure 1E). Interestingly, cells lacking BexR exhibit constitutively low-level expression of the reporter, suggesting that bistable expression from the bexR promoter also depends on BexR. Bistable expression from the bexR promoter was also observed at the colony level, suggesting long-term maintenance of the BexR expression state (Figure S1). The frequency of switching between expression states is similar for bexR and PA1202, further supporting the hypothesis that bistable expression of bexR is upstream of PA1202 bistability (Table 1 and Table S1). Truncation of the bexR upstream sequence indicated that a 195 bp fragment of upstream DNA is still sufficient to drive expression of a lacZ reporter (integrated in single copy in the chromosome) when bexR is expressed from a plasmid, whereas an 88 bp fragment is not (Figure 1F). Thus, the 195 bp region of DNA immediately upstream of bexR presumably contains the bexR promoter and BexR binding site(s). Thus, BexR positively regulates expression of PA1202 and of its own gene, and bexR is itself bistably expressed, suggesting that other BexR target genes may also be expressed in a bistable manner.
To determine the full extent of the BexR regulon in PAO1, we compared the mRNA content of PAO1 ΔbexR cells containing either a bexR expression vector or an empty vector in both mid-logarithmic and stationary phases of growth using DNA microarrays. A total of 71 genes exhibited between a 2- and 70-fold change in expression, with most genes upregulated by ectopic expression of bexR (Figure S2). PA1202 was upregulated 70-fold upon ectopic expression of bexR in mid-logarithmic phase. Several genes downstream of PA1202 were also strongly upregulated by ectopic expression of bexR, suggesting that these comprise a BexR-regulated operon. This putative operon includes PA1203, which is predicted to encode a redox protein, PA1204, which is predicted to encode a NADPH-dependent FMN reductase, and PA1205, which is predicted to encode a homolog of pirin, a widely conserved protein with oxygenase activity [18]. PA2698, which is also predicted to encode a hydrolase, was upregulated 7-fold by ectopic expression of bexR, suggesting that a cohort of several enzymes are coordinately regulated by BexR. Several multidrug efflux pumps appeared to be regulatory targets of BexR, as downregulation of mexEF-oprN by 6- to 10-fold and upregulation of mexGHI-opmD by 7- to 13-fold was observed during ectopic expression of bexR. Several quorum sensing-regulated genes encoding secreted proteins were also positively regulated by ectopic bexR expression, such as PA0572, which encodes a LasR-regulated Xcp secretion substrate with a predicted Zn-metalloprotease motif [19]–[21]. Finally, the LasR-regulated genes aprX, aprE, aprF and aprA, which encode components of the alkaline protease production and secretion machinery, were positively regulated by BexR. aprA, which encodes the alkaline protease precursor protein, plays a role in virulence in a Drosophila melanogaster orogastric model of pseudomonal infection, where it is thought to protect P. entomophila from antimicrobial peptides [22]. These results suggest that BexR controls the expression of a diverse set of genes, including some that encode predicted enzymes and others that encode quorum-regulated secreted proteases.
Because bexR is itself bistably expressed we would predict that the expression of BexR target genes in wild-type cells should co-vary with the bexR expression state. To test this prediction, we isolated mRNA from cultures of wild-type attB::PbexR-lacZ OFF, attB::PbexR-lacZ ON and ΔbexR attB::PbexR-lacZ reporter strains at both mid-logarithmic and stationary growth phases and profiled relative transcript abundance by qRT-PCR. We observed an approximately 10-fold difference in abundance of bexR transcripts between OFF and ON cultures in mid-logarithmic phase, and an approximately 6-fold difference between OFF and ON cultures in stationary phase (Figure 2A). Consistent with the idea that BexR target genes are expressed in a bistable manner in wild-type cells, expression of members of the putative PA1202 operon, from PA1202 to PA1205, all co-varied with bexR expression (Figure 2B), as did PA0572 and aprA (though for aprA the difference in transcript abundance between ON and OFF cultures was only 2-fold) (Figure 2C). We were unable to observe significant bistable expression of the other apr genes, possibly due to the relatively modest effect of BexR on their expression. The abundance of the lasA transcript was not significantly different between ΔbexR and wild-type cultures, suggesting that the observed bistability of aprA and PA0572 (which, like lasA, are LasR-regulated [19]) is not due to differences in LasR function between ON and OFF cultures (Figure 2C). Microarray analysis of cells ectopically expressing bexR suggests that two operons encoding multidrug efflux pumps are reciprocally regulated by BexR (Figure S2). However, this was not observed in wild-type cells in the OFF and ON states (data not shown). Taken together, our data indicate that BexR is responsible for coordinate bistable expression of a variety of genes in wild-type P. aeruginosa, including two that encode quorum sensing-regulated secreted proteases (PA0572 and aprA).
To address whether BexR directly regulates transcription of its target genes, we used chromatin immunoprecipitation (ChIP). We constructed a strain in which the native chromosomal copy of the bexR gene has been modified to encode a version of BexR containing a vesicular stomatitis virus glycoprotein (VSV-G) epitope tag at its C-terminus (BexR-V). This strain retained the ability to bistably express PA1202 lacZ on LB agar containing X-Gal, suggesting that the VSV-G epitope tag does not interfere with BexR activity (data not shown). We immunoprecipitated BexR-V-associated DNA from wild-type ON cultures grown to both mid-logarithmic and stationary phase and quantified occupancy of BexR-V at candidate target promoters relative to a control region not expected to bind BexR-V. BexR-V strongly occupies its own promoter, as well as those of PA1202 and PA0572 (Figure 3). Furthermore, BexR-V occupied the aprX and aprA promoters, but not the intervening DNA upstream of aprD. This suggests that BexR-V has at least two distinct binding sites within the apr locus. All occupancies were significantly higher than those observed in both wild-type OFF cultures and in a non-epitope tagged control strain (Figure S3). These results suggest that BexR regulates many of its target genes directly.
The evidence presented thus far suggests that bexR encodes a bistably expressed transcription regulator that positively regulates its own expression. This is reminiscent of the competence switch in B. subtilis. In this system, ComK, the master regulator of competence, positively regulates transcription of its own gene, thereby enabling a non-linear response to increasing concentrations of ComK, which leads to bistability in the development of competence. Using single-cell fluorescent reporter analysis, it has been shown that the ComK positive feedback loop is required for bistable expression of competence [14],[15]. We hypothesized that, in a similar manner, the positive feedback loop controlling bexR expression is required for bistable expression of the BexR regulon (i.e. positive feedback of bexR creates a condition of hypersensitivity to variation in levels of BexR protein). If this hypothesis is correct a gradual increase in basal bexR expression should increase the proportion of ON relative to OFF cells specifically in a strain with an intact positive feedback loop. In a strain that lacks this positive feedback loop, a graded increase in bexR expression should lead to a corresponding increase in expression of bexR-regulated genes with no detectable bistability.
Wild-type P. aeruginosa cells containing a PbexR-GFP-lacZ reporter construct integrated in single copy into the chromosome can be seen to exhibit BexR-dependent bistable expression of this reporter by fluorescence microscopy (Figure 1E). Consistent with this observation, quantification of the fluorescence of individual cells within a culture derived from either an ON colony or an OFF colony reveals that cells in the ON and OFF expression states can be distinguished from one another, and that each culture contains both ON and OFF cells (Figure 4). To analyze the effect of positive feedback on bexR bistability, we constructed a pair of strains containing the PbexR-GFP-lacZ reporter construct and an isopropyl-β-D-thiogalactoside (IPTG)-inducible copy of bexR (also provided in single copy from the chromosome from a different locus). One of these strains contained an unmarked, in-frame deletion of bexR (the minus feedback strain, Figure 5A), whereas the other contained wild-type bexR at its native locus (the plus feedback strain, Figure 5B). In the absence of IPTG, only cells of the reporter strain with the intact positive feedback loop displayed bistability, and contained two populations of cells corresponding to those in the ON and OFF expression states (manifest in Figure 5B [and Figure 6B] as a population of cells with an essentially bimodal distribution of fluorescence intensities). Furthermore, a gradual increase in ectopically expressed bexR resulted in an increase in the proportion of ON relative to OFF cells only in the plus feedback strain (Figure 5B); in the strain lacking the positive feedback loop, cells responded relatively uniformly to increasing synthesis of ectopically expressed bexR (manifest in Figure 5A as a population of cells with a normal distribution of fluorescence intensities, whose average fluorescence intensity increases with IPTG concentration). Importantly, for IPTG concentrations at which the average cell fluorescence intensity was similar between cells with and without feedback, two distinct cell populations (ON and OFF) were observed only in cells with an intact positive feedback loop (Figure 5). In particular, cells of the plus feedback strain at 0.5 mM IPTG had a mean fluorescence intensity of 1814 arbitrary units, which is similar to the mean fluorescence intensity of 1720 arbitrary units exhibited by the minus feedback strain at 4 mM IPTG. Whereas the mean reporter gene expression of these two cell populations, and thus the average abundance of BexR protein per cell, was quite similar under these two conditions, the existence of two subpopulations of cells occurred only in the presence of bexR autoregulation (Figure 5). These results suggest that positive feedback of bexR is necessary for bistability.
Feedback-mediated bistable systems often exhibit a capacity for history-dependent behavior, or hysteresis [reviewed in 23]. Systems exhibiting hysteretic behavior may have different responses under identical conditions, depending on the conditions previously experienced. For example, in bistable expression of the lac operon of E. coli at low concentrations of a non-metabolizable lactose analog, the concentration of inducer at which initially uninduced cells turn on is higher than that at which initially induced cells turn off [24],[25]. The behavior of this system at concentrations of inducer between these thresholds therefore depends on conditions previously encountered. Thus, systems with positive feedback can exhibit memory of previous expression states. To investigate the possibility that positive feedback of bexR can impart a memory of previous expression states on the system, we utilized the plus and minus feedback strains described above (Figure 5) and observed their response over time to a pulse of ectopically expressed bexR, induced by a 2 hour exposure to 20 mM IPTG. In cells without an intact positive feedback loop, the IPTG pulse was sufficient to raise the mean fluorescence intensity to the level seen in wild-type ON cells (Figure 6A and Figure 4). However, this degree of expression from the PbexR-GFP-lacZ reporter was quickly lost upon removal of IPTG and subculturing of cells into fresh media. In contrast, cells of the plus feedback strain maintained their induced state for many generations after the removal of IPTG, suggesting that a brief period in which cells experience a high intracellular concentration of BexR is sufficient to induce a long-lasting ON state (Figure 6B). Indeed, a pulse with IPTG for only 30 minutes is sufficient to induce a transition to a sustained ON state in the plus feedback strain (Figure S4). Only after 31 generations following removal of IPTG, do a portion of the cells begin to transition to the OFF state (Figure 6B). Taken together, the results of our single-cell population analyses suggest a mechanism in which variation in basal expression of bexR in OFF cells is amplified by a positive feedback loop in a stochastically determined subset of cells that then transitions to the ON state and is maintained in that state by continued autoactivation of BexR (Figure 7).
The results above characterize a heretofore undescribed bistable switch in P. aeruginosa that controls virulence gene expression. We have shown that bexR, which encodes a LysR-type transcription regulator, is bistably expressed, and that this bistability results in altered expression of several downstream genes, including those in the uncharacterized PA1202 operon and aprA, which encodes the virulence factor alkaline protease. Furthermore, reporter assays show that BexR can positively regulate its own expression. ChIP analysis indicates that BexR acts directly at the sites of many target promoters, including that of its own gene. Finally, single-cell analyses of the response of a cell population to a graded source of BexR, or a pulse of BexR, suggests that positive autoregulation is necessary for the observed bistability. Taken together, these results outline a novel feedback-mediated bistable switch in an opportunistic pathogen.
Bistability is a mechanism by which bacteria can introduce phenotypic heterogeneity within an isogenic population, thereby creating a subset of cells capable of surviving the onset of an otherwise lethal situation. For example, some bacteria have the ability to survive antibiotic treatment without evolving bona fide resistance by stochastically entering a dormant “persister” state during vegetative growth [26]. A recent study suggests that a bexR transposon mutant has 2-fold increased sensitivity to the fluoroquinolone antibiotic ciprofloxacin, which is used in treatment of P. aeruginosa infections in CF patients, though the potential mechanism for this increased sensitivity was not addressed [27],[28]. Although our findings raised the possibility that bistable bexR expression might lead to heterogeneity in ciprofloxacin resistance, we found no evidence that bexR contributed to the resistance of P. aeruginosa to ciprofloxacin, at least in strain PAO1 (data not shown).
Bistable expression of virulence factors has been previously reported in P. aeruginosa. For instance, the Type III secretion system is only expressed in a subset of cells grown in inducing conditions [29]. Additionally, the cupA fimbrial gene cluster is bistably expressed by P. aeruginosa when grown in anaerobic conditions [5]. Bistable expression of several virulence factors independently of one another may create several subtypes of cells with differing virulence potential within an isogenic population of infecting bacteria. Thus, bistable expression of virulence factors may represent a strategy employed by P. aeruginosa to generate cell types specialized to survive within different niches in the host.
In P. entomophila, AprA has a significant role in virulence in a D. melanogaster oral model of infection, where it is thought to protect the bacterium from the effects of host-produced antimicrobial peptides [22]. Although oral models of D. melanogaster infection with P. aeruginosa have been used to successfully characterize bacterial virulence, these models have not been used to test the role of AprA in P. aeruginosa virulence [30],[31]. If alkaline protease does play a role in defense against antimicrobial peptides in P. aeruginosa, upregulating aprA ∼2-fold in a subset of cells through BexR-mediated bistability may preemptively adapt a portion of the cell population to the sudden introduction to a particular host environment. P. aeruginosa alkaline protease has been shown to degrade a variety of human proteins and tissues and inhibit immune cell function, presumably by acting at the cell surface to modify phagocytic and chemotactic receptors (reviewed in [32]). Alkaline protease has also been suggested to play a role in corneal keratitis [33], although this role for AprA has been disputed more recently by the comparison of isogenic mutant strains [34]. However, our observation that wild-type strains of P. aeruginosa bistably express aprA may complicate the interpretation of earlier work. Interestingly, the rhizobacterium P. brassicacearum exhibits phenotypic variation in expression of an alkaline protease homolog, though whether this is mediated by bistability of a BexR homolog is unknown [35]. It has been suggested that heterogeneous production of extracellular proteases by an isogenic population of bacteria is an example of cooperative or altruistic behavior, as these proteases diffuse freely through the growth medium and can equally benefit all members of the population [36]. Thus, bistable production of alkaline protease or PA0572, a predicted protease, may serve to benefit both ON and OFF cells in a population. Whether bistable expression of aprA, or other members of the BexR regulon, has a role in mammalian virulence remains to be seen.
In contrast with aprA, many other regulatory targets of BexR are poorly characterized hypothetical genes. BexR-mediated bistability does not appear to be limited to P. aeruginosa PAO1, as the homolog of PA1202 in P. aeruginosa PA14, a more virulent clinical strain, is also bistably expressed in a BexR-dependent manner (Figure S5). This conservation across diverse strains of P. aeruginosa suggests an important biological role for BexR-mediated bistability. In this regard, a particularly interesting target of BexR is the PA1202 operon, which is strongly positively regulated by BexR. Several genes in this operon, such as PA1202 and PA1205, are predicted to encode enzymes with catabolic activity directed against small molecules. This may point to a role for the BexR regulon in the ability of P. aeruginosa to metabolize and thereby detoxify certain small molecules. Co-regulation of a diverse set of genes by BexR may indicate that it is involved in manifestation of more than one phenotype. That these genes are expressed in a bistable manner suggests that their expression or repression may be detrimental to growth under certain conditions.
We propose that positive feedback of bexR provides a mechanism for amplification and propagation of cell-to-cell variability in BexR levels. This regulatory circuit is similar to the one governing competence development in B. subtilis. Experiments in this system have suggested that noisy expression of comK results in ComK levels in a subpopulation of cells crossing a threshold level for comK autoactivation, causing differentiation into the competent state [14],[15],[37]. Noise in bexR expression may also provide the basis for generating cell-to-cell variability in BexR levels. The frequency of the BexR switch differs from that of the ComK switch. Whereas B. subtilis has been directly observed to enter a competent state in approximately 3.6% of cell division events [38], P. aeruginosa enters into the BexR-ON state approximately 10-fold less frequently, and the BexR-OFF state even less so (Table 1 and Table S1). These low frequencies are on par with classical phase-variation systems, but in the case of BexR, the expression state stability appears to be epigenetically mediated. This low switching frequency may be a function of the high degree of hysteresis observed in the BexR switch. Biological systems capable of hysteretic behavior can retain a memory of previous exposure to inducing conditions, and this has been observed in both naturally occurring and synthetic systems [25],[39]. Strictly speaking, hysteresis is not a necessary characteristic of bistable systems, as a synthetic feedback-mediated bistable system was observed to exhibit clear bistability but display no history-dependent response [40]. Nevertheless, hysteresis is often associated with bistable systems, and that it is observed in the BexR switch may suggest that retaining memory of previous conditions is beneficial to the cell.
In B. subtilis, regulation of ComK levels is achieved by degradation of ComK by the MecA/ClpCP complex and the inhibition thereof by ComS [41],[42]. Our single-cell population analyses indicate that directly modulating BexR levels by induction of ectopic synthesis can affect the frequency at which cells differentiate into the ON state (Figure 5). Modulation of BexR levels or activity in wild-type cells may provide a mechanism for fine-tuning the dynamics of this bistability. There may be accessory factors, perhaps themselves BexR-regulated, that affect BexR levels or activity. A mechanism for modulating BexR autoactivation dynamics may allow P. aeruginosa to regulate switching frequency in response to external conditions. As LysR-type transcription activators often bind to small molecules to alter their DNA-binding and regulatory properties, it is possible that the dynamics of the BexR switch may be tunable by a coinducer molecule [43]. However, no such molecule has yet been identified.
The results presented here outline a model for differentiation into the BexR-ON state, but do not address the mechanism by which a BexR-ON cell can revert to the BexR-OFF state. Previous studies suggest that escape from a positive feedback loop is often mediated by an accessory process. For example, escape from competence in B. subtilis occurs when reduction in ComS levels promotes ComK proteolysis by the MecA/ClpCP complex, relieving ComK autoactivation [38]. The switch from BexR-ON to BexR-OFF may also involve some antagonistic process. Unlike several other feedback-mediated bistable switches, the switch from ON to OFF in the case of BexR appears to occur only in a stochastically determined subset of cells. For example, escape from competence in B. subtilis occurs because comS transcription is repressed by ComK in the competent state and ComS protein gradually depletes in all cells [38]. In contrast, the BexR-ON state is relatively stable and heritable, and is lost only in a subpopulation of cells. The existence of a stochastic process mediating the switch to BexR-OFF that is distinct from the one mediating the switch to BexR-ON, is further supported by the ∼60-fold directional bias in switching frequencies (Table 1). This process may take the form of transcription regulation of bexR or post-translational modulation of BexR levels or activity, and we are currently investigating these possibilities.
P. aeruginosa strains PAO1 and PA14 were provided by Arne Rietsch (Case Western Reserve University). E. coli DH5α F'IQ (Invitrogen) was used as the recipient strain for all plasmid constructions, whereas E. coli strain SM10 (λpir) was used to mate plasmids into P. aeruginosa.
The PA1202 lacZ reporter strain (PAO1 PA1202 lacZ) contains the lacZ gene integrated immediately downstream of the PA1202 gene on the PAO1 chromosome and was made by allelic exchange. PCR products 486 bp and 513 bp in length flanking the 3′ end of PA1202 were amplified and spliced together to add KpnI, NcoI and SphI sites two bases after the PA1202 stop codon. This PCR product was cloned as a SacI/PacI fragment into pEXG2 [44]. The lacZ gene was subsequently cloned into this construct as a KpnI/SphI fragment derived from pP18-lacZ (Arne Rietsch, unpublished work), generating plasmid pEXF1202-lacZ. This plasmid was then used to create reporter strains PAO1 PA1202 lacZ and PA14 PA1202 lacZ by allelic exchange.
The deletion construct for the bexR gene (PA2432) was generated by amplifying regions 398 bp and 360 bp in length that flank bexR in the PAO1 genome by the PCR and then splicing the flanking regions together by overlap extension PCR; deletions were in-frame and contained the 9-bp linker sequence 5′-GCGGCCGCC-3′. The resulting PCR product was cloned on an EcoRI/BamHI fragment into plasmid pEX18Gm [45], yielding plasmid pEXM2432. This plasmid was then used to create strains PAO1 ΔbexR, PAO1 PA1202 lacZ ΔbexR and PA14 PA1202 lacZ ΔbexR by allelic exchange [45]. Deletions were confirmed by the PCR.
The attB::PbexR-lacZ reporter strains contain fragments of the bexR promoter fused to the lacZ gene and integrated in single copy into the attB locus in the PAO1 chromosome and were made by site-specific integration followed by backbone excision through transient synthesis of FLP recombinase from plasmid pFLP2 [17],[45]. PCR products spanning from 91, 198 or 297 bp to 3 bp upstream of the bexR start codon were amplified and cloned as EcoRI/XhoI fragments into mini-CTX-lacZ [17], which contains a consensus Shine-Dalgarno sequence upstream of lacZ, yielding plasmids mini-CTX-PF2432-lacZ.1, mini-CTX-PF2432-lacZ.2 and mini-CTX-PF2432-lacZ.3, respectively. These plasmids were then used to create reporter strains PAO1 attB::PbexR.1-lacZ, PAO1 ΔbexR attB::PbexR.1-lacZ, PAO1 attB::PbexR.2-lacZ, PAO1 ΔbexR attB::PbexR.2-lacZ, PAO1 attB::PbexR.3-lacZ and PAO1 ΔbexR attB::PbexR.3-lacZ. An EcoRI/XhoI fragment of mini-CTX-PF2432-lacZ.3 was subcloned into mini-CTX-GFP-lacZ [16], yielding plasmid mini-CTX-PF2432-GFP-lacZ.3. This plasmid was then used to create the fluorescent reporter strains PAO1 attB::PbexR.3-GFP-lacZ and PAO1 ΔbexR attB::PbexR.3-GFP-lacZ.
The BexR-VSV-G integration vector was generated by first cloning a PCR-amplified DNA fragment containing ∼300 bp of sequence from the 3′ portion of the bexR gene on a HindIII/NotI fragment into plasmid pP30Δ-YTAP [4], generating plasmid pP30Δ-BexR-TAP. This HindIII/NotI fragment was then subcloned into pP30ΔFRT-MvaT-V [46], generating plasmid pP30ΔFRT-BexR-V. This plasmid was used to make strain PAO1 PA1202 lacZ BexR-V by homologous recombination at the bexR locus followed by backbone excision through transient synthesis of FLP recombinase from plasmid pFLP2 [45]. Production of the BexR-V protein was confirmed by Western blotting with an anti-VSV-G antibody (Sigma).
Plasmid pBexR is a derivative of pPSV35 [44] and directs the synthesis of the BexR protein under control of the IPTG-inducible lacUV5 promoter. The plasmid was made by subcloning an EcoRI/HindIII DNA fragment containing a consensus Shine-Dalgarno sequence and the bexR gene into pPSV35.
The attTn7::TOPLAC-bexR strains contain a construct which directs the synthesis of the BexR protein under control of the IPTG-inducible TOPLAC promoter stably integrated into the genome in single copy at the attTn7 locus. The TOPLAC promoter in this construct is a derivative of the lac promoter that contains two lac operator sequences centered at positions −63.5 and +11. The sequence of this promoter is 5′-CACTACGTGCTCGAGGGTAAATGTGAGCACTCACAATTTATTCTGAAATGAGCTCTTTACACGTCCTGCTGCCGGCTCGTATGTTGTGTGGAATTGTGAGCGGATAACAATTAAGCTTAGTCGACAGCTAGCCGGATCC-3′, where the -35 and -10 sequences are underlined and the lac operator sequences are shown in bold. The bexR gene is inserted downstream of the TOPLAC promoter with a consensus Shine-Dalgarno sequence. This construct was inserted between the ends of the Tn7 transposon on pUC18-mini-Tn7T-LAC [47], generating plasmid pUC18-mini-Tn7T-TOPLAC-bexR. This plasmid was used to make strains PAO1 attB::PbexR.3-GFP-lacZ attTn7::TOPLAC-bexR and PAO1 ΔbexR attB::PbexR.3-GFP-lacZ attTn7::TOPLAC-bexR by site-specific recombination [47].
Cells were grown with aeration at 37°C to mid-logarithmic phase in LB supplemented as needed with gentamicin (25 µg/ml) and IPTG (0.1 mM). Cells were permeabilized with sodium dodecyl sulfate and CHCl3 and assayed for β-galactosidase activity as described previously [48]. Assays were performed at least twice in triplicate on separate occasions. Representative data sets are shown.
Cultures of PAO1 ΔbexR attB::PbexR.3-lacZ and PAO1 attB::PbexR.3-lacZ in the OFF and ON states were inoculated in quadruplicate at starting OD600 of ≈0.01 and grown with aeration to an OD600 of ≈0.55 (representing mid-logarithmic phase) and to an OD600 of ≈2.4 (representing stationary phase) at 37°C in LB. Cells were then harvested by centrifugation and RNA prepared essentially as described [49]. Transcripts were quantified by quantitative real-time RT-PCR as described [50].
Switching-frequency calculations were performed essentially as described [51], except that cells were plated on LB agar plates containing 50 µg/ml X-Gal and grown at 37°C. Error values represent 1 standard deviation (SD) from the mean switching frequency.
Cultures of PAO1 ΔbexR containing plasmid pPSV35 [44] or pBexR were grown with aeration at 37°C in LB containing gentamicin (25 µg/ml). Triplicate cultures of each strain were inoculated at a starting OD600 of ≈0.01 and grown to an OD600 of ≈0.5 (representing mid-logarithmic phase) and to an OD600 of ≈2.3 (representing stationary phase). RNA isolation, cDNA synthesis, and cDNA fragmentation and labeling were performed essentially as described previously [49]. Labeled samples were hybridized to Affymetrix GeneChip P. aeruginosa genome arrays (Affymetrix). Data were analyzed for statistically significant (p<0.05, fold change >2) changes in gene expression using GeneSpring GX.
Cultures of PAO1 PA1202 lacZ BexR-V in either the ON or OFF state were inoculated in quadruplicate at a starting OD600 of ≈0.01 and grown with aeration to an OD600 of ≈0.5 (representing mid-logarithmic phase) and to an OD600 of ≈2.0 (representing stationary phase) at 37°C in LB. ChIP was then performed with 3 ml of culture and fold enrichment values were measured by quantitative PCR relative to the PA2155 promoter essentially as described [46].
For fluorescence micrograph analysis, cultures were fixed with formaldehyde and glutaraldehyde at 2.4% and 0.04%, respectively, and cells were allowed to fix for 30 minutes at room temperature. Cells were washed three times with PBS and imaged on a Nikon TE2000 inverted microscope outfitted with a Nikon Intensilight illuminator, a Coolsnap HQ2 charge-coupled device camera from Photometrics and a Nikon CFI Plan Apo VC ×100 objective lens (1.4 NA) for differential interference contrast (DIC) imaging. For GFP images the ET-GFP filter set (Chroma 49002) was used. Images were captured using Nikon Elements software, which was also used for quantification of fluorescence in individual cells. This was done by automatically defining cell boundaries using the DIC image, excluding cells that were poorly focused, narrower than 0.5 µm, longer than 4.0 µm or shorter than 0.5 µm, and using those regions to quantify the GFP image. Values given are subtracted for background fluorescence. At least 400 cells were imaged for each timepoint, and the fluorescence intensities of a random subset of 250 cells are displayed in scatter plots. Images were exported to Adobe Photoshop CS4 for preparation.
For the hypersensitivity experiment (Figure 5), cells were grown with aeration at 37°C to mid-logarithmic phase in LB supplemented as needed with IPTG and prepared for microscopy as described above. The experiment was performed at least twice in duplicate on separate occasions. A representative data set from a single replicate is shown.
The hysteresis experiment (Figure 6) was performed by growing cells with aeration at 37°C in LB and either treating them with 20 mM IPTG for 2 hours or 30 minutes immediately before reaching mid-logarithmic phase, or not treating them with IPTG. A sample was then taken and prepared for microscopy (corresponding to the 0 generation time point) as described above while the remaining cells were washed with LB to remove the IPTG, and inoculated into fresh media at a 1∶4 dilution. Cells were then grown continuously for 2 generations to mid-logarithmic phase, a sample was taken and prepared for microscopy (corresponding to the 2 generation time point) and a fresh culture was inoculated at a 1∶16 dilution with the remaining cells. Cells were then grown continuously for 4 generations to mid-logarithmic phase, a sample was taken and prepared for microscopy (corresponding to the 6 generation time point) and a fresh culture was inoculated at a 1∶16 dilution with the remaining cells. Cells were then grown continuously for 5 generations to late-logarithmic phase, a sample was taken and prepared for microscopy (corresponding to the 11 generation time point) and a fresh culture was inoculated at a 1∶32 dilution with the remaining cells. Cells were then grown continuously for 5 generations to late-logarithmic phase and a fresh culture was inoculated at a 1∶16 dilution. Cells were then grown continuously for 3 generations to mid-logarithmic phase, a sample was taken and prepared for microscopy (corresponding to the 19 generation time point), remaining cells were allowed to grow for 1.5 generations to early stationary phase and used to inoculate a fresh culture at a 1∶100 dilution. Cells were then grown continuously for 7 generations (overnight) and used to inoculate a fresh culture at a 1∶100 dilution. Cells were then grown continuously for 3.5 generations to mid-logarithmic phase and a sample was taken and prepared for microscopy (corresponding to the 31 generation time point). The experiment was performed at least three times in duplicate on separate occasions. A representative data set from a single replicate is shown.
|
10.1371/journal.pcbi.1002127 | Genome-Scale Analysis of Translation Elongation with a Ribosome Flow Model | We describe the first large scale analysis of gene translation that is based on a model that takes into account the physical and dynamical nature of this process. The Ribosomal Flow Model (RFM) predicts fundamental features of the translation process, including translation rates, protein abundance levels, ribosomal densities and the relation between all these variables, better than alternative (‘non-physical’) approaches. In addition, we show that the RFM can be used for accurate inference of various other quantities including genes' initiation rates and translation costs. These quantities could not be inferred by previous predictors. We find that increasing the number of available ribosomes (or equivalently the initiation rate) increases the genomic translation rate and the mean ribosome density only up to a certain point, beyond which both saturate. Strikingly, assuming that the translation system is tuned to work at the pre-saturation point maximizes the predictive power of the model with respect to experimental data. This result suggests that in all organisms that were analyzed (from bacteria to Human), the global initiation rate is optimized to attain the pre-saturation point. The fact that similar results were not observed for heterologous genes indicates that this feature is under selection. Remarkably, the gap between the performance of the RFM and alternative predictors is strikingly large in the case of heterologous genes, testifying to the model's promising biotechnological value in predicting the abundance of heterologous proteins before expressing them in the desired host.
| Gene translation is a central process in all living organisms. However, this process is still enigmatic, and contradicting conclusions regarding the essential parameters that determine translation rates appear in different studies. We introduce a new approach for modeling the process of translation elongation. Taking into account the stochastic nature of the translation process and the excluded volume interactions between ribosomes, our model is aimed at capturing the effect of codon order and composition on translation rates. We demonstrate that in comparison to commonly used approaches, our approach gives more accurate predictions of translation rates, protein abundance levels and ribosome densities across many species. Using our model, we address the need for a better understanding of the inner workings of the translation process. To this end, we analyze large scale genomic measurements made in several organisms. Our analysis unravels several central and previously uncharacterized aspects of the translation process. For example, we show that in all organisms that were analyzed (from bacteria to Human), ribosome allocation is optimized to give maximal translation rate in minimal cost. In addition, we provide the first direct estimate for the effect of codon order on protein abundance, showing that in E. coli and S. cerevisiae it can solely account for more than 20% of the variation in this quantity.
| Gene translation is a complex process through which an mRNA sequence is decoded by the ribosome to produce a specific protein. The elongation step of this process is an iterative procedure in which each codon in the mRNA sequence is recognized by a specific tRNA, which adds one additional amino-acid to the growing peptide [1]. As gene translation is a central process in all living organisms, its understanding has ramifications to human health [2], [3], [4], biotechnology [5], [6], [7], [8], [9], [10], [11], [12] and evolution [4], [7], [11], [13].
In recent years there has been a sharp growth in the number of new technologies for measuring different features related to the process of gene translation [5], [6], [10], [14], [15], [16], [17], [18], [19]. However, this process is still enigmatic, with contradicting conclusions in different studies. In particular, the identity of the essential parameters that determine translation rates is still under debate [6], [20], [21]. Recent studies have suggested that the order of codons along the mRNA (and not only the composition of codons) plays an important role in determining translation efficiency [7], [20], [22], [23]. Starting with the seminal work of MacDonald et al. [24], [25] and the work of Heinrich et al. [24], [25] theoretical models for the movement of ribosomes (and other biological ‘machines’) have been presented [26], [27], [28]. Despite being relatively realistic these models haven't been used for the analysis of large scale genomic data. The models that have been used for this purpose, while making promising and worthy first strides, have not attempted to capture the nature of the translation elongation process on all its various physical aspects [6], [13], [26], [27], [28], [29], [30].
The most widely used predictors of translation efficiency are the codon adaptation index (CAI) [28] and the tRNA adaptation index (tAI) [27]. As we describe later, the tAI is the mean adaptation of a gene (i.e., of its codons) to the tRNA pool of the organism. The CAI is similar to the tAI albeit in this predictor the weight of each codon is computed based on its frequency in a set of highly expressed genes. Based on measures such as the tAI, it is possible to estimate the translation rate of single codons. Thus, it possible to study (local) translation rate profiles along genes [7], [31]. As we depict later, in this study we take into account some additional physical aspects of translation elongation.
The aim of the present research is twofold:
First, we address the need for a simple, physically plausible computational model that is solely based on the coding sequence (i.e. a vector of codons in each gene). In addition we further require that the model will allow for a computationally efficient analysis of the translation process on a genome-wide scale and across many species. Focusing on the coding sequence, we by no means wish to imply that it is only factor taking place in the determination of translation rates. Nevertheless, since it has been widely recognized as a prime factor in the translation elongation process, we will herby study it in isolation. To this end, we introduce a new approach for modeling translation elongation. Our model is aimed at capturing the effect of codon order on translation rates, the stochastic nature of the translation process and the interactions between ribosomes. We demonstrate that our approach gives more accurate predictions of translation rates, protein abundance and ribosome densities in endogenous and heterologous genes in comparison to contemporary approaches.
Second, using our model, we address the need for a better understanding of the translation process. Our analysis unravels several central and yet uncharacterized aspects of this process.
Our model is based on the Totally Asymmetric Exclusion Process (TASEP, see, for example [24], [25], and subsequent studies [32]. In the TASEP, initiation time as well as the time a ribosome spends translating each codon is exponentially distributed (mean translation times are of course is codon dependent). In addition, ribosomes span over several codons and if two ribosomes are adjacent, the trailing one is delayed until the ribosome in front of it has proceeded onwards (Figure 1A, Methods, see also Text S1).
Despite its rather simple description, the mathematical tractability of the model described above is poor and full, large scale, simulations of it are relatively slow. In order to allow for analytical treatment and in order to reduce simulation times, we introduced two simplifications. First, instead of describing the dynamics at the level of a single mRNA molecule we describe the dynamics after it was averaged over many identical mRNA molecules (Methods). Second, we limit ourselves to a spatial resolution that is of the size of a single ribosome. These simplifications will be further explained and justified later.
The simplified model, entitled Ribosome Flow Model (RFM), is illustrated in Figure 1 B–C. mRNA molecules are coarse-grained into sites of codons; (in Figure 1B C = 3); in practice, as we discuss with more details latter, we use C = 25 (unless otherwise mentioned), a value that is close to various geometrical properties of the ribosome such as its footprint on the mRNA sequence and the length of its exit channel [7], [14], [22], [33], [34], [35]. As we report later, the choice C = 25 is not arbitrary and was made since it gives the best predictions of protein abundance levels.
Ribosomes arrive at the first site with initiation rate but are only able to bind if this site is not occupied by another ribosome. The initiation rate is a function of physical features such as the number of available free ribosomes [7], [36], [37], the folding energy of the 5′UTRs [6], [20], the folding energy at the beginning of the coding sequence [6], [20], [38], [39] and the base pairing potential between the 5′UTR and the ribosomal rRNA [40]. As some of these features and their combined effect are unknown and out of the scope of this paper, we assume a global initiation rate or infer the initiation rate from the coding sequences (as we show in the section ‘Optimality of the translation machinery’). We do so for the sake of simplicity and in order to avoid over-fitting of data.
A ribosome that occupies the site moves, with rate , to the consecutive site provided the latter is not occupied by another ribosome. Transition rates are determined by the codon composition of each site and the tRNA pool of the organism. Briefly, taking into account the affinity between tRNA species and codons, the translation rate of a codon is proportional to the abundance of the tRNA species that recognize it (Figure 1, see more details in the Methods section).
Denoting the probability that the site is occupied at time by , it follows that the rate of ribosome flow into/out of the system is given by: and respectively. The rate of ribosome ‘flow’ from site to site is given by: (see the Methods section). As we discuss in details (see the Methods section and Figure 1D), the RFM and the full TASEP model, give similar predictions, yet the RFM runs markedly faster.
In this paper we focus on the steady state solution of the equations presented in Figure 1C and specifically in the rate of protein production at steady state. Steady state is a widely used assumption in cases like these (see, for example, [7], [32], [33]) and is hence a good starting point for a large scale study as the one conducted here. In addition, a pioneering analysis that took into account mRNA degradation and was not based on the steady state assumption, was unable to improve the predictive power of the model with respect to existing data (Methods). We note however, that this line of investigation is far from being exhausted and that it should be revisited once degradation rates of mRNA molecules and proteins become available (this data is currently lacking for the vast majority of genomes and heterologous genes).
We denote the steady state site occupation probabilities by and the steady state ribosome flow through the system by . The latter denotes the number of ribosomes passing through a given site per unit time and we note that this rate is nothing but the steady state rate of protein production.
One advantage of the RFM is its amenability to both analytical and numerical analysis. In particular one can study ribosome density profiles and protein production rates from the equilibrium dynamics of the translation process. The Methods section describes how to solve the model analytically under steady state conditions; in this section we discuss some of the basic properties of the solution.
We described a novel analysis of large scale genomic data by a predictor/model that is based on the physical and dynamical nature of gene translation. Given the copy numbers of the tRNA genes in the host genome, our model, the RFM, is based only on codon-bias; It can hence be applied when only the coding sequence of a gene is available and without additional data or information. Despite its relative simplicity, we show that our model predicts features such as protein abundance in endogenous and heterologous genes better than alternative (‘non-physical’) approaches. We demonstrate that the gap between the performance of the RFM and alternative predictors is especially large in the case of heterologous genes; thus, it should be very helpful in the common challenge of predicting the protein abundance of potential heterologous proteins before expressing them in the desired host (see, for example, [5], [6], [7], [21], [53], [54], [55], [56]). In addition, we have demonstrated that our approach can be used for accurately inferring various variables that cannot be inferred by the common predictors used nowadays.
From a Systems Biology point of view, by using our model we were able to demonstrate the global optimality of the process of gene translation [6], [7], [20]. We discovered that increasing the number of available ribosomes (or the initiation rate, ) increases the genomic translation rate and the mean ribosomal density only up to a certain point. After this point, the system is ‘saturated’: adding more ribosomes/increasing the initiation rate does not result in an increase of these two variables. Quite strikingly, in all the organisms we have analyzed, the global initiation rate is optimized to the pre-saturation point. The fact that similar results were not observed in artificial genes supports the conclusion that this feature is under selection.
Optimality of the translation machinery is perhaps not so surprising. Protein production is a central and complex process in the cell. For example, at any given time point there are around 60,000 mRNA molecules in S. cerevisiae [36] that are translated by 187,000 (±56,000) ribosomes [37]. The process of gene translation consumes a very large amount of energy and thus the problem of fine tuning the number of ribosomes and the translation rate should have a significant influence on the fitness of the organisms [6], [7], [20]. Specifically, increasing the translation rate of highly expressed genes (the ‘supply’) while decreasing the number of working ribosomes/ribosomal density (the ‘cost’) should improve the fitness of an organism. It was already suggested that there is selection for improving translation efficiency of highly expressed genes relatively to lowly expressed genes (see, for example, [6], [20]). By using our model, we can actually estimate the translation cost of highly and lowly expressed genes as the ratio between the translation rate and the average number of ribosomes working on the transcript. The number of proteins produced per unit time, per ribosome, for highly expressed genes (top 20%) is 0.000162/0.42 = 0.000386 (in arbitrary units). This number is 10% higher than that of the lowly expressed genes (lower 20%; 0.000125/0.36 = 0.000347). Again, this result demonstrates ‘optimality’: as highly expressed genes produce more mRNA molecules, decreasing the cost of translation should result in a much larger effect on the fitness of the organism.
Finally, the goal of this study was to model the process of translation elongation, emphasizing the effect of codon order. In the future, in order to decrease the gap between the predictions of our models and measurements of protein abundance, we intend to develop a more comprehensive model of this process. While promising strides in this direction were already made [57], [58], may features of the translation process are yet to be accounted for. Unfortunately, large-scale biological measurements of translation rates, initiation rates, tRNA levels, mRNA/protein degradation rates and many other quantities that are related to the process of gene translation are currently unavailable. Large scale measurements that are available (e.g. protein abundance) are related to the modeled process (Methods), but are indirect. This fact hinders the implementation and validation (as opposed to formulation) of more sophisticated models. In addition, it is important to note that the ability to predict measurements of protein abundance may also be hindered due to bias and noise in the current pool of existing data (see, for example, [17], [59]). As new data accumulates, the implementation of more comprehensive models will become possible and our understanding of the translation process will deepen further.
In the TASEP an mRNA transcript with codons is modeled as a chain of sites, each of which is labeled by the index , where . The first and last codons, , , are associated with the start and stop codons, respectively. At any time, t, attached to the mRNA are M(t) ribosomes. Being a large complex of molecules, each ribosome will cover codons. A codon may be covered by no more than a single ribosome. To locate a ribosome, we arbitrarily assume that the codon being translated is the one in the ‘middle’ of the ribosome. For example, if the first, (l+1)/2 codons are not covered, a ribosome can bind to the first codon on the mRNA strand, and then it is said to be “on codon ”. A complete specification of the configuration of the mRNA strand is given by the codon occupation numbers: if codon is being translated and otherwise. Note that when the (l−1)/2 codons before and after codon are covered by the ribosome that is on site . Since these codons are not the ones being translated, the codon occupations numbers for them are equal to zero.
We will now specify the dynamics of the TASEP model. A free ribosome will attach to codon with rate , provided that the first codons on the mRNA are empty. An attached ribosome located at codon will move to the next codon with rate , provided codon is not covered by another ribosome. In case (ribosome is bulging out of the mRNA strand) an attached ribosome will move to the next codon with rate .
In order to simulate this dynamics, we assume that the time between initiation attempts is distributed exponentially with rate . Similarly the time between jump attempts from site to site is assumed to be exponentially distributed with rate (The exponential distribution is of course, an approximation as the process of translating a single codon involves more than one step [1]). Note that in the case of the jump attempt is in fact a termination step. We define an “event” as an initiation, jump attempt, or termination step. From our definition it follows that the time between events is exponentially distributed (minimum of exponentially distributed random variables) with rate . Note that a jump attempt from codon can only be made if there is a ribosome translating this codon and hence the rate depends on the set of site occupation numbers.
The probability that a specific event was an initiation attempt is given by: . Similarly, the probability that a specific event was a jump attempt (or termination event) from site to site is given by .
At each step of the simulation, we determine the nature of the event and the time passed till its occurrence by these rules. The set of site occupation numbers are then updated accordingly and the simulation proceeds to the next event. For example if an initiation attempt was made, we check if the first codons on the mRNA are not covered. If so, we set , otherwise the attempt fails and remains as is. If a jump attempt from codon to codon was made, we check if site is not covered. If so, we set and , otherwise the attempt fails and remain as is.
Starting with an empty mRNA strand we simulate the system for 250,000 steps (events). The system is then simulated for an additional 1,000,000 steps where we keep track of the total number of terminations and the total time that have passed from the point this phase have started. The steady state rate of protein production was determined by dividing the number of termination events by the total time that has passed. The number of steps in the first and second stages was determined after observing that increasing the number of steps fourfold had a negligible effect on the predicted protein production rate.
The TASEP model mentioned above is a generalization (elongated particles and site dependent rates) of the simple TASEP model (see, for example, [60]). In the case of the ribosome flow model, we make two approximations. The first is coarse graining (dividing into chunks/sites), this approximation is quite common and was applied to various physical and biophysical problems. The second approximation is nothing but the mean field approximation. This means that in order to write the master equation for our model (Figure 1C) we have implicitly neglected the fact that there could be correlations between sites. We hence write approximate equations for the average (over many identical mRNA systems) occupation probabilities. Doing so, we assume that the probability that site i is occupied/empty and that site i+1 is occupied/empty is well approximated by the probability that site i is occupied/empty times the probability that site i+1 is occupied/empty. Although in general this is not always true, this approximation is also common in the TASEP literature.
Within the framework of the RFM, abortions were modeled by adding an abortion probability to the model. The abortion probability determines the percent of ribosome-ribosome collisions that will result in abortion, i.e., in premature detachment of the ribosome from the mRNA strand. Mathematically, abortion adds the following term to the model: where is the abortion probability. For every this term is added to the i-th and (i+1)-th rows of equation (1). This modification of the RFM corresponds to mutual abortion, i.e. for a situation where after an abortive collision both ribosomes will stop processing the mRNA transcript. Scanning different values for , we discovered that maximal correlations were obtain in the case of , i.e. in the limit were abortions due to ribosome-ribosome collisions are negligible.
In order to examine the steady state assumption (within the limitations of existing data), we analyzed the RFM model without it. Analysis was performed on the S. cerevisiae data where we simulated the model only for a time period proportional to the half life of the corresponding transcript [61]. In this case, steady state was not achieved and the translation rate was taken as the mean translation rate over the elapsing time period. This modification however, was unable to improve the predictive power of the model and in effect resulted in an opposite outcome.
Zhang model [33] similar to the TASEP model with the only change that the codon translation times are deterministic.
Here we would like to discuss the relation between translation rates and protein concentration/abundance. In what follows we will provide justification for the intuitive expectation that protein abundance should stand in high positive correlation with translation rates. Generally speaking, protein abundance levels are determined by a balance between protein production and degradation rates. Fixing the degradation rate, protein abundance levels will rise when the production rate is increased. Fixing the production rate, protein abundance levels will decrease when the degradation rate is increased. This said, one must also bear in mind that protein degradation rates are unavailable in most of the analyzed cases. And so, any current real data analysis is forced to average out the effect of protein degradation and focus on the contribution of the production rate to the determination of protein abundance levels.
Let denote the concentration of protein and let us assume that this protein is translated from a certain mRNA transcript whose copy numbers are denoted by . In general, the dynamics of this process may be described by the following differential equation: . Here and are the translation rate per mRNA molecule and the degradation rate of protein correspondingly. One possible choice for is: where is constant. Although this is a common first order approximation we will not base our conclusions on this particular choice and would only require that is a monotonically increasing function of the concentration . In general, the function depends on the protein , i.e. it can be different from protein to protein. Here however, we will replace the protein specific function with a genomic average degradation function which will be assumed monotonically increasing. Note that by definition, this function does not depend on the index .
The steady state solution of the above differential equation (with replaced by ) is: where is the steady state concentration of the protein . From the monotonicity of it follows that is a monotonically increasing function of .. This fact provides justification for the use of as a predictor for , i.e. one expects and to be positively correlated. Indeed, we have shown that this predictor performs very well, see Text S2. We will now show that itself can also be used as a predictor for , the advantage of this predictor is that it is solely based on the coding sequence and no additional information is required for its computation.
The set of mRNA copy numbers may generally depend on the set of translation rates , for example via the concentration of proteins that are involved in mRNA transcription and regulation. Fortunately, it is known that in endogenous genes translation rates are positively correlated with mRNA levels. Highly expressed genes are under selection to have higher mRNA levels, higher translation rate and higher protein abundance (note that this is not a causal relation; see, for example, [6]). Since mRNA levels are positively correlated with translation rates, higher values of do indeed imply higher values of and vice versa. Since in hetrogenouse gene expression mRNA copy numbers are usually independent of the mRNA variant of the protein, a similar trend is observed in this case as well. In building a predictor which is solely based on coding sequences, these empirical observation provide justification for using as a predictor for . Indeed, as we have demonstrated throughout the paper, this predictor out performs other commonly used predictors.
Our measure was based on the tAI [27]; as describe below, we adjusted it to our model:
Let ni be the number of tRNA isoacceptors recognizing codon i. Let tCGNij be the copy number of the jth tRNA that recognizes the ith codon, and let Sij be the selective constraint on the efficiency of the codon-anticodon coupling. We define the absolute adaptiveness, Wi, for each codon i as:The Sij-values can be organized in a vector (S-vector) as described in [27]; each component in this vector is related to one wobble nucleoside-nucleoside paring: I∶U, G∶U, G∶C, I∶C, U∶A, I∶A, etc.
Sensitivity analysis of the tAI of codons to Sij -values in S. cerevisiae showed that one codon (CGA) is extremely sensitive to these s-values. Increasing/decreasing the s-values by +−0.5 resulted in a change of up to one order of magnitude (usually much less) in all other codons. In the case of CGA, the change was up to 4000 times higher.
The tAI of this codon is relatively low and the model is sensitive to this value. Thus, we replaced the Wi of this codon by mean tAI of this codons over all possible changes (+−0.5) of Sij -values.
From Wi we obtain pi, which is the probability that a tRNA will be coupled to the codonThe expected time on codon is .
The expected time on a site is the sum of times of all the codons in the site.
The bottleneck was defined as the slowest window in a gene. The time of a window in the sum of times corresponding to its codons; the size of a window is 15 codons (the results were robust to small changes in the size of the window).
Figure S17 depicts the running time of our model as a function of λ and site. As can be seen, when the site size is larger than 10 codons, for all λ the typical running time for a gene is less than 0.1 second.
Measurements of ribosome densities in S. cerevisiae at a resolution of single nucleotides were downloaded from [14]. For comparison to the predictions of various models the profiles were aligned to the beginning of the coding sequences (similarly to the way it was done in [7], [20]). We computed and plotted the mean densities in sites of size 15 codons for each of the profiles (measured and predicted).
To estimate the dependence of the translation rate of genes (at their ‘working point’) on codon order, DTCO, we performed the following steps:
We call this quantity DTCO and we use it as a measure for the dependence of the translation rate on codon order.
To estimate the dependence of protein abundance on the codon order, DPCO, we performed the following steps:
To compute the ‘working point's of genes in a certain organism we first found the where the correlation between the mean predicted translation rate and protein abundance [16], [17], [62] is maximal. We computed the ratio (in percentages) between the mean genomic translation rate at this point and the mean maximal translation rate (for very large ); let Q% denote this value. (this value was 93%, 95%, and 99% in S. cerevisiae, S. pombe, and E. coli respectively)
The ‘working point’ of a gene in a certain organism is the where the translation rate of the gene is Q% of its maximal translation rate.
For each gene we computed the mean ratio between the synthetic version of the gene and its native version over 41 values of (between 0.0002 and 0.0094). The empirical p-value for the Spearman correlation is the probability that a random permutation of the two vectors will give higher correlation. It was computed by performing 100 such permutation and computing the Spearman correlation of each of them.
The Wilcoxon rank test that we used is a paired non-parametric test where we compared (1) the vector of distances between the predictions of our model and the real data (a distance for each point); (2) the vector of distances between the predictions of tAI and the real data; (3) the vector of distances between the predictions of Zhang model and the real data. We compared (1) to (2) and (1) to (3) and checked the following statistical question: “is there improvement (in terms of the distance between predicted and real data points) when a more sophisticated model (RFM) is used instead of a less sophisticated one (e.g. the tAI).
Jackknifing (see, e.g., [65]) was performed as described below.
Repeat 100 times:
Report the number of cases (0–100) that we get C = 25.
The result confidence level was 100 demonstrating a very high confidence.
|
10.1371/journal.pgen.1000241 | Essential Role of Chromatin Remodeling Protein Bptf in Early Mouse Embryos and Embryonic Stem Cells | We have characterized the biological functions of the chromatin remodeling protein Bptf (Bromodomain PHD-finger Transcription Factor), the largest subunit of NURF (Nucleosome Remodeling Factor) in a mammal. Bptf mutants manifest growth defects at the post-implantation stage and are reabsorbed by E8.5. Histological analyses of lineage markers show that Bptf−/− embryos implant but fail to establish a functional distal visceral endoderm. Microarray analysis at early stages of differentiation has identified Bptf-dependent gene targets including homeobox transcriptions factors and genes essential for the development of ectoderm, mesoderm, and both definitive and visceral endoderm. Differentiation of Bptf−/− embryonic stem cell lines into embryoid bodies revealed its requirement for development of mesoderm, endoderm, and ectoderm tissue lineages, and uncovered many genes whose activation or repression are Bptf-dependent. We also provide functional and physical links between the Bptf-containing NURF complex and the Smad transcription factors. These results suggest that Bptf may co-regulate some gene targets of this pathway, which is essential for establishment of the visceral endoderm. We conclude that Bptf likely regulates genes and signaling pathways essential for the development of key tissues of the early mouse embryo.
| While the chromatin of eukaryotes provides an efficient means to compact large amounts of DNA into a small nucleus, it renders the DNA relatively inaccessible. ATP-dependent chromatin remodeling complexes mobilize nucleosomes and provide a means to gain access to DNA in chromatin. While the biochemical functions of chromatin remodeling complexes is well-characterized, less is known of their biological functions. In this manuscript, we elucidate the biological functions of Bptf, a subunit of the NURF chromatin remodeling complex. Our studies show that Bptf is required for the establishment of the anterior–posterior axis of the mouse embryo during the earliest stages of development. To understand its functions in tissue differentiation, we generated and characterized Bptf-mutant ES cells. Mutant embryonic stem cells show significant defects in the differentiation of ectoderm, endoderm, and mesoderm. Genome-wide analysis of gene expression defects during differentiation has identified many Bptf-dependent pathways including key regulators of ectoderm, endoderm, and mesoderm differentiation. Moreover, we have identified critical functions for Bptf during the TGFβ/Smad-induced expression of visceral endoderm and mesoderm markers, an important signaling pathway in the early embryo. These results suggest that chromatin remodeling by Bptf regulates key signaling pathways in the early mouse embryo.
| The packaging of eukaryotic DNA into chromatin provides a general mechanism for the modulation of gene activity and DNA metabolism through alterations of chromatin architecture. The structure and composition of chromatin can be altered by a number of distinct pathways, including post-translational modification of histones, ATP-dependent remodeling of nucleosomes, and incorporation of histone variants [1]–[3]. ATP-dependent chromatin remodeling is catalyzed by the large and conserved SWI/SNF super family of multi-subunit chromatin remodeling enzymes that are classified into four major subfamilies (SWI/SNF, ISWI, CHD, and INO80), and distinguished by the common presence of a SWI2/SNF2-related catalytic ATPase subunit [4],[5].
The mammalian ISWI chromatin remodeling complexes contain either one of two related ISWI ATPases, Snf2l and Snf2h [6],[7]. The Snf2l ATPase is contained in two assemblies–NURF (Nucleosome Remodeling Factor), which is dedicated to the regulation of transcription, and the recently reported CERF [8],[9]. NURF is the founding member of the ISWI family of chromatin remodeling complexes, and was originally characterized in Drosophila [10]. Purified Drosophila NURF catalyzes ATP-dependent nucleosome sliding and promotes transcription from chromatin templates in vitro [9]. As shown by whole genome expression studies of mutants, NURF positively or negatively regulates transcription of several hundred Drosophila genes in vivo, including many genes important for fly development [11]. This is likely accomplished through recruitment of NURF301, the largest NURF subunit, by gene-specific transcription factors [11]–[13], and binding of a PHD finger of NURF301 to tri-methylated lysine 4 on histone H3 [14]. Human NURF contains the orthologs of three of four Drosophila NURF components–BPTF (Bromodomain PHD-finger Transcription Factor), the mammalian counterpart of NURF301, SNF2L (the ISWI ATPase) and RbAp46/48, a WD-40 repeat histone-binding protein found in several chromatin-related protein complexes [15]. Biochemical studies of human NURF have shown that it has similar properties to its Drosophila counterpart [15].
The physiological functions of an increasing number of mammalian chromatin remodeling complexes have been revealed by studies of mouse mutants for the catalytic ATPase. Mutations in Brg1, Brm, Chd4, Chd2, p400 and Etl1 have been shown to be required for proper embryonic development, hematopoiesis or postnatal survival [16]–[22]. A mutant for the Snf2h, one of the two murine ISWI ATPases, revealed severe proliferation defects in the early embryo, resulting in a peri-implantation lethal phenotype [23]. Given the presence of Snf2h in multiple chromatin remodeling complexes, the assignment of biological phenotypes to different enzyme complexes can be problematic [6],[7].
By analysis of mutations of unique subunits it is possible to identify the biological functions of different Snf2h-containing complexes. This has been accomplished for the Drosophila ISWI complexes. Studies of mutants for Drosophila NURF301, which is exclusive to the NURF complex [12], have revealed a late larval-lethal phenotype and mis-expression of homeotic selector genes and genes involved in the response to heat stress, cytokine and steroid hormone signals [11],[13]. These phenotypes do not overlap with those observed for mutants for Drosophila ACF1 (a component of the ISWI-containing complexes ACF and CHRAC) which are impaired in the establishment and/or maintenance of transcriptional silencing in pericentric heterochromatin and in repression by Polycomb-group genes [24].
To elucidate the biological roles of Bptf-containing complexes we have generated embryonic stem cell and mouse mutants for Bptf [15]. Our studies show that Bptf mutant phenotypes begin to manifest just after implantation stage, and mutant embryos are completely reabsorbed by embryonic day (E) 8.5. Genetic and molecular analysis in embryonic stem cells and the mouse suggest a role for Bptf in the development of visceral endoderm (VE) of the early mammalian embryo. We propose that Bptf is required for the development of the VE, and more importantly the distal visceral endoderm (DVE), in part through regulating cellular proliferation, the expression of homeobox-containing transcription factors and pathways regulated by the Smad transcription factors, a major conduit for cell signaling in development. These findings suggest a model in which the activities of Bptf-containing complexes, likely the NURF remodeling complex, regulate cell proliferation and embryonic development and therefore are essential in the post-implantation embryo.
BPTF, the mammalian ortholog of Drosophila NURF301, is a large, multi-domain protein that is apparently exclusive to the mammalian NURF complex (Figure S1A) [12],[15],[25]. An initial characterization of Bptf showed it to be nuclear in the P19 embryonic carcinoma cell line and can exist in at least two electrophoretic variants, which we termed Bptf-H and Bptf-L (Figure S1B, Figure S1C). As previously reported, Bptf is highly expressed in testis, spleen, brain and to a lesser extent in kidney by Western blotting (Figure S1D). Interestingly Bptf is highly expressed during embryonic development, and expression substantially declines upon birth (Figure S1E). The high levels of Bptf expression in the mouse embryo suggest it may have essential functions during embryonic development.
To elucidate the biological functions of mammalian Bptf during mammalian embryonic development we generated two mutant mouse lines. One line, designated as BptfXG023, was derived from an ES cell line carrying an in-frame gene-trap vector insertion between exons 15 and 16 of Bptf (Figure S2A, Figure S2B, S2C, S2D, S2E, S2F) (XG023; http://www.genetrap.org/) [26]. We identified the precise junction of the insertion site of the gene trap by DNA sequence analysis of the corresponding PCR products, and confirmed by RT-PCR that trapping of Bptf mRNA into vector sequences leads to loss of RNA splicing between exons 15 and 16, and reduced expression of Bptf sequences 3′ to the insertion site (Figure S3A, Figure S3B). In addition, we confirmed by 5′-RACE that the insertion resulted in an in-frame fusion between Bptf and β-galactosidase-neomycin phosphotransferase (β-Geo) sequences of the gene-trap vector. (Figure S3C). Consistent with previous findings, Northern blotting of adult tissue RNA showed that both wild-type Bptf and the Bptf–β-Geo fusion alleles are specifically expressed at high levels in the testis and at moderate levels in the lung, spleen, and brain (Figure S3D) [25]. These identical RNA expression patterns initially suggest that β-galactosidase is a faithful reporter of Bptf expression in adult tissues. These results indicate that the BptfXG023 mutation creates a truncated Bptf–β-Geo fusion carrying the N-terminal 1978 residues of the 2903-residue Bptf open reading frame, which eliminates the conserved glutamine rich region, PHD finger, and bromo-domains.
The second line, designated as BptfΔExon2, was generated by targeting loxP sites flanking exon 2 of Bptf (Figure S2A, Figure S2G, Figure S2H, Figure S2I, Figure S2J, Figure S2K, Figure S2L, Figure S2M, Figure S2N). The expression of the BptfΔExon2 allele was assessed by RT-PCR with the use of primer sets which amplify sequences on the 3′ end of the Bptf transcript. We found that the BptfΔExon2 allele slightly decreases expression of Bptf at the RNA level (Figure S4A). However, amplification and sequencing of the BptfΔExon2 mRNA from exons 1 to 8 shows an out-of-frame mutant mRNA, indicating that BptfΔExon2 behaves as a loss-of-function allele (Figure S4B, Figure S4C).
We intercrossed heterozygous mice to determine the biological function of Bptf during mouse development. With 210 mice genotyped for the BptfXG023 allele and 75 mice for the BptfΔExon2 allele, we did not find any surviving homozygous mice at weaning, indicating that Bptf is required for mouse development. An analysis of E6.5 to E18.5 progeny derived from heterozygous intercrosses confirmed that the homozygous mutant phenotype is embryonic-lethal between E7.5 to E8.5, with 100% penetrance (Table S1).
To identify defects in the embryonic development of BptfXG023 and BptfΔExon2 homozygotes, we performed whole mount examinations of mutant embryos. Growth defects increasing in severity from E5.5 to E7.5 were observed (Figure 1A) (Figure S5A). Dissections conducted at E8.5 and E9.5 were uninformative, because a majority of the embryos had been completely reabsorbed (data not shown). We crossed BptfXG023 and BptfΔExon2 heterozygous mice to generate the trans-heterozygous BptfXG023/BptfΔExon2 embryos. The trans-heterozygote recapitulated the growth defects of the BptfXG023 and BptfΔExon2 homozygotes, indicating that the two mutations are functionally equivalent (Figure S5B). To determine whether the developmental defects originated before implantation, we harvested E3.5 blastocysts and performed blastocyst outgrowth assays. We observed normal E3.5 homozygous BptfXG023 and BptfΔExon2 blastocysts and normal outgrowths from the blastocysts after tissue culture for 5 days, suggesting that either Bptf is not essential for pre-implantation development or that the maternal Bptf protein or mRNA can mask pre-implantation phenotypes (Figure S6). The masking of pre-implantation phenotypes by maternal Bptf is possible because it is highly expressed in oocytes [27],[28].
To further investigate the basis of the early embryonic lethal phenotype, we performed a histological analysis of BptfXG023 mutant embryos at E5.5, E6.5 and E7.5. Mid-sagittal sections of E5.5, E6.5 and E7.5 embryos stained with hematoxylin and eosin (H&E) showed a distinct proximal-distal (P-D) axis and the development of visceral endoderm but a significant decrease in size of the embryonic and extra-embryonic tissues. This was particularly evident in the embryonic ectoderm at E6.5 and E7.5 (Figure 1B). By E6.5, mutant embryos showed clear developmental defects. Although there was a clear boundary between extra embryonic and embryonic tissues, the absence of a primitive streak and any discernable mesoderm suggests that the anterior-posterior (A-P) axis did not form (Figure 1B).
A reduction in cell number can be a consequence of decreased cell proliferation, increased cell death (apoptosis), or a combination of both processes. To ascertain the extent of programmed cell death, we performed TUNEL assays on E5.5, E.6.5 and E7.5 homozygotes and found no increased numbers of apoptotic cells when compared to controls (Figure 1B). As a measure of cell proliferation we monitored phosphorylated histone H3 levels in the conceptus by immuno-histochemistry (Figure 1B). E5.5, E6.5 and E7.5 homozygotes showed ∼40–50% decrease in phosphorylated histone H3 levels indicating that a decrease in cellular proliferation may contribute to the mutant phenotype (Figure S7).
Subsequent to implantation of the mouse blastocyst there is rapid proliferation of the egg cylinder, which consists of three cell types: the more proximal extra-embryonic ectoderm, the more distal embryonic ectoderm or epiblast, and an outer layer of visceral endoderm [29]. The visceral endoderm originates from the primitive endoderm, a layer of cells organized at E4.5, which is composed of cells from the ICM of the E3.5 blastocyst expressing Gata6 but not Nanog [30]. At ∼E5.5, a specialized cluster of endoderm cells, the DVE, arises at the distal tip of the embryo. DVE cells migrate toward the prospective anterior, to form the anterior visceral endoderm (AVE). DVE/AVE cells secrete molecules such as cerberus (Cer1) and Lefty1, antagonists of the Transforming Growth Factor β (TGFβ)-related protein Nodal [29]. These antagonists restrict the activity of Nodal to the posterior pole of the embryo at E6.0 [31]. The primitive streak forms at E6.5, indicating that gastrulation has begun, and gives rise to the mesoderm and definitive endoderm germ layers [29].
As a first step in the molecular analysis of Bptf in embryonic development we examined the expression of Bptf using in situ RNA hybridization. We observed expression in the inner cell mass (ICM) and primitive endoderm at E4.5 and all embryonic tissues at E5.5 and E6.5. Interestingly we do not observe Bptf expression in the visceral endoderm at E5.5 and E6.5 (Figure 2A). We also monitored the activity of the β-galactosidase moiety of the Bptf–β-Geo fusion protein in heterozygous mice. Consistent with our in situ analysis a histochemical analysis of whole mounts showed that Bptf–β-Geo is expressed in the embryo proper at E5.5, E6.5 and E7.5 (Figure S8A, S8B, S8C, S8D, S8E, S8F, S8G). Further analysis of histological sections revealed that Bptf–β-Geo expression at E7.5 is primarily confined to the embryonic ectoderm, with reduced levels in mesoderm and no expression in the visceral endoderm (Figure S8A′, Figure S8B′). At subsequent stages, from E7.5 to E13.5, histochemical analysis of whole mounts showed widespread Bptf–β-Geo expression in the developing embryo (Figure S9A, Figure S9B, S9C, S9D, S9E, S9F). This temporal correlation between the earliest stages of Bptf expression and the stages when the mutant phenotype is revealed, suggests that there could be an essential requirement for Bptf as early as E4.5.
Our histological analysis suggests that Bptf mutants are defective in establishing an A-P axis. Apparent defects in A-P axis can be due to defects in the establishment or migration of the DVE [29]. To monitor the development of the DVE and its transition to the AVE, we analyzed the markers Otx2, Lefty1, Cer1, Hesx1, Hex1, Gata6, Nanog and Nodal by in situ hybridization in E4.5 to E6.5 mutant embryos [32]–[37].
An analysis of pre-implantation embryos suggested that Bptf mutants specify a functional primitive endoderm and ICM. The primitive endoderm of Bptf mutant embryos was found to express the primitive endoderm markers Gata6, Lefty1, and Hex1 at comparable levels to wild type controls (Figure 3B). The expression of these markers suggests that Bptf mutants are not defective in the differentiation of the primitive endoderm, the precursor of the visceral endoderm of E5.5 and later stage embryos. Consistent with a functional ICM, we observed normal expression of the pluripotency marker Nanog in Bptf mutants compared to controls (Figure 2B). Taken together, these results indicate that Bptf is not required for the specification of the primitive endoderm and the ICM in E4.5 embryos.
To assess the specification of the VE and DVE, we monitored the expression of Cer1, Hex1, Gata6, and Nodal in E5.5 Bptf mutant embryos. We observed that the markers Cer1, Hex1 are significantly reduced in Bptf mutants relative to the wild type controls (Figure 2B). The absence of expression of these markers indicates that the DVE does not form in the absence of Bptf. Interestingly Gata6 expression, normally expressed only in the VE at E5.5 and E6.5, is absent in the VE but present in the epiblast in mutants at E5.5 (Figure 2B). This suggests that Bptf could also have roles in specifying the VE as well as the DVE. A key regulator of DVE specification is Nodal. Nodal expression is found as early as E4.5 but is not significantly expressed in the epiblast and VE until implantation is well underway at E5.0 [36],[38]. In E5.5 Bptf mutant embryos we observe normal expression levels of Nodal in the epiblast and VE as in wild type embryos (Figure S10A). This E5.5 pattern of Nodal expression continues into E6.5 in Bptf mutant embryos (Figure S10B).
To examine the development of the AVE we monitored the expression of Cer1, Otx2, Hesx1, Lefty1 and Hex1 in E6.5 Bptf mutant embryos. Otx2 expression is required for the migration of the DVE to establish the AVE. We observed lower expression of Otx2 in the epiblast of Bptf mutant embryos relative to there wild type controls (Figure S11A). As expected we did not observe expression of AVE markers Cer1, Hesx1, Lefty1 and Hex1 in Bptf mutant embryos (Figure S11A). Combined with our analysis of E5.5 embryos our results strongly suggest that Bptf is required for the speciation of the DVE and the AVE.
Since Bptf mutant embryos are unable to form a functional DVE and AVE, we anticipated that they should be defective in specifying the primitive streak and differentiating mesoderm and definitive endoderm. Several critical transcription factors and signaling molecules such as T, Lhx1, Fgf8, Gsc, Foxa2, Nodal, and Cripto (Tdgf1) serve as effective markers for development of the primitive streak in the gastrulating embryo [34], [39]–[42]. Our analyses revealed that T, Lhx1, Fgf8, Gsc, and Foxa2 were undetectable at E6.5, and in the case of T, Fgf8 but not Lhx1, were delocalized in expression at E7.5 (Figure S11A, Figure S11B). The absence of expression of primitive streak markers at E6.5 confirms our histological analyses, and further supports the observation that gastrulation and mesoderm formation do not occur in Bptf mutants. Interestingly, the delocalized Nodal and Cripto expression patterns observed in the Bptf mutants at E6.5 are highly reminiscent of their expression patterns prior to the establishment of the DVE (Figure 2B) (Figure S11A) (Figure S10A, Figure S11B) [34],[39]. Taken together, the data suggests that Bptf mutant embryos arrest at a stage prior to DVE formation (<E5.5).
Recent models propose that the extra-embryonic ectoderm supports the growth of the embryo and is a source of signals for A-P axis establishment [29]. To address whether the developmental defects of Bptf mutants are caused by defective extra-embryonic ectoderm or by a lack of appropriate growth signals in the epiblast, we monitored the expression of the extra embryonic ectoderm (Bmp4, Erbb2, Fgfr2), trophectoderm (Mash2), angiogenesis (Vegf, Flk1) and a cell cycle regulator (JunB) markers [43]–[47]. We find Bmp4, Mash2, Erbb2, and Fgfr2 to be expressed normally in the mutant embryos at E6.5 and E7.5 (Figure S11A, Figure S11B). This suggests that the growth defects observed in Bptf embryos are not due to gross defects in the specification the extra-embryonic tissues. We also observe little to no expression of the angiogenesis markers Vegf and Flk1 in the extra-embryonic tissues at E7.5 (Figure S11B). However, expression of the cell cycle regulator JunB is increased in mutant embryonic ectoderm when compared to wild-type controls (Figure S11A). JunB is a member of the Ap-1 family of transcription factors which acts as a negative regulator of the cell cycle [48]. This up-regulation of JunB is consistent with our observations that Bptf mutants have reduced cellular proliferation (Figure 1B) (Figure S7).
In summary, our analysis of lineage markers by in situ RNA hybridization has revealed an essential role for Bptf in specifying the VE and the DVE of the E5.5 post implantation embryo. These defects likely lead to the observed absence of an AVE and primitive streak in E6.5 embryos. The absence of these developmental organizers arrests the growth of the embryo prior to gastrulation and is likely a major cause of the early embryonic lethal phenotype of Bptf mutant embryos.
To further explore the role of Bptf in cell differentiation we generated Bptf knockout mouse ES cells and examined their development in vitro and in vivo. By gene targeting and transient Cre expression we were able to generate eight independent homozygous BptfΔExon2 knockout ES cell lines with a euploid karyotype (Figure S12A). We observed varying degrees of reduced Bptf transcript levels by Northern blotting in mutant cell lines compared to that of wild type controls (Figure S12B). An analysis of BptfΔExon2 mRNA from exons 1 to 8 in wild type and knockout cell lines shows the message to be out of frame resulting in no observable protein in the knockout cell lines (Figure S12C, Figure S12D) (data not shown).
To address the possibility the Bptf is essential for cell viability and proliferation we measured the doubling time of knockout Bptf ES and MEF cell lines (Figure S13). Both the Bptf knockout ES and MEF cell lines were viable but exhibited slightly reduced cellular proliferation (Figure S14). These results show that Bptf is not required for cell viability and is only marginally required for cellular proliferation.
We measured the differentiation potential of the Bptf knockout ES cells in vitro as embryoid bodies and in vivo as teratomas. Bptf wild type and two independent knockout cell lines were subcutaneously injected into NOD/SCID mice and allowed to form teratomas over 8 weeks. Wild type ES cells formed teratomas in 8/11 injected mice. Bptf knockout lines did not form any observable tumors in 10 injections (data not shown). This study demonstrated that Bptf is essential for one or more biological processes including cell viability, proliferation or differentiation in the animal.
We next utilized an embryoid body analysis to monitor differentiation in Bptf knockout cell lines to ectoderm, mesoderm and endoderm cell lineages. Analysis of the mutant embryoid bodies showed little evidence of apoptosis by TUNEL, and similar densities of PCNA and phosphorylated histone H3 positive cells (Figure S15A). We did observe that Bptf knockout embryoid bodies were slightly smaller in size and did not form any observable endoderm (Figure S15B, Figure S15C). These results indicate that Bptf is not necessary for cellular survival or proliferation under conditions of embryoid body differentiation but is essential for differentiation of endoderm and possibly other advanced tissue lineages.
To investigate the differentiation defects of Bptf knockout embryoid bodies at a molecular level we monitored the transcription of well documented markers of endoderm, mesoderm and ectoderm differentiation (Figure 3A–3C). We observed minimal Bptf dependence for the primitive ectoderm markers FGF5 and Otx2 (Figure 3A). However, we did observe significant defects in Nestin transcription, a marker of neural stem cell progenitors derived from the primitive ectoderm (Figure 3A). These results indicate that primitive ectoderm lineages are less dependent on Bptf than the more differentiated lineages. As anticipated we observed severe defects in the expression of mesoderm and endoderm markers during the differentiation time course. We observed little to no activation of mesoderm markers T, FGF8, Evx1, Wnt3, and Gsc which are significantly activated in wild type cultures by day 5. (Figure 3B). Similarly, endoderm markers Sox17, Cer1, Hnf4a, and Foxa2 show severe expression defects in Bptf mutant embryoid bodies compared to controls (Figure 3C). In contrast to the large differences in expression of mesoderm and endoderm markers we observed less than two fold changes in expression of cell cycle regulators and pluripotency markers with the exception of Cyclin D1 (Figure S16A, Figure S16B). These defects in transcription and differentiation were rescued for two independent knockout lines by retargeting exon 2 to the BptfΔExon2 locus using the targeting vector. Our ability to rescue the expression defects of T, Gsc, Sox17 and Cer suggest that the observed phenotypes are due to Bptf mutation (Figure S17A, Figure S17B, Figure S17C) (data not shown). Taken together these results suggest that Bptf is essential for the formation of mesoderm, endoderm and more differentiated ectoderm lineages in the embryoid body.
We chose a microarray approach to investigate any differentiation defects of Bptf knockout ES cells in undifferentiated and early differentiation states. Differentiation was induced by LIF withdrawal (LIF−) or retinoic acid (RA) for three days before harvesting RNA. A comparison of Bptf-dependent genes by Venn diagram and manual clustering identifies six categories; those being affected under only one of the growth conditions (LIF+, LIF− or RA Regulation), those being affected in all three conditions (Constitutive Regulation), a class of genes which were regulated in the same direction in two of three conditions (Complex Regulation) and genes with mixed dependence (Mixed Regulation) (Figure 4A and 4B) (Dataset S1).
From our microarray analysis we observed large changes in gene expression under conditions of maintained pluripotency (LIF+) and particularly under conditions of differentiation (LIF− and RA) (Figure 4A and 4B). As expected many of the essential markers of early embryonic tissue differentiation that are dependent on Bptf in embryos are also Bptf-dependent in ES cells (Figure 4C). These markers include; pluripotency regulators Sox2, c-Myc, Nanog, visceral endoderm markers Lefty1, Cer1, Hex1, Foxa2 and the primitive streak markers Gsc, Lhx1, Wnt3, Fgf8, T (Figure 4C). These results reinforce the view that Bptf regulates the development of ectoderm, endoderm and mesoderm in both the early embryo and in ES cells.
A gene ontology (GO) analysis of Bptf-dependent expression datasets revealed an over representation of genes with “transcription factor activity”, genes involved in the biological processes of “development” and “morphogenesis” and the cellular processes of “cell death” and “cell proliferation” (Figure S18). Notable gene clusters include the consistent activation of genes correlated with “nervous system development” in all datasets, the repression of MHC I and II receptors during LIF−differentiation, the activation of genes correlated with “cytoskeletal components” during RA differentiation (Figure S18). In addition, we observed a striking over-representation of homeobox transcription factors within the “transcription factor activity” annotation. The homeobox-containing genes were almost exclusively up-regulated in each of the expression categories, and in some cases include almost the entire Hox gene cluster, indicating that Bptf is required for their repression in ES cells (Figure S19A, Figure S19B).
From our analysis we also observed that Bptf-dependent gene targets are more likely to be actively regulated genes, repressed in the presence of LIF or RA differentiation and conversely activated under conditions of LIF differentiation (Figure S20A, S20B, S20C). In support with these observations we observed histone modifications in Bptf knockout ES cell lines consistent with a repressed transcriptome under LIF+ growth conditions (Figure S20D). Interestingly some Bptf-dependent genes cluster together in the genome (Figure S21).
A diagnostic defect of Bptf embryos is an inability to form the DVE. In the absence of the DVE, the AVE cannot form preventing the necessary signals for the specification of the primitive streak. To further investigate the functions of NURF in the early embryo we focused on Smad mediated signaling pathways. Smad mediated signaling pathways are essential for the formation of the DVE in the embryo and the induction of mesoderm in differentiating ES cells, two prominent phenotypes of our in vivo and in vitro studies on Bptf [49],[50].
The most prominent ligand activating the Smad transcription factors in the early embryo is Nodal [51]. Nodal, and the closely related ligand activin, bind to type I and II TGFβ receptors resulting in the phosphorylation of the type I receptor. Phosphorylation of the type I receptor activates a kinase domain which phosphorylates Smads2/3. The phosphorylation of Smad2 or Smad3 transcription factors promotes interactions with Smad4 and triggers the translocation of the Smad complex into the nucleus [51]. Once in the nucleus, the Smad complex interacts with DNA sequence specific transcription factors to promote the regulation of Smad target genes [52].
Accordingly, we monitored the dependence of Smad-responsive genes on the presence of Bptf in ES cells. ES cells readily responded to activin-A as monitored by the phosphorylation of Smad2 (Figure S22A, Figure S22B). From these experiments we identified a number of Smad-dependent genes which completely or partially require Bptf for full activation. Genes requiring Bptf for full activation include Cer1, Gsc and T (Figure 5A). Genes which partially require Bptf include FGF8, Lefty1 and p21 (Figure 5A) (Figure S23).
To understand the relationship between Bptf and CBP/p300, known co-activators of the Smad transcription factors, we knocked down both Bptf and CBP/p300 using siRNA technology and monitored the activation of the Smad responsive genes Lefty1, FGF8, Gsc, T, and Cer in ES cells (Figure S22C). As in the BptfΔExon2 knockout ES cell lines we observed that each of these genes are dependent on Bptf for full activation (Figure 5B). While some genes differ in there requirement for Bptf, they are all dependent on CBP/p300 for activation (Figure 5B). Our results demonstrate that a genetic knockout and siRNA mediated knockdown of Bptf result in defects in the activation of Smad responsive genes to varying degrees. Taken together, these results indicate that Bptf, like CBP/p300, acts as a co-activator of Smad responsive genes in ES cells.
We also used the embryonic carcinoma cell line P19 to complement our findings with the Bptf knockout ES cells. We co-transfected P19 cells with DNA plasmids carrying four minimal Smad binding elements (SBE), or three activin response elements (ARE), linked to a core promoter and a luciferase reporter gene. The ARE and SBE elements were previously found to be the minimal Smad-responsive elements [53],[54]. Under conditions of Bptf knockdown, we observed a significant reduction in luciferase activity from both reporters (Figure S24A, Figure S24B).
To simulate Smad signaling in a different way we co-transfected combinations of Smads 2,4 with constitutively active TβRI (ALK5), the type I receptor for the TGFβ signaling pathway (Figure S24C) [55]. Using this system we observed efficient reduction in the activation of the SBE regulated luciferase reporter gene with a siRNA to Bptf but not a mock siRNA control (Figure S24D). We repeated the experiments using multiple unique siRNAs and measured the transcription of endogenous TGFβ regulated genes. In these experiments we used three individual Bptf siRNAs which were effective in knocking down protein expression after 2 days of culture (Figure S24E). We then stimulated the P19 cells with TGF-β1. Like nodal and activin-A, TGF-β1 stimulates the phosphorylation of Smad 2/3 through the dimerization and activation of similar type I and II receptors. We similarly observed a significant reduction of Cer1 and T transcription in Bptf depleted cells upon Smad2/3 activation with TGF-β1 (Figure S24F). We also used the human breast cancer cell line MCF10CA1 in similar assays to determine if BPTF could play a role in Smad signaling in humans [56]. In these experiments we used BPTF siRNAs which were effective in knocking down protein expression in MCF10CA1 cells after 2 days of culture (Figure S24G). We then stimulated the MCF10CA1 cells with TGF-β1 for 1 hour. We observed a significant reduction of PAI-1 but not SMAD7 induction suggesting that, as in the mouse, BPTF could regulate a subset of Smad responsive genes in humans (Figure S24H).
We next investigated whether the interaction between the BPTF-containing NURF complex and the Smads is direct or indirect using pull-down assays. Experiments with bacterially expressed GST-Smad2 or GST-Smad 3 showed an interaction to a degree between recombinant NURF complex and each of the Smad transcription factors, but not the GST or GST-βcatenin controls (Figure 5C). Smad transcription factors are composed of functional domains at the N-terminus (MH1 domain) and C-terminus (MH2 domain). The N-terminal+linker and C-terminal regions of Smad2 were used in similar pulldown experiments. We observed that recombinant NURF complex interacts specifically with the MH2 domain of Smad2 (Figure 5C). This interaction was also observed for the Bptf and Snf2l components of native NURF complex from crude ES cell nuclear extracts (Figure 5C). We also confirmed the reported interaction of the C-terminal MH2 domain of the R-Smads with the co-regulator CBP (Figure 5C) [57]. Hence, our results suggest that NURF, like p300 and CBP maybe recruited to the promoters of TGFβ responsive genes through direct interactions with the Smad transcription factors.
Our current model proposes that Bptf-containing complexes like NURF are recruited to the promoter of Smad regulated genes through direct interactions with the Smad transcription factors. To test this model, we initiated chromatin immuno-precipitation (ChIP) experiments to detect the Snf2l component of the NURF complex at Lefty1. As anticipated, we observed significant enrichment of Snf2l at the Neural Plate Specific Enhancer (NPE), a region which contains putative FAST and Smad transcription actor binding sites, of the Lefty1 promoter [58]. This enrichment was dependent on activin-A stimulation and on the presence of Bptf (Figure 5D). Snf2l enrichment correlates with the presence of the activating histone modification H3K4me3 at the 5′ UTR of Lefty1 (Figure 5D). These results are consistent with Bptf recruitment to the promoter of Lefty1 through the Smad transcription factors.
In this work we report a post-implantation lethal phenotype for mutations in Bptf, the previously characterized largest subunit of the NURF chromatin remodeling complex [15]. In the early embryo Bptf is expressed by E4.5 in the ICM and primitive endoderm. At later stages of development Bptf is expressed in both embryonic and extra-embryonic ectoderm by E5.5. Following gastrulation, Bptf is widely expressed in all germ layers to E13.5. Bptf expression is essential for early development because homozygous Bptf mutant embryos are reabsorbed by E8.5. No overt defects were observed in the mutant E3.5 or E4.5 embryo or its ability to proliferate in culture, suggesting that Bptf embryos undergo a normal initial specification and proliferation of the ICM, primitive endoderm and trophectoderm. However, mutant embryos exhibit diminished proliferation post-implantation as shown by defects in size of both extra-embryonic and embryonic tissues and decreased phosphorylated histone H3 staining. A histological analysis of mutant embryos at E6.5 and E7.5 revealed that they develop a VE but do not form a primitive streak or differentiate mesoderm. To investigate the causes of the defect in gastrulation, we monitored the expression of key markers prior to and during gastrulation in the embryo. As anticipated we failed to observe expression of primitive streak markers T, Foxa2, Gsc, Fgf8 or the posterior localization of Nodal and Cripto expression. Defects in the expression of primitive streak markers and the delocalized expression of Nodal and Cripto are likely due to the absence of the DVE/AVE. Defects in the DVE/AVE were confirmed by observing significantly reduced expression of the markers Cer1, Hex1, Lefty1 and Hesx1 in E5.5 and E6.5 embryos. Defects in Gata6 expression at E5.5 suggest that the defects in DVE specification are accompanied by general defects in VE specification. From these studies we conclude that a critical function for Bptf during mammalian development is directly or indirectly to specify the VE and DVE after implantation.
To identify Bptf-dependent gene targets we employed a microarray based approach on Bptf knockout ES cells during the early stages of differentiation and an embryoid body model. We discovered a role for Bptf in the regulation of gene clusters essential for development, morphogenesis, nervous system development and cell death and proliferation. Interestingly many transcription factors, primarily the homeobox-containing genes, are dependent on Bptf for their proper repression during undifferentiated and differentiated states. This dependence is interesting as NURF has been shown to be a activator of Hox gene transcription in more differentiated tissues in Drosophila and the mouse [13],[15]. As expected many markers of ectoderm (Nestin, Fgf5), mesoderm (Gsc, Lhx1, Fgf8, Tbx6, Wnt3) and endoderm (Lefty1, Cer1, Nodal, Hesx1) cell lineages require Bptf for their expression. These gene targets corroborate well with those identified from our in vivo studies on Bptf mutant embryos further supporting the conclusion that Bptf is essential for the development of ectoderm, mesoderm and endoderm. This also suggests that our microarray dataset obtained from ES cells is a reasonable approximation of the expression defects occurring in Bptf mutant embryos in vivo.
The functions for Bptf in the early embryo are undoubtedly complex. Our unbiased analysis of gene targets by microarray revealed many potentially Bptf-dependent differentiation pathways. Most relevant to this study include the regulation of the cell cycle and the differentiation of mesoderm and endoderm lineages, specifically the VE and DVE. The underlying cause for these defects could largely be due to the loss of chromatin associated complexes like the NURF chromatin remodeling complex. In the case of NURF the defects could be direct, as a remodeling activity at the promoter of genes necessary for cellular proliferation and differentiation, or indirect through the deregulation of master regulators of development like the homeobox-containing transcription factors. Moreover the underlying mechanism of Bptf action as a co-activator of some genes and a co-repressor of others is unclear. Further studies of nucleosome positioning and chromatin structure in mutants should clarify these possibilities.
In addition to system-wide defects with Bptf deletion, there could be specific defects in individual signal transduction pathways within the embryo or in the ability of the embryo to receive growth signals from the extra-embryonic tissues or the surrounding decidua. To identify potential Bptf-dependent signaling pathways we focused on its role in specifying the DVE. The pre-gastrulation embryo uses three well-known signaling pathways, the WNT/β-catenin, FGF/MAPK and Nodal/Smad pathways, to establish A-P asymmetry [29]. The three pathways can be distinguished by different A-P phenotypes. In mutants of FGF/MAPK signaling the epiblast has severe proliferation defects, do not specify primitive endoderm, are quickly reabsorbed and the blastocysts do not outgrow when grown in culture [30],[44],[59],[60]. Mutations in the WNT/β-catenin, but not Nodal/Smad pathways, develop the DVE and in some cases the AVE [50], [61]–[63]. The ability of the Bptf embryos to form blastocyst outgrowths, specify the primitive endoderm, but not form the DVE is reminiscent of mutants in the Nodal/Smad signaling pathway rather than a defect in FGF/MAPK or WNT/β-catenin signaling (Table S2).
Because Bptf has been associated with the NURF complex, a known regulator of transcription, the data suggests that Bptf is required for the expression of gene targets in the developing VE and DVE. In support of this hypothesis, we showed that Bptf is required for the regulation of endogenous promoters and Smad responsive promoter elements in ES, P19 and MCF10CA1 cells in tissue culture. Smad-dependent gene targets include those essential for cell proliferation (p21) and those essential for DVE function (Cer1, Lefty1) and primitive streak (T, Fgf8, Gsc). Moreover, pulldown assays showed that components of the NURF complex have direct interactions with the Smad transcription factors and it is recruited to the promoters of Smad regulated genes under conditions of activation. Taken together, our data suggest that Bptf can directly regulate Smad regulated genes, likely through the functions of the NURF remodeling complex, via recruitment by the Smad transcription factors (It is also possible that other as yet unidentified Bptf-containing complexes distinct from NURF function in this pathway). To address the possibility that the effects on the Nodal/Smad pathway are indirect we monitored the expression of key components of the pathway in embryos. We did not observe significant changes which could explain the observed defects in Smad signaling in the early embryo or during embryoid body differentiation in Bptf mutants (Figure S25).
Consistent with these findings, the ISWI ATPase, a component of Drosophila NURF, ACF and CHRAC has been reported to be important for transmitting Dpp/TGFβ signals to stem cells in the Drosophila ovary [64]. However, we wish to emphasize that the Bptf mutation likely affects many different pathways and biological processes, each of which may contribute to biological phenotypes. The challenge for the near future will be to uncover each of the many functions of Bptf in mammalian chromatin biology.
The RPCI21 mouse PAC library (MRC Genomic Resource Center, England) was screened using a random hexamer labeled probe to Bptf exon 2. Blotting was performed in hybridization bags using 0.25 M sodium phosphate pH 7.2, 1 mM EDTA, 7% SDS, at 65°C overnight with rocking. Blots were washed 5 times for 10 min with 0.25× SSC, 0.1% SDS at 65°C. Eight positive clones were identified using X-ray film as; 367-I21, 373-D5, 402-C16, 402-E15, 426-P15, 540-B9, 625-D23, 582-P16. Clone 367-I21 was confirmed by Southern blotting using Eco RI and Sal I digests and Bptf exon2 probe. Genomic sequence for the construction of the targeting vector was retrieved into bluescript SKII (Stratagene) and the integration of loxP sites and Neo selectable marker was performed using recombineering technology described previously [65]. The sequence of the targeting vector is available upon request.
Linearized vector was electroporated into CJ7 ES cells and 71 individual Neo resistant and HAT resistant clones were isolated according to previously published procedures [66]. Clones were screened for successful Bptf targeting using Eco RI, probe 29–31; Bam HI, probe exon2; and Sca I, probe exon2. 16 of 71 isolates were found to be correct for a recombination frequency of 22%.
To create heterozygous, conditional homozygous, homozygous knockout and rescue Bptf ES cell lines we transiently expressed Cre followed by retargeting. The piCre expression vector was electroporated into clone 7010 and conversion from Bptf/BptfFloxedNeo to Bptf/BptfΔexon2 was first screened for by PCR then confirmed by Southern blotting. Three clones were identified using this strategy, and clone 7004 was used for subsequent targeting. The wild type allele in the Bptf/BptfΔexon2 from clone 7004 was then retargeted using the Bptf exon 2 targeting vector. Individual conditional homozygous clones were first screened for by PCR then confirmed by Southern blotting. Two BptfFloxedNeo/BptfΔexon2 clones were obtained and named H12 and B19. The piCre expression vector was electroporated into H12 and B19 and conversion from BptfFloxedNeo/BptfΔexon2 to BptfΔexon2/BptfΔexon2 was first screened for by PCR then confirmed by Southern blotting. A total of 9 and 5 homozygous knockout clones were obtained from the H12 and B19 parental lines respectively. The karyotype of the knockout lines were confirmed using Giemsa staining and lines with an anuploid karyotype were discarded. ES cells were maintained on mitomycin C treated primary mouse embryonic fibroblasts (MEFs) and ES Cell Growth Media (15% FCS (ES Cell grade, Invitrogen), DMEM, essential amino acids,10 µM mercapto-ethanol, 2 mM glutamine, 100 U/ml LIF (Chemicon International), penicillin and streptomycin) throughout. Lines were passaged off MEFs onto gelatinized plates for 3 passages prior to any molecular analysis.
Rescue lines were made by retargeting one allele of BptfΔexon2 in clones Bptf mutant clones P2-G2 and P2-B9. Individual Neo resistant and HAT resistant clones were selected on mitomycin-C treated MEF feeder layers in ES Cell Growth Media (15% FCS (ES Cell grade, Invitrogen), DMEM, essential amino acids,10 µM mercapto-ethanol, 2 mM glutamine, 100 U/ml LIF (Chemicon International), penicillin and streptomycin) throughout. Successful retargeting events were confirmed by Southern analysis as described below.
Clone 7010 was used to make chimera mice as described previously [66]. Removal of frt-Neo-frt and deletion of LoxP-exon2-LoxP was accomplished by crossing to mice expressing Tg-CMV-Flp and Tg-CMV-Cre to BptfFloxedNeo to create the BptfFloxed and BptfΔexon2 lines respectively. Individual recombinants were identified by Southern blotting and maintained by backcrossing to C57B6/CRL.
Mice used in this study were maintained in a Specific Pathogen Free environment at the National Institutes of Health at Bethesda (MD, USA). Mice were maintained on a 12 hr light/dark cycle given NIH-13 blend lab chow and hypo chlorinated water ad libtium throughout the duration of the study. All experiments and animal maintenance procedures were approved by the Animal Care and Use Committee of the National Cancer Institute under protocol LMCB001 and its modifications.
NCI-Frederick is accredited by AAALAC International and follows the Public Health Service Policy for the Care and Use of Laboratory Animals. Animal care was provided in accordance with the procedures outlined in the Guide for Care and Use of Laboratory Animals” (National Research Council; 1996; National Academy Press; Washington, D.C.).
BptfXG023, BptfFloxedNeo, BptfFloxed, and BptfΔexon2 embryos on a B6/129 mixed background have been cryopreserved at the Cryopreservation and Assisted Reproduction Lab, National Cancer Institute, Frederick (MD, USA) as BPTFXG023, BPTFFloxedNEO, BPTFFloxed and BPTFdel-exon2 respectively. These mouse lines are available to the research community by request.
Mice were routinely genotyped by Southern blotting restriction digested tail DNA. Tail DNA was prepared using standard procedures [67]. ∼10 µg of DNA was digested with the appropriate restriction enzyme and resolved by 0.5% agarose gel electrophoresis for standard electrophoresis or 1.0% agarose using FIGE. DNA was transferred to Hybond XL (Amersham) using alkaline capillary transfer crosslinked with UV (Stratalinker) and probed with random hexamer labeled probes at 65°C using Perfect Hyb Hybridization Buffer (Sigma). Probes used for blotting were PCR amplified from pCH110 (Genbank # UO2445) using primers JL510 and JL511 for Probe B, and from mouse genomic DNA using primers JL299 and JL300 for Probe A; JL60 and JL61 for probe exon2; JL235 and JL236 for probe 29–31; JL293 and JL294 for probe 25–26; JL295 and JL296 for probe 20.2–20.7 using Taq polymerase (Invitrogen) (see Table S3 for primer sequences). Blots were washed twice with 2.0× SSC, 0.1% SDS and twice with 0.5× SSC, 0.1% SDS for 10 min @ 65°C for each wash. A Phosphoimager was used to detect the hybridization signal.
Embryos were genotyped using a PCR based strategy. Primers used for genotyping the BptfXG023 line were as follows, Primer A (JL511), Primer B (JL627), and Primer C (JL631) (see Table S3 for primer sequences). PCR reactions were performed in 25 µl volumes with 20 mM Tris-HCl (pH 8.4), 50 mM KCl, 1.5 mM MgCl2, 0.2 mM each dNTP, 0.2 µM each primer, 1 Unit Taq Polymerase (Invitrogen). Reactions were heated for 3 min at 94°C, and then cycled for 35 cycles at 30 sec at 94°C, 30 sec at 65°C, and 45 sec at 72°C, followed by one cycle for 2 min at 72°C. PCR products were resolved on 1.5% agarose gels. The genotype of embryos was confirmed by the presence of a 250 bp Bptf and/or 500 bp BptfXG023 bands.
Two different PCR strategies were used for genotyping the BptfΔexon2 line. The primers used in the first strategy are as follows, Primer A (JL648), Primer B (JL649), and Primer C (JL652) (see Table S3 for primer sequences). PCR reactions were performed in 25 µl volumes with 20 mM Tris-HCL (pH 8.4), 50 mM KCl, 1.5 mM MgCl2, 0.2 mM each dNTP, 0.2 µM each primer, 1 Unit Taq Polymerase (Invitrogen). Reactions were heated for 3 min at 94°C, then cycled for 35 cycles at 30 sec at 94°C, 30 sec at 60°C, and 45 sec at 72°C, followed by one cycle for 2 min at 72°C. PCR products were resolved on 1.5% agarose gels. The genotype of embryos was confirmed by the presence of a 250 bp Bptf, 325 bp BptfFloxed and BptfFloxedNeo or 550 bp BptfΔexon2 band.
The primers used in the second method are as follows, Primer C (JL651), Primer E (JL655), and Primer F (JL662) (see Table S3 for primer sequences). PCR reactions were performed in 25 µl volumes with 20 mM Tris-HCL (pH 8.4), 50 mM KCl, 1.5 mM MgCl2, 0.2 mM each dNTP, 0.2 µM each primer, 1 Unit Taq Polymerase (Invitrogen). Reactions were heated for 3 min at 94°C, then cycled for 35 cycles at 30 sec at 94°C, 30 sec at 65°C, and 45 sec at 72°C, followed by one cycle for 2 min at 72°C. PCR products were resolved on 1.5% agarose gels. The genotype of embryos was confirmed by the presence of a 375 bp Bptf, 475 bp BptfFloxed or 650 bp BptfΔexon2 bands.
To confirm BptfXG023 and BptfΔexon2 expression and an out of frame Bptf mRNA in the BptfΔexon2 line we designed primers with homology to the 3′ end and exons 1 through 8 of the Bptf mRNA. Total RNA from wild type and BptfΔexon2 homozygous embryos was converted to cDNA using superscript II according to manufacturer's procedures. Bptf expression was estimated by first normalizing cDNA to equal concentration with GAPDH amplification. Following GAPDH normalization 3′Bptf was amplified. Both GAPDH and Bptf were amplified in the linear range as follows. Reactions were heated for 10 min at 94°C, then cycled for 22 cycles for GAPDH and 30 cycles for Bptf at 20 sec at 94°C, 20 sec at 60°C, and 30 sec at 72°C. PCR products were resolved by native PAGE. The exon 1–8 junction was amplified using primers P3 and JL15 in 25 µl volumes with 20 mM Tris-HCL (pH 8.4), 50 mM KCl, 1.5 mM MgCl2, 0.2 mM each dNTP, 0.2 µM each primer, 1 Unit Taq Polymerase (Invitrogen) (see Table S3 for primer sequences). Reactions were heated for 3 min at 94°C, then cycled for 35 cycles at 30 sec at 94°C, 30 sec at 65°C, and 2 min at 72°C, followed by one cycle for 5 min at 72°C. PCR products were resolved on 1.5% agarose gels. PCR fragments were cloned using TOPO TA cloning system (Invitrogen) and sequenced with Big Dye V3.1 using M13 forward and reverse primers.
The exact integration site for the BptfXG023 insertion allele was confirmed by sequencing a PCR product generated by amplifying genomic DNA from a heterozygous mouse tail clipping. The DNA was amplified using primers JL511 and JL515 (see Table S3 for primer sequences). Reactions were heated for 3 min at 94°C, then cycled for 35 cycles at 30 sec at 94°C, 30 sec at 65°C, and 60 sec at 72°C, followed by one cycle for 2 min at 72°C. A ∼1.8 Kb PCR product was resolved on 1.5% agarose and purified. The product was sequenced using Big Dye v3.1 using primer JL511 to obtain BptfXG023 junction sequence.
BptfXG023 disruption was confirmed by 5′ RACE according to established procedures from Bptf/BptfXG023 total testis RNA according to procedures established by the Bay Genomics Gene Trap Consortium (http://baygenomics.ucsf.edu/). Two separate 5′ RACE reactions were sequenced giving identical results.
Timed pregnancies used to collect embryos were preformed using standard procedures [67]. Embryonic day 0.5 was recorded the morning after an observable plug was found. Portions of later stage E18.5 to E10.5 embryos were genotyped by Southern blotting and early E9.5 to E3.5 embryos and outgrowths were genotyped by PCR.
Embryos used for β-galactosidase staining, in situ hybridizations and phenotypic analysis were dissected from their deciduas in PBS at room temperature prior to staining/hybridization. Embryos were stained for β-galactosidase activity or used for in situ hybridizations using established protocols [67] (T. Yamaguchi, personal communication). Probes used for in situ RNA hybridizations were labeled with DIG according to standard procedures [67]. Plasmids or PCR products used for probe synthesis were as follows T, Lhx1, GATA6, Lefty1, FGF4, FGF8, HesX1, Hhex, Cripto, BMP4, Otx2, Nodal, Cer, Foxa2, Erb2, Fgfr2, Mash1, JunB, VEGF and Flk1. In situ signals were detected using BM Purple AP Substrate according to manufactures instructions (Roche). Embryos selected for sectioning were post fixed in 4% PFA overnight at 4°C, dehydrated in alcohol and embedded in paraffin and sectioned after β-galactosidase staining. Sectioned embryos were obtained using standard techniques [67]. Embryos genotyped following β-galactosidase staining or in situ hybridization were prepared post staining by digesting the embryo proper overnight at 55°C in 10 to 20 µl PBS, 0.1% tween 20, 2 mg/ml PCR grade protenase K (Roche). Proteinase K was heat inactivated at 100°C for 15 min and 2 µl was subsequently used for PCR genotyping.
Deciduas used for sectioning were dissected from the uterus in PBS, washed once with PBS and immersed in 4% PFA buffered with PBS overnight at 4°C. Deciduas were then dehydrated in an ethanol series and imbedded in paraffin. Paraffin blocks were trimmed and sectioned and sections were stained with hematoxylin and eosin, TUNEL assay and phosphorylated H3 according to standard procedures [67].
Embryos used for RT PCR analysis were dissected in DMEM, 25 mM HEPES pH 7.4, 15% FCS, 2 mM glutamine, penicillin and streptomycin. Embryonic and extra embryonic tissue was removed from the ectoplacental cone and frozen in dry ice. The ectoplacental cone was grown ex vivo in DMEM, 15% FCS, glutamine, penicillin and streptomycin using standard culture conditions. After 4 days growth extra embryonic cells were removed from the outgrowth and subjected to PCR genotyping as described above and qPCR as described below.
Outgrowth assays were preformed on E3.5 blastocysts. E3.5 blastocysts were isolated in 15% FCS, M2, 10 µM mercapto-ethanol, penicillin and streptomycin using standard procedures. Blastocysts were then transferred to gelatinized plates containing 15% FCS, M16, 10 µM β-mercaptoethanol, 2 mM glutamine with penicillin and streptomycin. Outgrowths were cultured under standard conditions for 5 days and observed for defects. At the end of the experiment the outgrowths were removed and genotyped by PCR using conditions above.
Mouse tissues were extracted from 3 month old male mice sacrificed by CO2 asphixation and immediately frozen in liquid nitrogen. Tissues were then extracted using TriReagent (Sigma) according to manufacturers established procedures. RNA precipitate was extensively washed in 80% EtOH, re-suspended in DEPC treated water and stored at −80°C. Protein precipitate was extensively washed with 0.3 M guanadininum hydrochloride, 95% ethanol and re-suspended in 8M urea, 1% SDS and stored at −80°C. Protein concentration was determined using the DC Protein Assay (BioRad) according to manufacturer's procedures.
mRNA from embryos successfully genotyped as WT or Bptf−/− from ectoplacental growth outgrowths was isolated using Quick Prep Micro mRNA Purification Kit (Amersham).
RNA samples were denatured, resolved by formaldehyde agarose electrophoresis and blotted to Gene Screen (Dupont) by high salt capillary transfer according to standard procedures. Blots were probed with random hexamer labeled DNA probes using Prefect Hyb hybridization solution (Sigma) at 65°C according to the manufactures procedures. Probes used were the Probe B and Exon2 probe described above. Blots were washed with four times with 1× SSC for 10 minutes at 65°C. Phosphoimager was used to detect the hybridization signal.
Protein samples were diluted in SDS sample buffer and resolved by PAGE. 50 µg total protein was transferred to PVDF (Biorad) using 10 mM CAPS, pH 10.5, 15% methanol, 3.5 mM DTT transfer solution for 17 hours at 20 mA and 20 V limits. Membranes were blotted with affinity purified anti-Bptf at 1∶10000 dilution [14], M2 anti Flag monoclonal antibody at 1∶5000 (Sigma), anti Smad2 and Smad2-phos (Cell Signaling Technologies) 1∶1000, anti SNF2H/L (Abcam) 1∶5000, anti-CBP and p300 1∶100 (Abcam) overnight at 4°C then anti rabbit or mouse HRP (Amersham) at 1∶20000 dilution for 2 hrs at RT in PBS with 5% NFDM and 0.1% tween 20 through out.
Histone Western blotting was performed essentially as above. Protein samples were extracted from ES cells grown on gelatinized plates in ES cell growth media containing 100 U/ml LIF using TriReagent according to manufacturer's procedures (Sigma). 50 µg of total protein was resolved by 15% SDS PAGE and transferred to PVDF (Biorad) using 12.5 mM Tris, 96 mM glycine pH 8.3, 20% methanol transfer solution for 1 hour at 200 mA and 25 V limits. Blots were probed with rabbit polyclonal antibodies to H4 ac-K5, H4 ac-K8, H4 ac-K12 (Serotech), H4 3me-K20, H4 acK16, H3 3me-K9, H3 3me-K27 (Upstate), g-H2AX, H3 3me-K4, H3 ac-K14, H3 ac-K18 (abcam) at 1∶2000 dilutions overnight at 4°C then anti rabbit HRP (Amersham) at 1∶4000 dilution for 1 hrs at RT in PBS with 5% NFDM and 0.1% tween 20 through out. Detection was preformed using Pico signal ECL according to manufacturer's procedures (Pierce).
ES cells were maintained on mitomycin C treated MEFs in ESC Growth Media (15% FCS (ES Cell grade, Invitrogen), DMEM, essential amino acids, 10 µM mercapto-ethanol, 2 mM glutamine, 100 U/ml LIF (Chemicon International), penicillin and streptomycin). Culture techniques used were described previously [66]. Lines were passaged 3 times onto gelatinized plates to remove MEFs prior to experiments.
Growth rates of ES cell lines were obtained on gelatinized plates in the absence of feeder layers. Cells were plated @ 1.0×105 in 6 well plates. Cell number was determined every day for 4 consecutive days with changes in ESC Growth media occurring every day.
Activin-A stimulation of ES cells was performed as follows. ES cell lines were started on gelatinized plates in ESC Growth Media. Cultures were then grown for 2 days in ESC Growth Media without LIF. On the third day, cells were cultured in low serum ESC Growth Media without LIF (same as ESC Growth Media except 0.1% FCS is used instead of 15% FCS) in the presence or absence of 30 ng/ml activin-A (R & D Systems) overnight. Cells were processed using TriReagent according to manufactures standard protocol.
p300, CBP siRNA knockdowns were preformed as follows. ES cells were grown in ES Cell Growth Medium with 100 U/ml LIF on gelatinized plates. Cells were collected using trypsin and transfected using an Amaxa Nucleofector device using solution ES Cell and program A-30 according to manufacturer's procedures. 2.0×106 ES cells were transfected with 400 nmoles siRNA duplex (Darmacon) or a GFP-Max nucleofection control. Post nucleofection 1.0×106 cells were plated to each well of a 6 well gelatinized tissue culture plate. Cells were stimulated with or without Activin-A in low serum ES Cell growth media as described above. Cells at specified time points were lysed using TriReagent reagent (Sigma) and RNA and protein were purified according to manufacturer's procedures.
MCF10CA1h cells were grown in a 1∶1 mixture of DMEM and Ham's F12 medium (Gibco) supplemented with 5% horse serum. Cells were collected using trypsin and transfected using an Amaxa Nucleofector device using solution V and program T-27 according to manufacturer's procedures. 2.0×106 cells were transfected with 400 nmoles siRNA duplex (Darmacon) or a GFP-Max nucleofection control. Post nucleofection 1.0×106 cells were plated to each well of a 6 well tissue culture plate. Cells were grown for 2 days in growth media and then serum deprived in 1% FBS, DMEM, nonessential amino acids, 10 µM β-mercaptoethanol overnight. Following serum shock cells were incubated with or without 5 ng/ml TGF-β1 (R&D Systems) for 1 hour. Cells were lysed using TriReagent reagent (Sigma) and RNA and protein were purified according to manufacturer's procedures.
P19 cells were obtained from the ATCC (ATCC Number CRL-1825). Cells were grown in 10% FBS, DMEM, 2 mM glutamine, penicillin and streptomycin. Cells were transfected using an Amaxa Nucleofector device using solution V and program C-20 according to manufacturer's procedures. 2.0×106 P19 cells were transfected with 400 nmoles siRNA duplex (Darmacon) or a GFP-Max nucleofection control. Post nucleofection 3.0×105 cells were plated to 12 well plates. Cells were allowed to grow for two days then were serum starved for 18 hours in 0.1% FBS, DMEM, 2 mM glutamine, penicillin and streptomycin. Cells were induced with 5 ng/ml TGF-β1 (R&D Systems) in low serum growth medium overnight. Cells were lysed using TriReagent reagent (Sigma) and RNA and protein were purified according to manufacturer's procedures.
Luciferase assays were performed in 12 well format by transfecting 0.6 µg pGL3ti-(SBE)4 or pAR3-Luc, 0.012 µg pCH110 with or without 0.6 µg TβRI or 0.6 µg Smad 2-HA and 0.006 µg Smad4-HA, 40.0 pmoles siRNA duplex per well using Lipofectamine 2000 (Invitrogen). In experiments using TGF-β1 (R&D Systems) or Activin-A (R&D Systems) cells were allowed to grow for 48 hours in 10% FBS, DMEM, 2 mM glutamine prior to adding TGF-β1 at 5 ng/ml or Activin-A at 30 ng/ml. Cells were induced for 24 hours then lysed with 250 µl siGlow Lysis Buffer (Promega) according to manufacturer's procedures. Transfections utilizing the constitutively active receptors were incubated for 48 hours prior to lysis as above. 50 µl or 100 µl of lysate was assayed with 100 µl Bright-Glow Luciferase Assay Reagent (Promega). Transfection efficiency was normalized to β-galactosidase levels by assaying 10 µl lysate to 100 µl Beta-Glo Assay Reagent (Promega). Relative Luciferase units were obtained by dividing the luciferase activity levels by the β-galactosidase levels. Experiments were repeated at least twice and yielded essentially the same results.
We used P19 cells to test specificity of TGFβ induction of Bptf protein levels. P19 cells were grown in 10% FBS, DMEM, 2 mM glutamine, penicillin and streptomycin. Ligands were added at the following concentrations; TGFβ1 (R&D Systems) at 5 ng/ml, FGF4 (R&D Systems) and heparin at 25 ng/ml and 1 ug/ml respectively, BMP4 (R&D Systems) at 10 ng/ml and Wnt3a or L cell control conditioned medium at 1∶1 with 10% FBS, DMEM, 2 mM glutamine, penicillin and streptomycin. Cells were incubated overnight and harvested for protein with TriReagent reagent (Sigma) according to manufacturers procedures. Bptf Western blotting was performed as described above.
Embryoid bodies were cultured as follows. ES cells were treated with trypsin briefly to retain cells in medium sized clumps. Cells were then centrifuged at low speed and resuspended in ESC Growth Medium without LIF and cultured in bacteriological grade petri dishes. ESC Growth Medium without LIF was changed every day for 9 days with minimal disturbance to the embryoid bodies. Samples were taken at specified time points using TriReagent according to manufactures standard protocol or fixed in 10% NBF for histology.
Knockout of Bptf in MEFs was accomplished by infecting litter mate wild type and conditional Bptf (-/Floxed genotype) MEF lines with an adenovirus expressing CMV-Cre. Infection was performed in 6 well plates at sub-confluence (2×105 cells) with 1 ml growth media containing 45 µl Adenovirus @ 1×1010 PFU/ml. Cells were grown for two days in the presence of virus. Following infection cells were divided into 5 wells of a 6 well plate. Cell numbers were recorded every 24 hours with changes in media occurring every 48 hours. A sample of cells was removed at day 4 for Western analysis of Bptf protein levels. Experiment was repeated with two wild type and conditional knockout MEF lines from littermate embryos.
ES cells were maintained on mitomycin C treated MEFs in ESC Growth Media (15% FCS (ES Cell grade, Invitrogen), DMEM, essential amino acids, 10 µM mercapto-ethanol, 2 mM glutamine, 100 U/ml LIF (Chemicon International), penicillin and streptomycin). Culture techniques used were described previously [66]. Lines were passaged 3 times onto gelatinized plates to remove MEFs prior to experiments.
Activin-A stimulation of ES cells was performed as follows. ES cell lines were started on gelatinized plates in ESC Growth Media. Cultures were then grown for 2 days in ESC Growth Media without LIF. On the third day, cells were cultured in low serum ESC Growth Media without LIF (same as ESC Growth Media except 0.1% FCS is used instead of 15% FCS) in the presence or absence of 30 ng/ml activin-A (R & D Systems) overnight. Following activin-A induction cells were washed with PBS and fixed in 2 mM EGS (Pierce) in 25% DMSO/75% PBS for 30 min followed by 1% PFA in PBS for 30 min. For histone ChIP, cells were fixed in 1% PFA in PBS for 15 min. Following fixation cells were washed 3X in PBS and removed from the tissue culture dish with a cell scraper. Cell pellets were then frozen at −80°C. The following day the pellets were thawed and processed using the ChIP procedure published by Upstate biologicals. Antibodies used for pulldown were ChIP grade SNF2H/L (Abcam), H3 3me-K4, H3 3me-K27 (Upstate) and pan H3 (Abcam).
Quantitation of pulldown was performed by real time PCR using 2× DyNAmo syprogreen qPCR kit (New England Biolabs) according to manufacturers procedures. Briefly, reactions were composed of 5 µl 1.2 µM forward and reverse primers, 5 µl diluted cDNA template and 10 µl 2× qPCR mix (see Table S3 for primer sequences). Reactions were heated for 10 min at 94°C, then cycled for 40 cycles at 20 sec at 94°C, 20 sec at 60°C, 30 sec at 72°C. After each cycle the sample was heated to 78°C for 10 sec prior to reading sample fluorescence. Pulldowns were quantified as a percentage of input using a dilution series as a standard curve. Histone modification pulldowns are expressed as enrichment relative to histone H3 occupancy under +activin-A conditions. SNF2H/L pulldowns were first normalized to signal at Gapdh and are expressed as the enrichment during +activin-A relative to −activin-A conditions.
RT reactions were performed on RNA extracted from ES cells, embryoid bodies or P19 cells using TriReagent according to manufacturer's procedures. 5 µg of total RNA was reverse transcribed with Superscript II using oligo dT priming according to manufactures procedures (Invitrogen). cDNA reactions were diluted 5–10 fold and used in a template for PCR.
Real time PCR was performed using 2× DyNAmo syprogreen qPCR kit (New England Biolabs) according to manufacturers procedures. Briefly, reactions were composed of 5 µl 1.2 µM forward and reverse primers, 5 µl diluted cDNA template and 10 µl 2× qPCR mix (see Table S3 for primer sequences). Reactions were heated for 10 min at 94°C, then cycled for 35 cycles at 20 sec at 94°C, 20 sec at 60°C or 52°C, 30 sec at 72°C. After each cycle the sample was heated to 78°C for 10 sec prior to reading sample fluorescence. Reactions were done in triplicate. ΔΔCt method was used to quantify the relative levels of expression to the Gapdh or β-actin house keeping genes. Expression levels were then normalized to un-induced cells control cells.
QPCR was performed on E7.5 embryo cDNA as follows. Briefly, reactions were composed of 5 µl 1.2 µM forward and reverse primer pair, 5 µl diluted cDNA template and 10 µl 2× qPCR mix (see Table S3 for primer sequences). Reactions were heated for 10 min at 94°C, then cycled at 20 sec at 94°C, 20 sec at 60°C, and 30 sec at 72°C. The number of cycles to maintain linearity was determined using the real time analysis software. PCR reactions within the linear range were resolved using native PAGE.
For microarray experiments we used one wildtype and two independent Bptf mutant cell lines in 3 culture conditions (LIF+, LIF−, and RA+). Experiments were carried out with 3 biological replications. At the day 3, Triazole (1 ml/well; Invitrogen, USA) was added to the well and total RNAs were extracted using Phase lock gel columns (Eppendorf/Brinkman) according to the manufacturer's protocol. Total RNAs were precipitated with isopropanol, washed with 70% ethanol, and dissolved in DEPC-treated H2O. 2.5 g of total RNA samples were labeled with Cy3-CTP using a Low RNA Input Fluorescent Linear Amplification Kit (Agilent, USA). A reference target (Cy5-CTP-labeled) was prepared from the Universal Mouse Reference RNA (Stratagene, USA). Labeled targets were purified using an RNeasy Mini Kit (Qiagen, USA) according to the Agilent's protocol, quantified by a NanoDrop scanning spectrophotometer (NanoDrop Technologies, USA), and hybridized to the NIA Mouse 44K Microarray v2.2 (whole genome 60-mer oligo; manufactured by Agilent Technologies, #014117) [68]. Transcript copy number estimation using a mouse whole-genome oligonucleotide microarray according to the Agilent protocol (G4140-90030; Agilent 60-mer oligo microarray processing protocol - SSC Wash, v1.0). All hybridizations were carried out in the two color protocol by combining one Cy3-CTP-labeled experimental target and Cy5-CTP-labeled reference target. Microarrays were scanned on an Agilent DNA Microarray Scanner, using standard settings, including automatic PMT adjustment.
Differential gene expressions in various cell lines in the standard culture condition were analyzed using the NIA Array Analysis software (http://lgsun.grc.nia.nih.gov/ANOVA/) which implements ANOVA statistics with two additional methods to reduce the number of false positives: (1) small error variances were replaced with the average error variance estimated from 500 genes with similar signal intensity, and (2) false discovery rates (FDR<0.05) were used to select genes with differential expression, instead of p-values [69]. The FDR method accounts for the effect of multiple hypotheses testing.
Gene Ontology analysis and clustering was accomplished using DAVID (http://david.abcc.ncifcrf.gov/) and the NIA Mouse Gene Index (http://lgsun.grc.nia.nih.gov/geneindex/mm8) using default settings [70].
In vitro GST pulldown assays were performed in 50 µl volumes in binding buffer (25 mM Hepes pH 7.4, 100 mM NaCl, 0.5 mM MgCl2, 0.01% NP40), ∼10 µg GST protein fusion bound to a glutathione Sepharose support, 100 ng human NURF complex. Proteins were allowed to bind for 1 hour at 4°C with occasional mixing. The beads were washed with 500 µl binding buffer three times at 4°C and re-suspended in a final volume of 30 µl.
NURF and CBP pulldowns from ES Cell extracts were preformed essentially as described above from 500 mM KCl nuclei extractions made as previously described [71]. The extract was diluted 1∶3 with 25 mM Hepes pH 7.4, 0.5 mM MgCl2, 0.01% NP40 with protease inhibitors (Roche) prior to binding to resin bound GST-Smads. The beads were washed with 500 µl binding buffer three times at 4°C and re-suspended in a final volume of 30 µl. Proteins were eluted from the beads by adding 5 µl 6× SDS sample buffer and incubating at 37°C for 30 min. Proteins were resolved on 4% or 8% polyacrylamide gels and transferred to PVDF (Biorad) and prepared for Western blotting as described above.
|
10.1371/journal.pbio.0050192 | Myc Dynamically and Preferentially Relocates to a Transcription Factory Occupied by Igh | Transcription in mammalian nuclei is highly compartmentalized in RNA polymerase II-enriched nuclear foci known as transcription factories. Genes in cis and trans can share the same factory, suggesting that genes migrate to preassembled transcription sites. We used fluorescent in situ hybridization to investigate the dynamics of gene association with transcription factories during immediate early (IE) gene induction in mouse B lymphocytes. Here, we show that induction involves rapid gene relocation to transcription factories. Importantly, we find that the Myc proto-oncogene on Chromosome 15 is preferentially recruited to the same transcription factory as the highly transcribed Igh gene located on Chromosome 12. Myc and Igh are the most frequent translocation partners in plasmacytoma and Burkitt lymphoma. Our results show that transcriptional activation of IE genes involves rapid relocation to preassembled transcription factories. Furthermore, the data imply a direct link between the nonrandom interchromosomal organization of transcribed genes at transcription factories and the incidence of specific chromosomal translocations.
| Many different types of cancer result from gene translocations. Specifically, two different chromosomes can be joined that fuse growth control genes with powerful regulatory elements, leading to unrestricted control of cell growth. Translocation partner genes must physically encounter each other in the nucleus to undergo a translocation; how they find each other in the crowded nucleus is unknown. We showed previously that gene transcription occurs at a few hundred discrete nuclear sites called transcription factories. In the current study we investigated the effects of activation of the Myc proto-oncogene and examined its location with respect to transcription factories and its common translocation partner, the immunoglobulin heavy chain (Igh) gene. We found that switching on the Myc gene leads to its rapid relocation to a transcription factory. Surprisingly, we found that the activated Myc frequently chooses the same transcription factory as the highly transcribing Igh gene. This close juxtaposition of translocation partner genes at a shared transcription factory may provide the opportunity for a chromosomal translocation, and thus may be the first step in the genesis of several types of cancers.
| Interphase chromosomes are organized in tissue-specific arrangements in nuclei, suggesting that chromosomal position and juxtaposition play a role in gene expression [1–5]. Nonrandom chromosome positioning has also been implicated in the frequency of specific chromosomal translocations. For example, Chromosomes 12 and 15, which contain the frequent B cell translocation partners immunoglobulin heavy chain (Igh) and the proto-oncogene Myc, are preferred neighbors in mouse splenic lymphocytes [3]. Similarly, in human lymphoid cells MYC and IGH are found in the same vicinity in about one-third of nuclei [6]. Any process that brings these genes together would obviously be expected to increase the risk of a translocation between them; however, almost nothing is known about the forces that organize chromosomes in the nucleus.
Nascent transcription occurs at RNA polymerase II (RNAPII)-rich nuclear foci known as transcription factories [7–11]. These sites are highly enriched in the hyperphosphoryated forms of RNAPII involved in transcription initiation and elongation [7,11]. Previous findings suggest that active mammalian genes are transcribed in bursts of activity punctuated by long periods of relative inactivity [9,12–14]. This concept is supported by recent live-cell studies showing that gene expression [15,16], and in particular gene transcription [17], occur in discrete pulses. We and others [9,10] have observed a virtually absolute correspondence between transcriptional activity at individual gene alleles and their positioning within transcription factories, whereas identical inactive alleles, often in the same cell, are clearly positioned away from factories. Collectively, these data could be interpreted to imply that the engagement of genes at factories is dynamic; however, they could equally be construed to indicate that a transcription factory nucleates around an individual gene during a transcriptional burst. Arguing against the latter interpretation is the finding that multiple genes in cis and trans can frequently share the same factory, which strongly suggests that genes migrate to preassembled transcription sites for transcription [9].
In this study, we investigated the positioning of immediate early (IE) genes relative to transcription factories and other B cell expressed genes during IE induction. We found that before activation the majority of IE alleles are not associated with transcription factories, whereas upon induction, IE genes rapidly relocate to preformed transcription factories. Remarkably, we observed preferential recruitment of the proto-oncogene Myc to the same transcription factory that is occupied by its frequent translocation partner, Igh. Our results suggest that this frequent and preferential juxtaposition may provide the opportunity for a chromosomal translocation, and may in part dictate the incidence with which specific chromosomal translocations occur.
Resting B cells can be stimulated through the B cell receptor signaling pathway to rapidly increase transcription and mRNA expression of the IE genes Fos and Myc [18]. We used RNA fluorescent in situ hybridization (FISH) with gene-specific intron probes to investigate the transcriptional activity of several genes during IE gene induction in mouse B lymphocytes (Figure 1). We found that transcription frequencies vary for several genes in B cells, similar to our previous observations in erythroid cells [9]. Transcription signals for the B cell-specific gene Igh are present at approximately 90% of alleles both before and after induction (Figure 1A). In the vast majority of cells both alleles are actively transcribed (>80% of cells), whereas a smaller percentage of cells have only single signals, consistent with our previous findings [19]. The immunoglobulin light chain genes Igk and Igl also have transcription signals at approximately 90% of alleles, with approximately 80% of cells having two transcription signals (Figure 1B and 1C). These data show constitutive transcriptional activity at nearly all alleles that is unchanged upon IE induction. In contrast, transcription of the IE proto-oncogenes Fos and Myc is significantly lower in unstimulated cells, with 20% and 26% of alleles, respectively, displaying transcription signals (Figure 1D and 1E). Most cells (~60%) have two silent alleles, some cells have one active allele (~30%), and a minority of cells have two active alleles (<10%). Upon induction, the percentage of loci with transcription signals for Fos and Myc rises dramatically within 5 min to 53% and 75%, respectively. This rise is the result of a dramatic increase in the percentage of cells with two active alleles and, to a lesser extent, an increase in cells with one active allele. The fold increases in the percentage of active alleles are in precise agreement with run-on transcription studies in mouse B cells that demonstrated a 2- to 3-fold induction of Fos and Myc transcription in stimulated cells [18]. These results suggested the possibility that increased IE expression could be accounted for by transcriptional recruitment of additional IE alleles rather than an increase in the “basal” rate of transcription at all IE alleles.
We questioned whether the increase in nascent transcription levels seen in run-ons could be accounted for solely by the transcriptional recruitment of additional IE alleles. We used a sensitive reverse transcription PCR (RT-PCR) technique capable of quantitating the average absolute number of primary transcripts per cell [20]. In this method serial dilutions of a known amount of a spiked competitor RNA that contains a small internal deletion is compared to the amount of endogenous primary transcript in total RNA preps from a known number of cells. We adapted this method by using PCR primers that flank a 5′ splice donor site. Cleavage of the primary transcript at the 5′ donor site occurs soon after the RNA polymerase has passed the 3′ splice acceptor site, at the end of the intron sequence [21]. Quantitative detection of RT-PCR products from this part of the primary transcript provides an extremely sensitive measure of the number of transcripts being synthesized over that intron. An estimate of the average number of primary transcripts per gene can be calculated by extrapolating to the full size of the transcription unit.
We found that the average number of Igh primary transcripts does not change during B cell induction (Figure 1F; Table 1), consistent with our FISH data, which indicate that Igh transcription is unaffected by stimulation. We detected approximately 4.3 and 4.5 unspliced primary transcripts on the Igh intron per cell in unstimulated and stimulated cells, respectively. As this intron is approximately half the Igh transcription unit length, we calculated that there are an average of 8.5 and 9.0 primary transcripts being produced from the two Igh transcription units per cell (4.3 and 4.5 per Igh allele). Since RNA FISH shows that 90% of Igh alleles are transcriptionally active, we estimate that there are approximately five transcripts being produced on each active allele.
The picture for the IE Fos gene is very different. In unstimulated cells we found that the extrapolated, average absolute number of Fos primary transcripts per cell is 1.73 copies. This is less than one primary Fos transcript per allele, indicating that not all Fos alleles are transcriptionally active in unstimulated cells. If we calculate the average number of Fos transcripts per active allele based on our RNA FISH data in which 20% of Fos alleles showed a transcription signal, we arrive at an average number of 4.3 Fos primary transcripts per active allele. In stimulated cells Fos primary transcript intron copies per cell increase approximately 2.7-fold, consistent with the 2.65-fold increase in the percentage of actively transcribed Fos alleles determined by RNA FISH. Comparison of the number of Fos primary transcripts per active allele indicates that the number of Fos transcripts per active allele does not change upon induction (4.3 versus 4.4 copies). These data show that our RNA FISH technique is very sensitive and is capable of measuring very small numbers of primary transcripts at a transcription site. In addition, these results show that when we do not see an RNA FISH signal over a particular allele, it is truly “off” and has no primary transcripts associated with it. Collectively, these results show that Fos IE gene induction occurs via transcriptional activation of additional, previously inactive alleles, rather than by simply increasing the “basal” rate of transcription of all alleles.
Our previous studies in erythroid cells suggested that gene association with transcription factories is dynamic, with genes moving to preformed factories in order to transcribe [9]. Inducible gene expression in B cells permitted us to examine the dynamics of transcription in relation to transcription factories. We speculated that the activation of previously quiescent IE alleles upon B cell induction may involve repositioning of alleles to transcription factories. However, the Myc gene has a well-characterized attenuation site that is thought to block passage of RNAP II, resulting in a stalled polymerase [22]. This observation leaves open the possibility that silent Myc alleles may be pre-positioned in factories awaiting removal of a transcriptional block.
We first examined the positions of actively transcribed genes relative to transcription factories, using RNA immuno-FISH (Figure 2). We found that 92% of Myc RNA FISH signals are associated with strong RNAP II foci (Figure 2A). Similarly, 90% of transcriptionally active Igh alleles are associated with strongly staining RNAP II foci (Figure 2B), consistent with previous observations of erythroid-expressed genes [9,10]. Others have shown that the remaining 10% of RNA FISH signals localize to weakly staining RNAPII foci [10], indicating that essentially all transcriptionally active alleles associate with transcription factories. This observation agrees with previous studies that showed a good correspondence between pulse-labeled nascent RNA and RNAPII foci [7,11]. We conclude that all Myc and Igh transcription occurs at transcription factories.
Next, we investigated the position of nontranscribing alleles by DNA immuno-FISH, which detects DNA of both active and inactive alleles and RNAP II proteins. We found that approximately 30% of Myc DNA FISH signals overlapped with RNAP II foci in unstimulated cells, while 70% were not associated with RNAP II foci. These results are consistent with the percentage of transcriptionally active alleles detected by RNA FISH (Figure 3A and 3C), and show that the inactive Myc alleles are not associated with transcription factories, but are instead positioned away from these sites. Three-dimensional DNA FISH measurements between the inactive Myc alleles and the nearest transcription factory show that on average silent Myc alleles are 500 nm from the nearest factory (Figure 3D).
After 5 min of stimulation the percentage of Myc loci associated with transcription factories increased to 65%, in agreement with the increased percentage of actively transcribing Myc alleles determined by RNA FISH (Figure 3B and 3C). These results show that Myc induction involves an increase in the percentage of Myc alleles associated with transcription factories. The increase of gene association with factories could be achieved in two ways. Myc transcriptional induction could involve the nucleation of transcription factories on newly activated Myc alleles. Alternatively, transcriptional induction could involve the rapid relocation of silent Myc alleles to preassembled transcription factories.
The synchronous induction of IE gene alleles described above allowed us to directly test these alternate scenarios. The Igh locus on mouse Chromosome 12 is positioned 28 Mb telomeric to the Fos locus. Approximately 90% of Igh alleles exhibit RNA FISH signals in B cells (Figure 1A) [19], and are associated with transcription factories, indicating that the Igh locus undergoes nearly constant transcription, similar to the highly expressed Hbb locus in erythroid cells [9,23]. We therefore used Igh RNA FISH signals as factory reference points and scored the percentage of Igh transcription signals that have a colocalizing (overlapping) Fos signal, before and after induction using double-label RNA FISH. In unstimulated cells approximately 7% of Igh signals had a colocalizing Fos signal (Figure 4A and 4C). After induction, colocalization nearly tripled, with 20% of Igh signals having a colocalizing Fos signal. These results suggest that a significant proportion of the newly activated Fos alleles move to a factory that is already occupied by an Igh allele rather than forming their own factory.
We previously showed a low but significant level of interchromosomal associations between the highly transcribed Hbb and Hba genes in erythroid cells [9]. We hypothesized that the preferred neighbor arrangement of Chromosomes 12 and 15 in B cells [3] might allow Myc and Igh to co-associate with the same transcription factory in trans. Using double-label RNA FISH as above, we found that approximately 6% of Igh signals have a colocalizing Myc signal in unstimulated cells (Table 2). Comparing the percentage of active Myc alleles that colocalized with an Igh signal to those that did not, we found that a remarkable 25% of the transcribing Myc alleles colocalized with Igh in trans prior to induction. Upon induction, as we observed a 2.9-fold increase in the percentage of transcribing Myc alleles, we found a 2.5-fold increase in the percentage of Igh alleles with a colocalizing Myc signal (Figure 4B and 4C; Table 2). Again, by comparing colocalizing versus noncolocalizing Myc signals we found that approximately one-fourth of the active Myc alleles (22–24%; Table 2) were associated with Igh alleles upon induction. Thus, one-fourth of the newly activated Myc alleles, which were previously located away from transcription factories, had moved to a factory occupied by Igh. We confirmed that colocalizing Myc and Igh transcription signals co-associated with a shared transcription factory using triple-label RNA immuno-FISH to detect transcriptionally active Myc and Igh alleles and RNAP II foci (Figure 4D). We found that all colocalizing Myc and Igh signals overlapped with the same transcription factory.
In order to put this extraordinarily high frequency of interchromosomal Myc-Igh colocalization into perspective we compared the colocalization frequencies between Igh and five other B cell-expressed genes. One gene, Eif3s6, is located approximately 20 Mb from Myc on Chromosome 15, and four other genes, Igk, Igl, Uros, and Actb, are on Chromosomes 6, 16, 7, and 5, respectively. Since Chromosomes 12 and 15 are preferred neighbors in B cells [3], we considered the analysis of colocalization between Eif3s6 and Igh to be of particular interest. If the high level of interchromosomal colocalization between Myc and Igh were simply due to the fact that the genes are on neighboring chromosomes, then we might expect Eif3s6 and Igh to colocalize at similar frequencies when transcribed. We found that Eif3s6 and Igh colocalize, but at significantly lower levels than Myc-Igh. Only 11% of Eif3s6 signals colocalized with Igh, compared to approximately 25% for Myc. For the other genes we found that 9% of Uros, 8% of Igk, 6% of Igl, and 2% of Actb transcribing alleles colocalized with Igh (Figure 5). These considerably lower frequencies of co-association with Igh clearly demonstrate that trans co-association frequencies between different gene pairs can vary greatly. For example Igh-Myc trans colocalization is over 10-fold higher than Igh-Actb, indicating that the Myc and Igh trans colocalization frequency is statistically highly significant. However, the Myc-Igh co-association is also highly preferential, as indicated in the comparison between Myc and Eif3s6. Myc co-associates with Igh at a greater than 2-fold higher frequency than Eif3s6. This result demonstrates that not all genes on neighboring chromosomes co-associate at equal frequencies and shows that Myc and Igh preferentially co-associate in trans.
Our results suggest that 5 min after induction, many previously inactive Myc alleles are moving to preformed factories that contain transcriptionally active Igh alleles. If this interpretation is correct we would expect to see a net movement of Myc alleles toward Igh alleles upon stimulation of IE gene expression. To directly test this hypothesis we carried out 3D DNA FISH, measuring the separation distances between Myc and Igh alleles in unstimulated and stimulated cells. We found a statistically significant shift in the distribution of measurements upon B cell stimulation, changing from a mean separation distance of 2.16 μm to 1.83 μm (p = 0.005) (Figure 6A). This shows that across the population of cells Myc and Igh alleles are significantly closer together 5 min after induction. In contrast, we found no significant change in the distributions of measurements between Igh and two other genes in trans, Actb and Uros (Figure 6B and 6C). These results show there is no net movement between Actb-Igh and Uros-Igh upon B cell activation, indicating that these genes are not significantly changing their location relative to one another. However, there is net movement of Myc alleles toward Igh upon induction. We did not detect any net movement between Igh and Fos upon induction (Figure 6D), most likely because the range of separation between these physically linked genes is too small to detect subtle changes in relative positioning via light microscopy [9]. We conclude that increased Myc expression during IE gene induction involves the rapid relocation of Myc alleles to preassembled transcription factories, with many alleles migrating to a factory containing the Igh gene.
We also assessed separation distances between Myc and Igh alleles in two other tissue types, adult kidney and fetal liver erythroid cells. We detected much greater separation distances between Myc and Igh in these tissues compared to unstimulated B cells (Figure 6E). Myc and Igh were separated by an average of 2.78 and 3.20 μm in kidney and erythroid cells respectively, compared to an average of 2.16 μm in unstimulated B cells. These results are consistent with previously reported observations of tissue-specific positioning of genes [6,24] and chromosomes [3,25] and show that Myc and Igh are already in the same “nuclear neighborhood” in unstimulated B cells, which most likely facilitates their increased proximity and colocalization upon stimulation.
To corroborate the FISH results, we used the capturing chromosome conformation (3C) assay [9,26,27], which measures ligation frequency between in vivo formaldehyde cross-linked chromatin fragments. Ligation products were detected with four different primer pairs within the Igh and Myc loci in stimulated B cells (Figure S1). The primer pair that produced the most robust product (primer pair d/g) was used to detect ligation products in unstimulated and stimulated B cells. Over multiple experiments, ligation products were always detected in stimulated B cells, while in unstimulated cells the products were usually, but not always detected. In contrast, Igh/Myc ligation products were never detected in brain or kidney cells, indicating that juxtaposition of Myc and Igh is restricted to tissues in which both genes are expressed.
Importantly, Myc and Igh are the two most common translocation partners in Burkitt lymphoma and mouse plasmacytoma. Of these cancers 80% harbor Myc-Igh translocations, while the remaining cases contain Myc-Igk (15%) or Myc-Igl (5%) translocations [28,29]. To establish whether there is a relationship between the frequency of these translocations in plasmacytomas and the co-association frequencies of transcriptionally active alleles in normal B cells, we measured the extent to which the transcriptionally active Myc colocalized with Igk and Igl alleles in 10-min stimulated cells by RNA FISH. We found that 11% of transcribing Myc alleles colocalized with Igk and 7% colocalized with Igl, compared to 22% with Igh (Figure 7). Thus the frequencies of co-association between Myc and the immunoglobulin loci in transcription factories are in line with the appearance of their respective translocation frequencies in mouse plasmacytomas.
Our results show that IE gene induction involves the rapid nuclear relocation of previously inactive genes to preassembled transcription factories. This dynamic transcriptional organization is nonrandom and leads to the preferential juxtaposition of the Myc and Igh genes at transcription factories. Transcriptional colocalization may provide an opportunity and therefore an increased risk of illegitimate recombination resulting in a chromosomal translocation [30,31]. We cannot discount the possibility that differences in oncogenic potential result in selective outgrowth of one type of translocation versus another. However, it is striking that the co-association frequencies echo the appearance of specific translocations in plasmacytomas, suggesting that the juxtaposition frequency of specific genes in a transcription factory has a direct effect on their translocation frequency.
Upon B cell induction, signaling pathways converge upon the IE genes, a process that causes their relocation to transcription factories. Others have shown recently that upon activation, genes can undergo directed, actin and myosin-dependent relocalization, moving between 1 and 5 μm [32]. We cannot discount the possibility that similar forces may be involved in the relocation of genes to factories. However, we found that inactive Myc alleles are positioned on average 500 nm from the nearest factory, a distance that could conceivably be covered by random chromatin movements [33–35].
A key question concerns the basis of the preferred co-associations of specific genes in a common factory. The relative positions of genes in cis or on preferred neighbor chromosomes would be expected to affect the frequency of co-association in a factory [9] as it does recombination frequency [36]. On the other hand, it is possible that tissue-specific chromosomal positioning is driven by the net effect of thousands of preferential interchromosomal interactions between active (and inactive [37]) genes that serve as dynamic anchor points that facilitate chromosome positioning [38]. Preferential co-associations in factories may be the result of 3D spatial clustering of genes with related functions or genes coordinately regulated by common factors [9,37,39]. Genome-wide examination of the subsets of genes that preferentially co-associate may provide valuable information about these influences.
Igh translocations are presumed to occur through aberrant repair during programmed recombination, by recombinase activating gene protein (RAG) during V(D)J recombination, and activation-induced deaminase (AID) during somatic hypermutation and class switching [40]. Cleavage by RAG complexes at cryptic RAG recognition sites at other genes, and altered DNA structures have been implicated in the generation of some human IGH translocations [41,42]. However, the current consensus view is that cryptic RAG sites are not present in the major Myc breakpoint region. Igh translocations within the Igh diversity and joining regions occur in the bone marrow pre-B cells, which undergo V(D)J recombination [40]. However, most Igh translocations to Myc are found in the Igh class switch region and are believed to occur in germinal center B cells, the site of class switching [40]. There is strong evidence to suggest that genes may be susceptible to double-stranded breaks during transcription. The process of transcription creates considerable torsional stress [43], which can be relaxed by topoisomerases via introduction of transient double-stranded breaks. In fact, topoisomerase type IIβ-generated double-stranded breaks in the promoter regions of some genes have recently been shown to be required for regulated transcription [44]. Topoisomerase cleavage sites are common features of translocation hot spots [45]. Significantly, topoisomerase type IIβ binding sites have been mapped to the major breakpoint region of the human MYC gene, at the 5′ end of the first intron [46]. Further evidence of a link between transcriptional organization and recombination is suggested by two papers, one of which showed that actively transcribed yeast tRNA genes cluster in the nucleolus [47], and another which showed that recombination is higher between actively transcribed tRNA genes compared to inactive tRNA genes [48]. Interestingly, double-stranded break repair enzymes Ku70/80 are also associated with transcription factories [49]. In summary, the introduction and repair of double-stranded breaks may be commonplace in transcription factories. Therefore, interchromosomal co-associations between genes in factories may be expected to result in a heightened risk of aberrant repair of double-stranded breaks resulting in chromosomal translocations.
It is curious that evolution has allowed the interchromosomal juxtaposition of the Myc and Igh loci in transcription factories to persist, considering the potentially grave risks of such an organization. However, the apparent dangers of illegitimate recombinations may be outweighed by advantages of clustering transcribing genes, which may make efficient use of shared resources, or perhaps provide a degree of transcriptional coordination of subsets of genes.
CD43− resting B cells were isolated from spleens of 6- to 8-wk-old BALB/c mice by magnetic cell sorting using CD43 microbeads (Miltenyi Biotec, http://www.miltenyibiotec.com) to deplete other cell types. Induction of B cells was done in PBS supplemented with 10 ng/ml recombinant mouse IL-4 (Stemcell Technologies, http://www.stemcell.com), 20 μg/ml purified rat anti-mouse monoclonal antibodies to CD40 (clone HM40–3, Serotec, http://www.serotec.com) and 10μg/ml goat anti-mouse IgM μ chain, F(ab′)2 fragment (Jackson Immunoresearch, http://www.jacksonimmuno.com) at room temperature for up to 15 min before fixation for FISH.
RNA FISH was carried out as described previously [50,51]. We visualized Igh transcription with a dinitrophenol-labeled single-stranded DNA probe to the intronic enhancer region [19], followed by Texas Red detection. We prepared digoxygenin or biotin-labeled single-stranded DNA probes to detect Fos, Myc, Eif3s6, Uros, Actb, Igk, and Igl, intron sequences as described [19]. Primer sequences used to PCR-subclone the various probes are listed below. DNA FISH was carried out as previously described [52]. The following BAC clones (BACPAC Resources, http://bacpac.chori.org) were used: 234 kb RP23-98D8 for Myc; 178 kb RP24-233K8 for Fos; 166 kb RPCI24-258E20 for Igh; 151 kb RP24-132K17 for Uros; and 213 kb RP23-97O1 for Actb. For double-label experiments, we labeled one of the DNA FISH probes directly with AlexaFluor 594 and labeled the other probe with digoxigenin, detected with fluorescein-conjugated antibodies. Immunofluorescence and immuno-FISH was carried out as described [9,52], using a CTD4H8 antibody (Upstate Biotechnology, http://www.upstate.com) that was raised against a Ser5-phosphorylated CTD. This antibody is specific to phosphorylated forms of RNAP II [53].
Primers used to amplify RNA FISH probes were the following. Fos intron 1 sense, 5′-GCTTTGTGTAGCCGCCAGGT-3′; antisense 5′-AGAGGAAAGCGGAGGTGAGC-3′. Fos intron 2 sense, 5′-AAGTAGAGCTGGTGAGCAGCGATT-3′; antisense, 5′-AGAAAAGGACCAACATTCAGTTAAGG-3′. Myc intron 1 sense, 5′-AGCACAGATCTGGTGGTCTTTC-3′; antisense, 5′-CTCCTTCGAGCAGGGACTTAG-3′. Myc intron 2 sense, 5′-CTTCTCCACCACTCATTGGCATTA-3′; antisense, 5′-GGGAGGAAGTGGAAGATCACAGTT-3′. Eif3s6 intron 1 sense, 5′-GTGAGGAAGCTTTGAGAAGGAGGA-3′; antisense, 5′-ATTAATTTTGCTGTTCCCTGCTGA-3′. Uros intron 6 sense, 5′-TCAGCGCCACAGCAAGGGTT-3′; antisense, 5′-GCCTTCCCTCCTTTGTTCCCAGT-3′. Actb intron 1 sense, 5′-TCGCTCTCTCGTGGCTAGTA-3′; antisense, 5′-TGGCGAACTATCAAGACACA-3′. Igh Iμ intron sense, 5′-AGCTGTGGCTGCTGCTCTTA-3′; antisense, 5′-AGCCTCGCTTACTAGGGCTCTC-3′. Igl J-C intron probe 1 sense, 5′-TGAGTGACTCCTTCCTCCTTTG-3′; antisense, 5′-TGGAGGCAGTGTGTAAAGTGTC-3′. Igl J-C intron probe 2 sense, 5′-GTTGTCTTGCAAGGGTCTTTTT-3′; antisense, 5′-GTGCGAATAAAAGAAGGGATTG-3′. Igk J-C intron sense, 5′-AAGACACAGGTTTTCATGTTAGGA-3′; antisense, 5′-AATAGAATTATGAGCAGCCTTTCC-3′.
We examined RNA FISH signals on an Olympus BX41 epifluorescence microscope, and assessed 200 loci for each probe combination, except for Uros-Igh, for which 140 alleles were assessed. Transcription signals scored as colocalizing if the red and green signals overlapped to create a visible yellow signal. To assess the association of Myc DNA FISH signals and RNAP II foci, we captured image stacks of nuclei, using an Olympus BX41 epifluorescence microscope, equipped with a UPlanApo 100× oil objective to reduce chromatic aberration, and fitted with a motorized stage. Images were captured and analyzed using Analysis 3.2 image capture software, fitted with a RIDE module (SIS, http://www.sis.com). The stacks were deconvoluted using a nearest-neighbor algorithm with 85% haze removal, and analyzed. We analyzed 81 and 88 alleles in unstimulated and stimulated B cells, respectively. Statistical analysis was carried out using a two-sided Fisher exact test. For Myc alleles that were not co-associated with an RNAPII focus, we measured the separation distance from the edge of the gene signal to the edge of the nearest RNAP II immunofluorescence signal, and analyzed 27 alleles.
To measure the distances between Igh and genes in trans by DNA FISH, we collected image stacks using a Zeiss 510 Meta confocal microscope. Separation distances for each Igh allele and the nearest Myc, Fos, Uros, or Actb allele were measured on 3D-reconstructed image stacks using Volocity image analysis software (http://www.improvision.com/products/volocity). In all cases, we made measurements from center to center of the two gene signals. We analyzed at least 83 measurements for gene pairs in unstimulated and stimulated B cells, adult kidney cells, and E14.5 fetal liver cells. Changes in the distributions of measurements were assessed by two-sided Student t-test.
The assay was carried out essentially as described [20]. RNA competitor fragments were generated by cloning 230 bp Igh and 236 bp Fos fragments that span 5′ exon-intron junctions. The plasmid containing the Igh fragment was digested with HindIII and AflII to release a 17 bp fragment, then religated. The Fos deletion was generated by BsmFI and NheI digestion to remove a 24 bp fragment. RNA was transcribed from linearized plasmids, then checked by gel electrophoresis and quantitated by UV spectrometry. Dilutions of controlled amounts of RNA was spiked into Trizol reagent that contained a known number of cells. RNA was extracted, reverse transcribed, and PCR amplified with nested primers.
To measure the numbers of primary transcripts over the length of the intron, the relative intensity of the endogenous RT-PCR product was compared to the spiked competitor RT-PCR product, measured using AIDA quantitation software. For Igh, the lanes with four copies of spiked competitor per cell was used for quantitation. For Fos, quantitation was obtained from the average from the lanes with 1 copy per cell and 0.5 copy per cell.
The primers used were as follows. Igh-f, 5′-CCTGGGAATGTATGGTTGTGGCTTC-3′; Igh-r, 5′-CCCCCTAAAGCAAT:GACTGAAGACTCA-3′; Igh nested-f, 5′-CCTCGGTGGCTTTGAAGGAACAAT-3′; Igh nested-r, 5′-CCCTAAAGCAATGACTGAAGACTCAGT-3′; Fos-f, 5′-AGCATCGGCAGAAGGGGCAAAGTA-3′; Fos-r, 5′-TGAAGTAGGAAGCTGTCAGGGAAACTG-3′; Fos nested-f, 5′-AGAAGGGGCAAAGTAGAGCAGGTGA-3′; Fos nested-r, 5′-TGTCAAAATCTGACAAGGGAGGGAAAG-3′.
We carried out the 3C assay as described previously [9]. We fixed B cells that had been stimulated for 5 min in 2% formaldehyde for 10 min at room temperature, and digested 1 × 106 nuclei overnight with 600 units of BglII. We ligated digested chromatin (2 μg) with 2,000 units of T4 DNA ligase in a final volume of 800 μl. We cloned ligation product detected by primer pair d/g (see list below)and verified it by DNA sequencing. We tested the specificity of the 3C primers as previously described [9]. The primers used for 3C analysis were the following. Myc a forward, 5′-TCTACACCCCATACACCTCCA-3′; Myc a nested, 5′-CGAGAATATGCCATGAATTGG-3′. Myc b forward, 5′-GGGGAGGGAATTTTTGTCTATT-3′; Myc b nested, 5′-GGACAGTGTTCTCTGCCTCTG-3′. Myc c forward, 5′-TGCCCTCTCAGAGACTGGTAA-3′; Myc c nested, 5′-TTCCCCTTTCCTCTGTCATCT-3′. Myc d forward, 5′-ATTCTTCCAGGTGGTGATGTC-3′; Myc d nested, 5′-CTTCCCACAGCTCTCTTCCTT-3′. Igh e forward, 5′-AACCCATCTACCCATGTAGCC-3′; Igh e nested, 5′-CCTCTGACTGCCTCTTTTCCT-3′. Igh f forward, 5′-ACTGTGATCGGTTTTGGAGTG-3′; Igh f nested, 5′-CTGGGAGGGTTTGGTTCTTAC-3′. Igh g forward, 5′-CCCAGAACCTGAGAAGGAAGA-3′; Igh g nested, 5′-ACAGAACCGAACCATGACTTG-3′. Igh h forward, 5′-TTGGGCACTAAACACCACTTC-3′; Igh h nested, 5′-GGTGTGTGCAGGTTTTTGTCT-3′. Hbb-b1 forward, 5′-CTCAGAGCAGTATCTTTTGTTTGC-3′; Hbb-b1 nested, 5′-AGGATGAGCAATTCTTTTTGC-3′. Calreticulin Cal1, 5′-CTCCAGATAAACCAGTATGAT-3′; Cal2, 5′-AAACCAGATGAGGGCTGAAGG-3′. Actb Actb 1forward, 5′-CGGTGCTAAGAAGGCTGTTCC-3′; Actb 1nested, 5′-AGCAAGAGAGGTATCCTGACC-3′. Actb Actb 2forward, 5′-TGTGACAAAGCTAATGAGG-3′; Actb 2nested, 5′-TGAGTAGATGCACAGTAGG-3′.
|
10.1371/journal.pcbi.1002032 | How Molecular Motors Are Arranged on a Cargo Is Important for Vesicular Transport | The spatial organization of the cell depends upon intracellular trafficking of cargos hauled along microtubules and actin filaments by the molecular motor proteins kinesin, dynein, and myosin. Although much is known about how single motors function, there is significant evidence that cargos in vivo are carried by multiple motors. While some aspects of multiple motor function have received attention, how the cargo itself —and motor organization on the cargo—affects transport has not been considered. To address this, we have developed a three-dimensional Monte Carlo simulation of motors transporting a spherical cargo, subject to thermal fluctuations that produce both rotational and translational diffusion. We found that these fluctuations could exert a load on the motor(s), significantly decreasing the mean travel distance and velocity of large cargos, especially at large viscosities. In addition, the presence of the cargo could dramatically help the motor to bind productively to the microtubule: the relatively slow translational and rotational diffusion of moderately sized cargos gave the motors ample opportunity to bind to a microtubule before the motor/cargo ensemble diffuses out of range of that microtubule. For rapidly diffusing cargos, the probability of their binding to a microtubule was high if there were nearby microtubules that they could easily reach by translational diffusion. Our simulations found that one reason why motors may be approximately 100 nm long is to improve their ‘on’ rates when attached to comparably sized cargos. Finally, our results suggested that to efficiently regulate the number of active motors, motors should be clustered together rather than spread randomly over the surface of the cargo. While our simulation uses the specific parameters for kinesin, these effects result from generic properties of the motors, cargos, and filaments, so they should apply to other motors as well.
| The spatial organization of living cells depends upon a transportation system consisting of molecular motor proteins that act like porters carrying cargos along filaments that are analogous to roads. The breakdown of this transportation system has been associated with neurodegenerative diseases such as Alzheimer's and Huntington's disease. In living cells, cargos are typically carried by multiple motors. While some aspects of multiple motor function have received attention, how the cargo itself affects transport has not been considered. To address this, we developed a three-dimensional computer simulation of motors transporting a spherical cargo subject to fluctuations produced when small molecules in the intracellular environment buffet the cargo. These fluctuations can cause the cargo to pull on the motors, slowing them down and making them detach from the filament (road). This effect increases as the cargo size and viscosity of the medium increase. We also found that the presence of the cargo helped the motors to bind to a filament before it drifted away. If other filaments were present, then the cargo could bind to one of them. Our results also indicated that it is better to group the motors on the cargo rather than spread them randomly over the surface.
| Cells are highly organized, and much of this organization results from motors that move cargos along microtubules. The single-molecule properties of molecular motors are relatively well understood both experimentally and theoretically. With this as a starting point, we investigated how the presence of the cargo itself alters transport. Aside from exerting viscous drag, the cargo could in principle alter single-motor based transport both by changing the motors' diffusion and ability to contact the filament (a free motor diffuses very differently from a cargo-bound one), and also by exposing the motor to the random forces resulting from thermal fluctuations of the cargo which depend on the size of the cargo and the viscosity of the environment. Whether such effects are significant are investigated here.
Recent studies show that cargos in vivo are frequently moved by more than one microtubule-based motor [1], [2], [3], [4]. This raises the question of how multiple motors function together, the subject of recent theoretical and experimental work [1], [5], [6], [7]. In vitro, when more than one motor is actively hauling a cargo, the run length, i.e., the distance that the cargo travels along the microtubule before detaching, increases with the number of active motors. However, the presence of the cargo itself may be important when there are multiple motors. In addition to possibly changing the single-molecule's function, the cargo's size may alter the relationship between the total number of motors present and the number of motors actively engaged in transporting the cargo (assuming random motor organization on the cargo's surface). If motors are not randomly organized, details of this organization will also be important. How each of these factors contributes to overall transport is unknown.
To approach these problems requires a new theoretical framework: past studies simplified the problem using essentially one-dimensional models [5], [6], [8], [9] that had the motors attached to the cargo at a single point, with the cargo represented by a single point (though potentially experiencing viscous drag proportional to a specific diameter). Here we have developed a bone-fide three dimensional Monte Carlo simulation that allows us to directly investigate how the presence of the cargo itself affects single-motor driven transport and motor-microtubule attachment, as well as how the relationship between cargo size and the arrangement of motors on the cargo affects ultimate cargo motion, all within the context of a cargo experiencing random Brownian translational and rotational motion.
The attachment of motors to a cargo of finite size, rather than an idealized point mass, has a number of ramifications. First, the function of the motor(s) might be altered by the translational and rotational diffusion of the cargo; the larger the cargo, the more effect it has on the motors' diffusion, and thus, potentially, on the motors' ability to contact/interact with a microtubule. Second, when a motor is attached to both the microtubule and the cargo, it will feel instantaneous forces due to the cargo's thermal motion. These forces will depend on the cargo's size; and the random thermal ‘tugs’ from the cargo could slow the rate of travel of a motor and, in principle, induce the motor to detach from the filament. Third, there is a relationship between the cargo size, the total number of motors present, how they are arranged, and how many can be engaged. To illustrate this, imagine one cargo that is 50 nm in diameter, and another that is 500 nm in diameter. In the first case, even if the motors are randomly distributed on the cargo, because the length of an individual motor is more than 100 nm, all of those on the lower half of the cargo, and some on the upper half, will be able to reach a nearby microtubule (Figure 1A). In contrast for the 500 nm cargo, most motors will be unable to reach if they are randomly distributed on the cargo (Figure 1B). However, if all the motors were clumped at a single point, the size of the cargo essentially becomes irrelevant, because if one motor can reach, they all can (Figure 1C).
We thus set out to answer the following questions:
We organized the presentation of our results according to these questions.
To address these questions, we developed three-dimensional Monte Carlo simulations. Generally speaking, Monte Carlo is an approach to computer simulations in which an event A occurs with a certain probability PA where 0≤PA≤1. In practice, during each time step, a random number x is generated with uniform probability between 0 and 1. If x≤PA, event A occurs; if x>PA, event A does not occur.
Our simulations were carried out as follows. We started with a three dimensional spherical cargo, subject to rotational and translational diffusion according to the equations presented below and in the Text S1. To this cargo, we attached kinesin motor(s) that are modeled as bungee cords, i.e., they behave as springs with a spring constant of 0.32 pN/nm [5], [10] when stretched beyond their relaxed length of 110 nm but produce no force when compressed. We started the simulation so that potentially one or more motors could bind to a cylindrical microtubule (25 nm diameter). The motors then moved the cargo along the microtubules, taking 8 nm steps. While technical details of the simulation are in the Text S1, the general idea is that at each time step Δt, we consider all motors present, calculate all forces acting upon them, and then ask what each of them does.
We start by describing how we simulate transport of a cargo with motors attached. Our basic algorithm is as follows. Consider one or more motors attached at random points to the cargo surface. The cargo is then suspended above the microtubule, with a well-defined separation distance between the bottom of the cargo and the top of the microtubule, and the motors are each given an opportunity to attach to the microtubule. If none do (either because none can reach, or because although they can reach, they stochastically are not able to attach in the allotted time with the ‘on’ rate assumed to be ∼2/sec [11], [12], [13]), we use one of two initial conditions. If we want to find the time it takes for a cargo with a single motor to attach, then the cargo is allowed to rotate consistent with Brownian diffusion, and the procedure is repeated. Eventually, the motor binds. The time between when the simulation is started and when the motor attaches is the ‘on’ rate for the cargo; since only one motor is present, it reflects how the presence of the cargo affects the motors' on-rate.
The other initial condition is used if there are multiple motors and we are more interested in transport along the microtubule after the motors attach to the filament. In this case, if none of the motors attaches after being given the opportunity to do so, the cargo is rotated so that at least one motor attaches to the microtubule.
Once some subset of the motors is attached, the cargo travels along the microtubule. At each time step of the simulation, each motor on the cargo is given the opportunity to detach from the MT if it is attached, or attach if it is detached (and geometrically can reach the MT). If a motor is attached to a MT, then there is some probability that it will bind and hydrolyze ATP, and subsequently take a step. Although kinesin is a two headed motor, we model each motor by a single kinesin head that hydrolyzes ATP in such a way that Michaelis-Menten kinetics is obeyed. The probabilities of a motor detaching from the MT, releasing ATP, and taking a step are all dependent on the load on the cargo because the cargo exerts force on the motors (see Text S1. This load has contributions from the externally applied force, the other motors which are pulling the cargo, and from thermal fluctuations. The thermal fluctuations randomly rotate and translate the cargo which, in turn, can stretch the motor linkage and exert a load on the motor. (See below for further details on thermal fluctuations.) Once all the motors have been given a chance to step, the cargo is translated and rotated according to the force and torque to which it is subjected. The cargo travels along the microtubule until all the motors detach from the microtubule, and the ‘run’ ends; this then determines the run length of the cargo. The velocity is calculated by dividing the distance the cargo moves by the travel time τ, where τ is typically 1 msec but may be as long as 10 msec. Averaging over these velocities gives the average velocity. To get good statistics, we simulate a specified number of runs with the same initial conditions to get a set of runs. We also simulate a number of sets with different initial conditions to obtain good statistics.
In our simulations, the spherical cargo is subjected to thermal fluctuations which we can divide into translational and rotational components. The equation of the cargo's translational motion is given by the Langevin equation:(1.1)where m is the cargo's mass and is the cargo's velocity. The drag force on the cargo is proportional to its velocity with the drag coefficient , where R is the cargo's radius and is the coefficient of viscosity which is the kinematic viscosity multiplied by the specific gravity of the fluid. is the sum of the forces due to an external force of magnitude FL and the force of the engaged motors pulling on the cargo. We solve this equation in the Text S1, and quote the solution here for the position of the cargo at time step t+Δt:(1.2)where is the standard deviation of a normal distribution and is a vector in Cartesian coordinates of the laboratory frame of reference that represents three independent random variates drawn on a normal distribution having zero mean and unit standard deviation.
For the cargo's rotational motion, the corresponding Langevin equation is(1.3)where is the moment of inertia of a solid spherical cargo, and is the drag coefficient proportional to the angular velocity . is the torque on the cargo referenced from the center of mass due to the engaged motors. is the rapidly varying random torque due to the thermal fluctuations of the environment. We solve this equation in the Text S1 where we give the formulas for the change in orientation of the cargo at each time step. These formulas are analogous to Eq. (1.2). As we shall see, rotational diffusion due to thermal fluctuations can play a significant role in limiting the distance that a cargo can travel.
After considering motors randomly attached anywhere on the cargo, we consider cases which have a restricted region of the cargo surface area where motors can attach. For these cases, we start each simulation with N motors randomly attached to the cargo's surface within a region specified by the cone angle as shown in Figure 2. The area available for attachment can be described by a cone with its apex at the center of the sphere. A line extends from the apex to the base of the cone. The cluster angle φ is the angle between this line and the side of the cone. The intersection of the cone with the surface of the cargo defines the allowed region of motor attachment. The cluster angle can vary between 0 and 180 degrees. A cluster angle of 90 degrees defines the lower hemisphere of the cargo. A cluster angle of 180 degrees corresponds to the entire spherical surface, and means that the motors can attach anywhere on the sphere.
We organize our results according to the questions posed in the introduction.
Our study of the effects of the cargo on transport has a number of ‘take-home’ messages. The first is that, at both the single-motor and multiple-motor levels, the presence of the cargo can significantly alter the effective ‘on’ rate/probability of successful binding of the motor(s) to the filament, because the center of mass of the cargo diffuses away from the microtubule relatively slowly, and while this is occurring, its rotational diffusion frequently brings the motor close enough to the microtubule to allow attachment. Thus, the cargo ‘helps’ the motor to attach, though the degree of assistance depends on cargo size and viscosity of the medium surrounding the cargo. Rapidly diffusing cargos might not linger long in the vicinity of a microtubule, but in a cell where there are multiple filaments available, these cargos could quickly find and bind to a filament.
Second, in order to for a motor to attach to the filament in a reasonable amount of time, the motor length needs to be longer or comparable to the radius of the cargo which may explain why motors are 60 to 110 nm in length.
Third, if motors are randomly arranged on the cargo's surface, the relationship between the number of motors present and the number of actually engaged motors depends strongly on the cargo size, so that different simple models of regulating cargo motion by recruiting motors to the cargo surface (either by a specified change in total number of motors, or by a specified change in local motor surface density) will have different effects on overall cargo motion as a function of cargo size. Thus, in order to have regulation affect a set of cargos equally, independent in variations in cargo size, it is best to have motors clustered in a small region on the cargo.
A further finding also supports the utility of motor clustering: for large cargos, if motors are randomly placed, achieving a reasonable number of engaged motors (n = 3–6) would require a large number of motors (50–100) to be present on the cargo, which appears inconsistent with biochemical characterizations of cargo-bound microtubule motors [4], though it is consistent with biochemical characterizations of cargo-bound myosin motors [18] which are likely randomly arranged on cargos [18], [19]. Overall, our findings suggest that, in vivo, microtubule motors are likely organized into clusters when present on large cargos, but that such clustering is unnecessary for small cargos.
In addition, a reasonable number of engaged motors would be required for long travel distances of several microns but not for short run lengths. Since microtubules can be tens of microns long compared to actin filaments which have a typical decay length of 1.6 microns [18], we expect long travel distances along microtubules but relatively short run lengths along actin filaments. Thus we predict the microtubule motors kinesin and dynein to be clustered on cargos while we expect the actin motor myosin V to bind randomly to cargos. There is clear experimental evidence for the random arrangement of myosin on cargos in vivo, and weak experimental evidence for the clustering of kinesin and cytoplasmic dynein [19].
For the purposes of this paper, we have assumed that the points where motors are attached to the cargos are fixed on the cargo's surface. This is true in some cases, e.g., when motors bind to dynactin which in turn binds to spectrin which is a filament that coats some vesicles [20], [21]. However, in other cases, the attachment points can diffuse through the fluid membrane of the vesicle and cluster at one location. An example of this is an experiment showing that motors dynamically accumulate at the tip of membrane tubes growing out of a vesicle as a consequence of the fluidity of the membrane [13], [22].
Clustering does not seem to affect the rate at which the first motor of a cargo attaches to a microtubule unless the cargo is large (greater than 200 nm) and the viscosity is high. Motor proteins are sufficiently long (greater than 50 nm) and rotational diffusion sufficiently rapid that the number of motors on a cargo does not significantly affect the rate at which the cargo binds to the microtubule.
|
10.1371/journal.pntd.0005800 | The location of Australian Buruli ulcer lesions—Implications for unravelling disease transmission | Buruli ulcer (BU), caused by Mycobacterium ulcerans, is increasing in incidence in Victoria, Australia. To improve understanding of disease transmission, we aimed to map the location of BU lesions on the human body.
Using notification data and clinical records review, we conducted a retrospective observational study of patients diagnosed with BU in Victoria from 1998–2015. We created electronic density maps of lesion locations using spatial analysis software and compared lesion distribution by age, gender, presence of multiple lesions and month of infection.
We examined 579 patients with 649 lesions; 32 (5.5%) patients had multiple lesions. Lesions were predominantly located on lower (70.0%) and upper (27.1%) limbs, and showed a non-random distribution with strong predilection for the ankles, elbows and calves. When stratified by gender, upper limb lesions were more common (OR 1·97, 95% CI 1·38–2·82, p<0·001) while lower limb lesions were less common in men than in women (OR 0·48, 95% CI 0·34–0·68, p<0·001). Patients aged ≥ 65 years (OR 3·13, 95% CI 1·52–6·43, p = 0·001) and those with a lesion on the ankle (OR 2·49, 95% CI 1·14–5·43, p = 0·02) were more likely to have multiple lesions. Most infections (71.3%) were likely acquired in the warmer 6 months of the year.
Comparison with published work in Cameroon, Africa, showed similar lesion distribution and suggests the mode of M. ulcerans transmission may be the same across the globe. Our findings also aid clinical diagnosis and provide quantitative background information for further research investigating disease transmission.
| Buruli ulcer is an emerging tropical disease that is also increasingly common in the temperate Australian state of Victoria. The mode of transmission of this geographically restricted infection remains elusive. We have accurately mapped the location of 649 PCR-confirmed Buruli lesions affecting 579 patients and displayed their position on front and back human body diagrams. Lesion distribution density was assessed with computer-generated heat-maps. Buruli lesion distribution was most common on exposed parts of the body (distal limbs). However, even on exposed areas, lesion distribution was highly unevenly distributed and focused towards ankles, backs of calves and elbows. The palmar and plantar surfaces of hands and feet were rarely affected. We propose that targeting behavior by biting insects rather than direct contact with a contaminated environment best explains the lesion distribution we observed.
| Buruli ulcer (BU), listed by the World Health Organisation (WHO) as a neglected tropical disease, is a destructive infection of the subcutaneous tissue caused by the acid-fast bacillus Mycobacterium ulcerans [1,2]. It is endemic in at least 33 countries and is the third most prevalent mycobacterial disease worldwide after tuberculosis and leprosy [3]. The natural history of BU begins as a small nodule or plaque that usually progresses into a large, necrotic ulcer if left untreated. In severe cases, the disease can result in significant cosmetic and functional deformities. Despite vastly different climatic and socioeconomic conditions, West Africa and the Australian state of Victoria are the two most commonly affected regions globally [4]. In Victoria BU occurs in well-defined endemic zones including the Mornington and Bellarine Peninsulas near Melbourne [5]. These are coastal regions in a climactically temperate region where the annual incidence of BU has progressively increased since the mid-1990s [5–8].
A characteristic of BU is its sharp geographical restriction, often to quite small areas of just a few square kilometres [5]. However, the reservoir and mechanisms of disease transmission are yet to be determined. There are two general theories of transmission. The first is inoculation from a contaminated environmental source through contact with sharp leaves, thorns and bushes (for example), or through direct exposure of existing wounds [4]. The second is inoculation via an insect vector which has itself been previously contaminated [4]. As the pathogen grows best at temperatures slightly lower than core body temperature, some researchers have also considered a third possibility involving acquisition of M. ulcerans by aerosol or inoculation, followed by silent dissemination and local reactivation at relatively cooler peripheral body sites [9,10].
The distribution of BU lesions on the human body was investigated recently in Cameroon using spatial analysis in 88 confirmed cases [11]. The main finding was a non-random distribution with lesions tending to cluster at large joints such as the ankles and elbows. These observations support previous research indicating the upper and lower limbs are the most commonly affected body sites [6,12]. Using a similar method, we have conducted our own investigation on a much larger cohort of confirmed cases from Victoria. We aimed to quantitatively assess lesion location to better understand transmission and to define differences and similarities between disease in temperate and tropical zones. We also intended to determine more comprehensively the clinical pattern of disease presentation in Australian populations to assist clinicians recognise BU earlier and minimise morbidity.
This was a multisite, retrospective, observational study of patients diagnosed with BU in Victoria. We aimed to include all cases from 1 January 2004 when the disease was first made legally notifiable by the Department of Health & Human Services (DHHS) until December 31, 2015. Notifications prior to January 2004 were voluntary. DHHS keeps records of cases including clinical and epidemiological information as part of their enhanced surveillance program. Additional cases diagnosed between 1998 and 2003 by clinicians involved in the study were also included. Cases were confirmed either by a positive culture or IS2404 PCR for M. ulcerans (generally both) [13]. If patients developed additional BU lesions more than 12 months apart, these were defined as re-infections and included as separate BU episodes.
Primary data collected for all patients included the physical location of lesions on the body, gender, patient age at diagnosis, and the geographic endemic region where the patient was most likely infected. Endemic areas comprised the Bellarine Peninsula, the Mornington Peninsula, and ‘Other’ made up of Phillip Island, the Frankston Area, Geelong, East Gippsland, Melbourne’s South East Suburbs and Interstate endemic regions. When patients had travelled or resided in multiple different endemic areas they were excluded from the sub-analyses that compared patients by region.
To increase the accuracy of lesion location descriptions, we contacted clinicians responsible for treating patients from seven major tertiary hospitals in Victoria, and requested they map the location of lesions from their own records in a systematic, standardised manner. Clinicians used a variety of sources to identify lesion locations including clinical notes, referral letters, pathology reports and photographic evidence. When lesions were large, we asked them to locate the likely initial origin and map this point. All mapping from clinicians was initially completed on hard-copy ‘front’ and ‘back’ templates of the human body which were printed onto standard A4 graph paper.
Coordinates for each lesion were then collected and recorded electronically. Front and back shapefiles were created separately by inputting x and y coordinates from outlines of the hardcopy templates and adding this data to blank ArcGIS maps. Coordinates of lesions were then inputted and similarly added as another layer to the previously created templates. A kernel density analysis and raster clipping tool within the software was used to visualise lesion distribution. All electronic mapping and spatial analysis was conducted using ESRI’s ArcGIS ArcMap (Economic and Social Research Institute, Redlands, USA, version 10·3·1), RStudio (RStudio, Boston, USA, version 0·99·893) and R (The R Foundation for Statistical Computing, version 3·3·1).
The localisation of lesions was categorised into single specific body regions (e.g. foot, ankle) or groups of specific body regions. These were then analysed to determine if there were specific distribution patterns that varied by gender, age, the presence of multiple lesions or by average maximal daily temperatures for the endemic area. The groups of specific body regions included; upper limb (hand, wrist, forearm, elbow, arm, shoulder); lower limb (foot, ankle, leg, knee, thigh, buttocks); distal lesions (hand, wrist, forearm, elbow, foot, ankle, leg, knee); proximal lesions (arm, shoulder, thigh, buttock, face, neck, abdomen, back, chest); arm and shoulder combined; hand, wrist, forearm and elbow combined.
To determine the likely date of infection we analysed a subset of patients from one health service (Barwon Health) with known duration of symptoms prior to presentation. The estimated date of infection was determined by subtracting from the date of diagnosis the number of days of symptoms prior to diagnosis and the estimated mean incubation time for M. ulcerans in Victoria (135 days) [5]. We then categorised the dates of estimated infection in groups of 3 months from hottest to coldest according to the average daily temperatures in Victoria from the Australian Bureau of Meteorology [14]. Associations between the hottest 3 months and the coldest 6 months were determined using univariate analyses with the two coldest month categories combined (May to October) to provide a sample size large enough to allow meaningful comparative analyses.
Data was analysed using STATA 14 (StataCorp, Texas, USA). Univariate analyses were performed using Mantel-Haenszel and multivariate analyses were performed using logistical regression analyses. A p-value of less than 0.05 was deemed significant.
Our study was performed as part of the ongoing enhanced surveillance program conducted by the Department of Health and Human Services of Victoria (DHHS). Low risk ethics approvals were obtained from the Institutional Review Board at each participating clinical site to allow us to access any missing demographic patient data and permit the most accurate possible lesion location through each patient’s treating clinician (IRBs: Austin Health, Barwon Health, Peninsula Health, Monash Health, Royal Children’s Hospital, Melbourne Health, Alfred Health). Institutional Review Boards did not require us to obtain consent as data we were collating and analysing had already been notified to DHSS under the Public Health and Wellbeing Act 2008 and patients were not re-contacted. All patient data analysed were anonymized.
Of the 694 patients diagnosed with BU during the study period, including 27 who were diagnosed prior to 2004, we have been able to review the clinical records of 538 patients (77·5%). Combined with an additional 41 patients with adequate data on lesion location, the final analysis included 579 patients (83·4%) who contributed to a total of 585 episodes of infection and 649 lesions (Fig 1). There were six patients (1·0%) with two episodes of infection that were more than 12 months apart and 32 patients (5·5%) with multiple lesions, including those with likely reinfection and multiple lesions (n = 2), with the median number of lesions being 2 (IQR 2–3; Range 2–13).
Among the 585 total episodes of infection, 54·5% (n = 319) were from male patients and 45·5% (n = 266) were from female patients. The median age was 55 years (interquartile range = 30 to 71 years) while the age range was 1 to 95 years. Children 15 years and under accounted for 12·8% of patients (n = 75) while 51·5% (n = 301) were between 15 and 65 years, and 35·7% (n = 209) greater than or equal to 65 years. Most patients (n = 538; 92·0%) reported exposure in one endemic region, with 67·5% (n = 363) exposed on the Bellarine Peninsula, 19·5% (n = 105) exposed on the Mornington Peninsula and 13·0% (n = 70) exposed in other regions. A further 6·5% (n = 40) reported exposure in more than one endemic location and no information about the location of exposure could be obtained for the remaining 7 (1.2%) patients.
As shown in Figs 2–5 and Table 1, we observed a qualitatively non-random distribution of BU lesions on the human body. When comparing body regions (Fig 2 & Table 2), lesions were most common on the upper and lower limbs accounting for 27·1% and 70·0% of all lesions respectively. There were comparatively very few lesions on the trunk (1·7%) and the head or neck regions (1·2%). On limbs, it was the distal regions that were most affected. The calves and ankles showed the greatest density on the lower limb while the elbow and dorsal surfaces of the forearm and hand were the regions mainly affected on the upper limb.
There were no obvious differences comparing the right or left side of the body with similar density patterns on each. There were two midline lesions from two different children, one at the natal cleft and the other on the nose. Notably, only one lesion was found on the sole of the foot and none were found on the palms of the hands. Lesions were over a joint in 35.6% of cases. Comparing distribution between large joints of the upper and lower limbs (Table 1), the ankles accounted for 15·7% (n = 102) while the elbow and knees were the next most affected with 9·7% (n = 63) and 6·9% (n = 45) of all lesions respectively. An example of a severe case of Buruli ulcer crossing a joint and acquired on the Mornington Peninsula is shown in Fig 6.
When stratified by gender and adjusted for age, the odds of males having a lesion on the upper limb was almost twice that compared to females (OR 1·97, 95% CI 1·38–2·82, p<0·001) but halved when comparing with females on the lower limbs (OR 0.48, 95% CI 0.34–0.68, p<0.001). Males also had a significantly higher likelihood of having lesions on distal regions of the upper limb, that is, the hand, wrist, forearm and elbow combined (OR 2·41, 95% CI 1·61–3·61, p<0·001) but a significantly lower likelihood of having lesions on the foot (OR 0·37, 95% CI 0·19–0·71, p = 0·002).
When stratified by age and adjusted for gender, patients aged ≥65 years compared to those <65 years were significantly less likely to have proximal lesions (OR = 0·57, 95% CI 0·36–0.91, p = 0·02), significantly less likely to have a lesion on the arm and shoulder combined (OR = 0·28, 95% CI 0·11–0·67, p = 0·001) but significantly more likely to have a lesion on the hand, wrist, forearm and elbow combined (OR 1·91, 95% CI 1·30–2·81, p = 0·001).
Qualitative comparisons in BU distribution revealed no obvious differences between exposure on the Bellarine and Mornington Peninsulas (Fig 5).
Of the 32 patients who had multiple lesions at the time of presentation, 17 (53%) had lesions located on separate limbs. Patients aged ≥ 65 years (OR 3·22, 95% CI 1·53–6·78, p = 0·001) and those with a lesion on the ankle (OR 2·71, 95% CI 1·23–5·98, p<0.01) were significantly more likely to have multiple lesions. There was no significant difference when comparing by gender in this regard (male compared with female OR 1.08, 95% CI 0.52–2.21, p = 0.84).
The probable date of infection was estimated for 338 (59%) patients. These patients were grouped in 3-month intervals from hottest to coldest in descending order as shown in Table 3. Lesions on the arm and shoulder combined (OR 3·37, 95% CI 1·45–7·86, p = 0·003), as well as proximal lesions (OR 1·78, 95% CI 0·98–3·21, p = 0·05), were significantly associated with the hottest 3 months (Dec-Feb: 285 cases) when compared to the coldest 6 months (May-Oct: 103 cases).
We have shown a focal, non-random distribution of BU on the human body in Victoria, a Buruli endemic region with increasing disease incidence and a temperate climate. Areas of the body most affected include the ankles, calves, elbows, knees, and dorsal surfaces of the hands and forearms. BU was rare on the palms of the hands, soles of the feet, head, neck, and trunk. Combined, these findings suggest BU lesions are generally found on the distal regions of the upper and lower limbs, except the palms and soles, particularly around large joints. Our findings are similar to previous research on lesion localisation and the similarity in BU distribution in Victoria when compared with studies carried out in Africa raises the possibility of a unifying mechanism of transmission worldwide [6,11].
The distribution pattern of BU we have observed appears to correlate with exposed skin areas not covered by clothing. Indeed, a failure to wear protective clothing has been previously documented as a risk factor for developing the disease [15]. The warmer summer months of the year may therefore be periods of high exposure risk when less clothing is worn. This is supported by our findings of an increased proportion of cases likely acquired in warmer months, and the fact that warmer months were associated with an increased likelihood of having lesions on proximal body regions that are less likely to be protected by clothing during these periods.
Stratification by age and gender revealed further differences in distribution. Compared to younger individuals, those aged greater than 65 had fewer lesions on proximal areas of the upper limbs. Similar findings have been noted in Africa where children and young adults were found to have more proximal lesions than those older [12]. Again, this may be due to clothing choices as older individuals may tend to wear longer sleeved upper garments. Comparing between genders revealed that upper limb lesions were more common in males, particularly at the elbows, while lower limb lesions were more common in females, especially the dorsum of the foot. This may be due to females choosing to wear more open footwear and longer-sleeved upper garments. Alternatively, males may be more prone to trauma, for example, when working as manual labourers. Future research is required to further explore these hypotheses.
While the exact mechanism of BU transmission remains unclear, an inoculating event such as direct trauma or insect bites is commonly thought of as a pre-requisite for disease emergence [15]. One theory of transmission is inoculation through direct contact with contaminated environmental sources such as sharp leaves or thorns or through existing wounds [4]. It is plausible that random trauma through environmental contact could produce a specific distribution pattern of BU. Additionally, previous research has shown body regions such as the ankles, shins and elbows, where skin lies close to bone, as being more prone to injury in children [16]. However, we have also observed BU to cluster at the calves, regardless of age. The predilection for BU on this relatively protected part of the body may therefore suggest transmission in this body region is less likely to occur via existing wounds or casual trauma.
We feel our data from Victoria fits best with the hypothesis that M. ulcerans is transmitted via insect vectors such as mosquitoes [4,17,18]. With the recent discovery that possums, a mammal native to Australia, can harbour M. ulcerans and subsequently develop clinical disease, it has been proposed that mosquitoes act as mechanical vectors connecting possums with humans living nearby [10,17]. Mosquitoes use a variety of visual, chemical and thermal cues to target a suitable location for blood meals in humans and published data exists showing a species-dependent selection of biting sites which are characteristically non-random [19,20]. As BU occurs mainly at exposed large joints and the back of the legs, these areas could be targets for Australian mosquito populations in endemic zones. Mapping of species-specific mosquito biting preferences and comparison with our density maps could provide a unique opportunity to test this hypothesis. Future research could also compare our maps with whole body thermographs to examine the theory that M. ulcerans has a preference for cooler body sites.
We accept that there may be different modes of Buruli transmission in different regions of the world. Studies of lesion location performed in Africa have sometimes shown a predominance of right-sided lesions, suggesting that acquisition of infection follows contact between the preferred leading arm or leg and a contaminated environment [21,22]. However we did not find evidence for this in our study nor is this pattern uniform in all African case series [23]. So far, investigation of field captured mosquitoes [24] and small animals [25] in Africa has not identified an analogue to the possum-amplifier-mosquito-vector scenario we have proposed for southern Australia [17]. Nevertheless, the generally held view that M. ulcerans is widespread and free-living in endemic environments and that direct contact is all that is required to acquire infection needs to be challenged. This view is not supported by our findings, or those from Africa showing sparing of the soles of the feet and palms of the hand. Furthermore, analysis of the whole genome sequence of M. ulcerans suggests a niche-adapted rather than free-living existence, probably in association with other biota, as there are more than 700 gene deletions or interruptions compared with its M. marinum progenitor [26]. In support of this prediction M. ulcerans has only been isolated in pure culture directly from the environment once, and that was from a water insect captured in Benin [27]. Furthermore, while there are differences, there are also similarities between our results and the distribution of lesions reported by Bratschi et al in Cameroon on which we have based this research [11]. Hence it is possible that biting insects are the predominant mode of transmission of M. ulcerans everywhere. If true, prevention of insect bites should substantially reduce the incidence of Buruli ulcer in people living in endemic areas.
Limitations of our research include being unable to collect data on approximately 17% of the total number of patients notified as cases during our study period. However, we have mapped a large cohort of patients and it is unlikely missing cases would significantly alter our principal findings. There may also be a possible loss of spatial resolution due to the 2-dimensional nature of our heat-map representation. In total, there were 185 lesions (28·5%) which that were located on the medial or lateral sides of the body. Nevertheless, our interest in lesion localisation was the anterior or posterior aspects of a body region or joint and our mapping process was standardised to include side lesions on the front template only, for easier visualisation.
In conclusion, our results provide clear evidence for a highly specific, non-random distribution of BU lesions on the human body. Our study was significantly larger than a BU lesion distribution study performed in Cameroon but we found a generally similar pattern suggesting M. ulcerans transmission and pathogenesis may be similar across the world despite very different geographical and climatic conditions. Our study will also inform clinicians who need to consider the differential diagnosis of skin lesions in routine clinical practice by being able to compare with our density maps, as well as guide future researchers interested in understanding disease transmission and its prevention.
|
10.1371/journal.pbio.1000155 | CFTR Delivery to 25% of Surface Epithelial Cells Restores Normal Rates of Mucus Transport to Human Cystic Fibrosis Airway Epithelium | Dysfunction of CFTR in cystic fibrosis (CF) airway epithelium perturbs the normal regulation of ion transport, leading to a reduced volume of airway surface liquid (ASL), mucus dehydration, decreased mucus transport, and mucus plugging of the airways. CFTR is normally expressed in ciliated epithelial cells of the surface and submucosal gland ductal epithelium and submucosal gland acinar cells. Critical questions for the development of gene transfer strategies for CF airway disease are what airway regions require CFTR function and how many epithelial cells require CFTR expression to restore normal ASL volume regulation and mucus transport to CF airway epithelium? An in vitro model of human CF ciliated surface airway epithelium (CF HAE) was used to test whether a human parainfluenza virus (PIV) vector engineered to express CFTR (PIVCFTR) could deliver sufficient CFTR to CF HAE to restore mucus transport, thus correcting the CF phenotype. PIVCFTR delivered CFTR to >60% of airway surface epithelial cells and expressed CFTR protein in CF HAE approximately 100-fold over endogenous levels in non-CF HAE. This efficiency of CFTR delivery fully corrected the basic bioelectric defects of Cl− and Na+ epithelial ion transport and restored ASL volume regulation and mucus transport to levels approaching those of non-CF HAE. To determine the numbers of CF HAE surface epithelial cells required to express CFTR for restoration of mucus transport to normal levels, different amounts of PIVCFTR were used to express CFTR in 3%–65% of the surface epithelial cells of CF HAE and correlated to increasing ASL volumes and mucus transport rates. These data demonstrate for the first time, to our knowledge, that restoration of normal mucus transport rates in CF HAE was achieved after CFTR delivery to 25% of surface epithelial cells. In vivo experimentation in appropriate models will be required to determine what level of mucus transport will afford clinical benefit to CF patients, but we predict that a future goal for corrective gene transfer to the CF human airways in vivo would attempt to target at least 25% of surface epithelial cells to achieve mucus transport rates comparable to those in non-CF airways.
| The ciliated epithelium that lines the conducting airways of the lung normally functions to transport hydrated mucus secretions out of the airways to maintain respiratory sterility. Cystic fibrosis (CF) lung disease results from reduced airway surface hydration leading to decreased mucus clearance that precipitates bacterial infection and progressive obstructive lung disease. CF is a genetic disease, and the mutant protein is a chloride ion channel (CFTR) that normally regulates ion and fluid transport on the airway surface. Restoration of corrected CFTR function to the airway epithelium of CF patients by delivering a new CFTR gene to airway epithelial cells has long been envisioned as a therapeutic strategy for CF lung disease. Towards this goal, we use a novel viral vector to deliver CFTR to a culture model that represents the ciliated airway epithelium of CF patients and show that this strategy restores airway surface hydration and mucus transport to levels of that in non-CF individuals. This study demonstrates efficient and efficacious CFTR delivery to CF ciliated airway epithelium and that CFTR delivered to approximately 25% of the surface epithelial cells restores normal levels of airway surface hydration and mucus transport. These studies serve as a benchmark for the efficiency of CFTR gene delivery to CF airways for future CF gene therapy studies in vivo.
| Cystic fibrosis (CF) is the most common recessive lethal genetic disorder in Caucasian populations and results from a defect in the CFTR gene. Although CF affects many organs, the pulmonary manifestations account for over 90% of the morbidity and mortality [1]. Dysfunction of CFTR in CF airway epithelium perturbs the normal regulation of ion transport, leading to a reduced volume of airway surface liquid (ASL), mucus dehydration, decreased mucus transport (MCT), and mucus plugging of the airways, which are hallmarks of early CF lung disease. Failure of effective mucus clearance initiates and exacerbates CF lung disease, resulting in an inability to effectively prevent or eradicate bacterial infection, typically dominated by Pseudomonas aeruginosa. Persistent neutrophil-mediated inflammation in CF airways further compromises defective clearance and, over several decades, results in airway destruction and fatal decline of lung function.
Airway mucus clearance is dependent on MCT facilitated by ciliated cell function and cough clearance, and constitutes “mechanical” innate defense of the lung [2]. Both MCT and cough clearance require sufficient hydration of mucus secretions for effective airway clearance. The currently accepted model for CF lung pathogenesis is that absence of CFTR function leads to ASL volume reduction that results in mucostasis. A logical therapeutic strategy to reverse CF lung pathogenesis would therefore replace CFTR function to CF airway epithelium, restoring ASL volume regulation and MCT. CFTR gene delivery strategies remain a rational approach towards this goal. To date, however, clinical trials in CF patients using CFTR gene delivery techniques have resulted in limited successful gene delivery that is widely considered to be insufficient for therapeutic benefit to CF patients. The fundamental hurdle to these approaches is the low efficiency of CFTR gene delivery to human conducting airway epithelial cells that regulate MCT.
In non-CF human airways, CFTR is expressed in ciliated airway epithelial cells of the surface and submucosal gland ductal epithelium [3], and in the fluid-secreting cells of the submucosal glands [4]. Ciliated cells are the predominant luminal epithelial cell type present throughout the proximal and distal conducting airways and are critical cell types for facilitating MCT [5]. These properties of ciliated cells make them an abundant and relevant target for CFTR gene delivery. Since restoration of MCT or ASL volumes to normal non-CF levels has not been described after delivery of CFTR to CF airway epithelium either in vitro or in vivo, it remains unknown how much CFTR or how many surface epithelial cells will be required to express CFTR to restore mechanical innate defense to the airways.
In vitro models of human ciliated airway epithelium (HAE) recapitulate the morphology and physiology of the human airway epithelium and have been a valuable tool in the study of cell physiologic mechanisms that regulate ion and fluid transport and MCT in the human ciliated conducting airways [6]–[8]. Importantly, HAE retain phenotypic differences between non-CF and CF airway epithelium, i.e., CF HAE exhibit reduced or absent cAMP-mediated chloride ion (Cl−) secretion, dysregulation of sodium ion (Na+) absorption, excessive ASL absorption, dehydration of secreted mucus, and mucostasis [6]. This model has been predictive of the in vivo efficacy of drug- and gene-based therapeutics in human clinical studies [9]–[16]. In particular, this model has been utilized to determine that currently available gene transfer vectors approved for clinical testing (e.g., adenovirus, lentivirus, AAV, and nonviral vectors) are inefficient at delivering CFTR to sufficient numbers of surface epithelial cells to restore CFTR function or ASL volume for effective MCT. The target number of cells in a ciliated airway epithelium needed to express CFTR for restoration of ASL and MCT is currently unknown.
We have shown that human parainfluenza virus (PIV) selectively targets ciliated cells of HAE after luminal delivery [17]. Since recombinant PIV can be re-engineered to express large transgene inserts [18], we used PIV to: (1) generate PIV-expressing CFTR as an additional gene (PIVCFTR); (2) test whether delivery of CFTR to CF ciliated cells restored mechanical innate defense, i.e., MCT, to human CF airway epithelia; and (3) determine the numbers of CF epithelial cells requiring CFTR to restore MCT rates to normal non-CF HAE levels. Using CF HAE, CFTR expression levels per cell and numbers of cells expressing CFTR were correlated with correction of ion transport, ASL volume regulation, and MCT rates to assess the relationship between gene transduction and restoration of normal mucociliary transport. We show that PIV-mediated delivery of CFTR to ciliated cells of CF HAE resulted in functional CFTR channel activity with restoration of ASL volume homeostasis and MCT. Further, we show that CFTR expression in individual ciliated cells does not require tight regulation of expression and that restoration of MCT rates to those measured in non-CF HAE required CFTR delivery to at least 25% of surface epithelial cells or approximately 30% of ciliated cells. Hence, we describe the first demonstration, to our knowledge, of efficient CFTR gene delivery to CF ciliated airway epithelium that is sufficient to correct the fundamental physiological dysfunctions that precipitate CF lung pathogenesis.
Recombinant PIV-expressing GFP (PIVGFP) infects ciliated cells of an in vitro model of human airway epithelium that recapitulates the morphology of the human ciliated airway epithelium in vivo (Figure 1A, 1B, and 1C). Inoculation of freshly excised human tracheobronchial airway epithelium showed PIVGFP also targeted ciliated cells under noncultured conditions (Figure 1D).
To express CFTR in CF ciliated cells, a PIV with CFTR inserted into the viral genome was constructed (PIVCFTR, Figure S1). Apical surface inoculation of CF HAE with PIVCFTR or PIVGFP at 106 plaque-forming units (PFU) (100 µl of 107 PFU/ml for 2 h: multiplicity of infection [MOI] ∼3 for all lumenal cells and ∼5 for ciliated cells) resulted in infection of a significant number of cells 48 h postinoculation (pi) as detected by immunolocalization of PIV fusion (F) glycoprotein viewed en face (Figure 2A and 2B). Immunolocalization of PIV-mediated GFP and PIV F expressed by PIVGFP or F glycoprotein expressed by PIVCFTR confirmed targeting of ciliated cells (Figure 2C and 2D). Quantitation of lumenal cells infected by PIVGFP or PIVCFTR revealed that similar numbers of ciliated cells were infected by each virus (Figure 2E). Although ciliated cell numbers were variable in HAE derived from different donors (range 60% to 80%), the percentage of ciliated cells per square centimeter of epithelium surface area determined by β-tubulin IV immunodetection showed that on average approximately 70% of surface cells were ciliated cells. Since the total number of lumenal surface cells in HAE approximates 0.3×106 cells, we estimate that approximately 42% of surface cells or approximately 60% of ciliated cells were infected by PIV under these conditions.
The PIV vector replicates in ciliated cells and by 24 h pi, sheds progeny virions onto the culture surface, resulting in further rounds of ciliated cell infection. However, when low viral titers (103 PFU) were used to inoculate HAE, evidence of further rounds of infection (i.e., spread) was not apparent until 48 h pi [17]. To determine whether the high efficiency of ciliated cell infection by PIVCFTR was dependent on spread of progeny virus from ciliated cell to ciliated cell, we compared infection rates at 24 h versus 48 h pi. HAE inoculated with PIVCFTR (106 PFU) for 2 h showed similar high infection rates at 24 h pi as for 48 h (2-h data for Figure 2F vs. 2E), indicating that the initial high titer inoculum and not cell–cell spread of virus mediated the highly efficient targeting of ciliated cells by PIV. Prolonging or decreasing inoculation times to 8 h or 5 min produced modestly increased or decreased infection rates, respectively (Figure 2F). The high numbers of PIVCFTR-infected cells with only 5-min inoculation time highlight the high efficiency of PIV infection and suggest that short exposure times in the airways will be sufficient for efficient targeting of ciliated cells in vivo.
Although equally efficient at targeting ciliated cells as PIVGFP, PIVCFTR stimulated significantly lower amounts of epithelial cell-derived inflammatory mediators associated with in vivo viral pathogenesis. Indeed, PIVCFTR only induced inflammatory mediator secretion to levels stimulated by inoculation with UV-inactivated PIV (Figure 2G for CXCL8, and Figure S2 for CXCL10, IL-6, IL-12p40, MCP-1, and RANTES). Since PIVCFTR produced 10-fold fewer progeny virions than PIVGFP due to the insertion of the relatively large CFTR insert (Figure 2H), it is likely that the generation of inflammatory mediators is proportional to the rate of PIV replication in ciliated cells.
Expression levels of transduced CFTR in CF HAE were determined by comparing the levels of exogenous CFTR mRNA expressed by PIVCFTR relative to endogenous CFTR mRNA in CF HAE and non-CF HAE using quantitative RT-PCR. Previously, it has been estimated that human airway cells contain only approximately 10 CFTR transcripts/cell [19]. In our experiments, we found that CF HAE inoculated with PIVCFTR produced a 236-fold increase in CFTR mRNA when compared to cultures inoculated with PIVGFP or mock (Figure 3A). Since this large-fold increase in CFTR mRNA is in part reflective of the low endogenous copy number of CFTR, we also assessed CFTR protein levels semiquantitatively by western blot. CF HAE inoculated with PIVCFTR expressed large amounts of mature CFTR (Figure 3B, lane 3, band C), whereas no mature CFTR protein was detected in CF HAE inoculated with vehicle alone (lane 1) or PIVGFP (lane 2). Serial 10-fold dilutions of total protein lysates of CF HAE inoculated with PIVCFTR (lanes 4 and 5) provided a semiquantitative measurement of the amounts of exogenous CFTR protein in CF HAE compared to CFTR protein levels in non-CF HAE (lane 6). We estimate that an approximately 50-fold increase in mature CFTR protein in transduced CF HAE was achieved compared to non-CF HAE. Therefore, two independent measures of CFTR abundance indicate a significant overexpression of both CFTR mRNA and protein in transduced CF HAE. Note, these measures are likely an underestimate given that not all ciliated cells are infected by PIVCFTR. Because approximately 60% of ciliated cells were infected in these experiments, and ciliated cells on average comprise approximately 70% of surface cells within a culture, we estimate that individual infected ciliated cells likely overexpress CFTR protein by at least 100-fold over non-CF ciliated cells.
Apical localization and overexpression of CFTR above endogenous levels in ciliated cells was confirmed by immunodetection of CFTR in CF HAE and non-CF HAE (Figure 3C). For these studies, we chose to engineer both GFP and CFTR into a single PIV vector (PIVGFPCFTR) to enable identification of infected cells by GFP fluorescence. In CF HAE infected with PIVGFPCFTR, CFTR was immunolocalized only to ciliated cells that were also positive for GFP (Figure 3Ci) and concentrated in apical membrane domains at the base of the cilial shafts. Although endogenous CFTR in non-CF ciliated cells in vitro is localized to these regions [3], subapical membrane CFTR immunoreactivity was also detected after PIVGFPCFTR, likely suggesting the increased presence of CFTR in recycling endosomes. Infection of CF HAE with PIVGFP alone showed that ciliated cells positive for GFP were negative for CFTR immunoreactivity (Figure 3Cii). When non-CF HAE were infected by PIVGFPCFTR, endogenous CFTR was present in GFP-negative ciliated cells and overexpressed in GFP-positive cells (Figure 3Ciii). For non-CF HAE infected with PIVGFP, GFP-positive and -negative ciliated cells showed only endogenous CFTR apical membrane immunoreactivity (Figure 3Civ). CFTR (endogenous or PIV-delivered), GFP, or PIV antigens were never detected in cell types that did not posses cilia (Figure S3).
To determine whether PIV-mediated CFTR delivery to ciliated cells resulted in functional CFTR anion channel activity in CF HAE, we maximally stimulated cAMP-mediated anion transport capacity using forskolin (Fsk), an activator of CFTR. Figure 3D shows bioelectric short-circuit current (Isc) traces obtained in Ussing chamber experiments with CF HAE inoculated with mock (vehicle alone), PIVGFP, or PIVCFTR. For comparison, a Isc trace from a non-CF HAE is also shown. Whereas Isc responses to Fsk were not observed in mock- or PIVGFP-inoculated CF HAE, CF HAE inoculated with PIVCFTR exhibited rapid and sustained increases in Isc that were rapidly inhibited by a CFTR-specific inhibitor (CFTR172 [20]). Experiments using CF cells derived from four different donors revealed that the kinetics and magnitudes of the Fsk responses in PIVCFTR-corrected CF HAE were indistinguishable from those observed for non-CF HAE (Figure 3D and 3E; range 6.7–42.0 µA/cm2 for PIVCFTR, 0–1.5 µA/cm2 for CF HAE controls [mock and PIVGFP] and 7.8–70.3 µA/cm2 for non-CF HAE). An additional control using PIV expressing the nonfunctional CFTR mutant ΔF508CFTR (PIVΔF508) confirmed that functional CFTR was required for bioelectric correction of CF HAE (Figure 3E). These data show that delivery of functional CFTR to CF ciliated cells fully restored maximally stimulated CFTR anion channel activity to normal non-CF levels. PIVCFTR did not significantly affect UTP-mediated Cl− secretion in CF HAE beyond that of PIVGFP (Figure S4).
Although others have shown that overexpression of CFTR was not detrimental to airway epithelial cell integrity in vitro and in vivo [21],[22], we had anticipated that Fsk-stimulated Cl− secretion would exceed that of non-CF HAE since CF ciliated cells significantly overexpressed CFTR, i.e., CF HAE would be “supercorrected.” That CF HAE overexpressing CFTR exhibited identical anion secretion as measured in non-CF HAE with endogenous CFTR levels suggested that the ceiling for anion secretion rates was not solely related to the absolute quantity of CFTR present in ciliated cells. Several explanations appeared plausible to account for this observation.
A first explanation is that transduced CFTR was not trafficked to the apical membrane of CF ciliated cells. As shown in Figure 3C, this was not the case, as immunofluorescent localization revealed clear targeting of transduced CFTR to apical domains of CF ciliated cells. It is possible, however, that not all correctly trafficked CFTR was inserted into the apical membrane in regions that facilitate function. Certainly, our immunolocalization data suggest CFTR is present in subapical membrane structures likely representing recycling endosomes.
A second explanation is that overexpression of CFTR resulted in mislocalization of a fraction of CFTR to basolateral membranes of ciliated cells. This event would be predicted to dampen Fsk-induced CFTR responses as suggested for adenovirus-mediated CFTR delivery to airway epithelia [23]. Although CFTR immunoreactivity was restricted to the apical domains of ciliated cells, even when overexpressed (Figure 3C), we further tested for this possibility by using CFTR172 as a probe to measure CFTR functional activity in apical and/or basolateral compartments of PIVCFTR-corrected CF HAE. Addition of CFTR172 to apical surfaces 15 min before or during Fsk-induced anion secretion resulted in rapid and complete inhibition of secretion (Figure 4A). In contrast, CFTR172 applied to basolateral surfaces before or during Fsk-induced anion secretion had no effect on Fsk-induced anion secretion, suggesting that no significant functional CFTR was present in the basolateral membranes of ciliated cells.
A third explanation is that the Fsk responses are limited by the apical membrane driving force for Cl− secretion. We tested this possibility by comparing the Fsk responses with PIVCFTR-corrected CF HAE (>100-fold increased CFTR) to non-CF HAE (endogenous CFTR levels) with protocols designed to make the electrochemical driving force for Cl− secretion nonlimiting (bathing solutions changed from Cl− replete [Krebs bicarbonate Ringer, KBR] to Cl− deplete [high potassium low chloride, HKLC]). As shown in Figure 4B, Fsk responses in CF HAE overexpressing CFTR and non-CF HAE were similar under conditions of physiological Cl− secretory driving forces (KBR). Importantly, when the apical membrane Cl− secretory driving force was made large and not limiting by lumenal Cl− substitution, Fsk again stimulated similar responses in CF HAE overexpressing CFTR compared to non-CF HAE (Figure 4B, HKLC). These observations strongly suggest that overexpressed levels of CFTR do not produce reduction in Cl− secretory driving forces that offset the additional quantity of CFTR in the apical membrane and hence buffer the Cl− secretion rates. As an additional control, UTP-mediated Cl− secretion (via calcium-activated Cl− channels) was increased >20-fold in HKLC bathing solution compared to KBR (Figure 4C), indicating that the maximal anion secretory capacity of cultures had not been reached.
Collectively, these data suggest that CFTR overexpressed in CF ciliated cells is selectively trafficked to the apical domains of these cells. However, a ceiling of CFTR Cl− secretion is reached that approximates that of endogenous CFTR in non-CF HAE. We speculate that this ceiling reflects the limiting requirement of the number of potential apical membrane docking sites for CFTR and/or the limited amount of accessory/regulatory proteins localized at apical membranes of CF cells required for CFTR function. It is likely that CFTR insertion into the apical membranes of ciliated cells is tightly regulated with recycling and replenishment of CFTR governed by CFTR-rich recycling endosomes. To determine whether limited availability of docking sites or accessory proteins was unique to CF ciliated cells, non-CF HAE were inoculated with PIVCFTR and PIVGFP, and Fsk responses compared (Figure 4D). Overexpression of CFTR in non-CF HAE provided only moderately increased Fsk-mediated Cl− secretion compared to cultures inoculated with PIVGFP. Although modest but significant differences in resistance were measured before/after Fsk treatment in mock-treated non-CF HAE, no significant differences in resistance were measured before/after forskolin treatment after PIVCFTR versus PIVGFP inoculations: (Resistances [Ω.cm2] before/after Fsk: Mock, 550±49/401±0.5; PIVGFP, 452±39/402±0; PIVCFTR, 419±12/401±0.2; n = 5 for each). Collectively, these data suggest replacement of a corrective CFTR gene to CF ciliated cells is the only manipulation required for correction of the CF defect since both CF and non-CF HAE regulate CFTR activity similarly.
The dehydrated airway surface phenotype characteristic of CF results from the inability to induce Cl− secretion and the failure to regulate Na+ absorption to maintain ASL height at approximately 7–10 µm, i.e., physiologic “thin-film” volumes. We investigated whether expression of CFTR in CF ciliated cells corrected both the Cl− secretory and Na+ hyperabsorptive phenotype of CF HAE by measuring responses to specific antagonists on transepithelial potential difference (Vt) with microelectrodes under thin-film conditions when the height of the ASL was at steady state. Measurement of the basal contribution of Cl− transport to Vt by blocking basolateral membrane cellular Cl− entry with bumetanide (10−4 M) showed that 40% of the Vt in non-CF HAE was accounted for by Cl− ion transport (Figure 5A, white bars). In contrast, in CF HAE, there was no detectable bumetanide-sensitive Vt, consistent with the absence of functional CFTR (black bars). However, after PIVCFTR, but not PIVGFP, the contribution of Cl− transport to CF HAE Vt was indistinguishable from that for non-CF HAE (approximately 40%, red bars). These data obtained under thin-film conditions confirm that delivery of CFTR to CF ciliated cells fully corrected the Cl− secretory defect to non-CF levels.
The epithelial Na+ channel (ENaC) is rate-limiting for Na+ absorption and is negatively regulated by CFTR expression [24]–[26]. Since ENaC activity is regulated by mediators present in the ASL (e.g., nucleotides and proteases [27],[28]), and Ussing chamber (“thick-film”) studies result in washing away of these critical ENaC regulatory factors, we determined how CFTR delivery to CF ciliated cells affected the regulated activity of ENaC under thin-film conditions using microelectrodes. The change in Vt, in response to the Na+ channel blocker, benzamil (10−5 M), indicated that Na+ transport accounted for approximately 40% of the total Vt in non-CF HAE (Figure 5A, white bars) and approximately 80% of transport in CF HAE (black bars), consistent with a Na+ hyperabsorptive phenotype for CF airway epithelia. Expression of CFTR, but not GFP, in CF ciliated cells significantly reduced the contribution of CF HAE Na+ transport to levels measured in non-CF HAE (Figure 5A, red bars). These data show that delivery of CFTR to CF ciliated cells restored both Cl− secretion and the regulation of Na+ absorption by CF airway epithelia to normal, non-CF HAE levels. The simplest conclusion drawn from these data is that ENaC and CFTR both reside in ciliated cells in the human airway epithelium.
During these experiments, it was noted that the lumenal surfaces of PIVCFTR-corrected CF HAE appeared hydrated compared to the dehydrated surfaces of CF HAE, suggesting that the rebalancing of Na+ absorption and Cl− secretion consequent to delivery of CFTR to ciliated cells restored hydration to the lumenal surfaces of CF HAE. Therefore, we initiated experiments to measure ASL height regulation in CF HAE in the absence or presence of transduced CFTR in ciliated CF cells.
ASL volume regulation was assessed by measuring ASL height with XZ-plane confocal microscopy 48 h after addition of 25 µl of PBS to the apical surfaces of CF HAE. In control CF HAE (inoculated with mock or PIVGFP), CF epithelia absorbed almost all fluid from their surfaces, resulting in an ASL height of 3 µm, i.e., the minimal space of compacted folded-over cilia and consistent with mucostasis (Figure 5B) [29]. However, in PIVCFTR-corrected CF HAE, ASL height stabilized at approximately 8 µm, a height similar to that of non-CF HAE (Figure 5B). These data show that delivery of CFTR to CF ciliated cells fully restored the regulation of ASL height to levels maintained by non-CF HAE, thus establishing the critical role of CFTR and ciliated cells in ASL height homeostasis.
We next investigated whether the depletion of ASL over time impaired cilia beat by measuring ciliary beat frequency (CBF) of CF HAE under thin-film conditions. Immediately after addition of 25 µl of PBS to the apical surface of CF HAE, CBF was approximately 8 Hz, a value not different than exhibited by non-CF HAE (Figure 5C). However, consistent with the decreased ASL height in control CF HAE, effective CBF was reduced 48 h later in CF HAE compared to non-CF HAE (Figure 5C). In contrast, effective CBF was maintained in PIVCFTR-inoculated, but not PIVGFP-inoculated, CF HAE at levels similar to those measured in non-CF HAE (Figure 5C). These data strongly argue that ineffective cilia beat observed in CF HAE reflects a defect in ASL height regulation, not ciliary function, and restoration of ASL height regulation with PIVCFTR is sufficient to restore effective cilia beat in CF HAE.
A novel feature of the HAE model is the recapitulation of MCT, reflecting coordinated cilia beat that produces rotational flow of mucus over hydrated epithelial apical surfaces [6]. In CF HAE, the rotational flow is abolished due to ASL depletion, mimicking mucostasis described for CF airways in vivo [6]. To determine whether expression of CFTR in CF ciliated cells could prevent mucostasis, CF HAE were inoculated with PIVCFTR, PIVGFP, or mock and, 24 h later, a small bolus of 1-µm fluorescent beads added to the apical surfaces and cultures maintained at >95% humidity for 24 h. For control CF HAE, rotational flow of beads was rarely observed at 48 h pi (Figure 5D, 5Ei, and 5Eii), i.e., mucostasis occurred. In contrast, PIVCFTR-corrected CF HAE exhibited significant rotational flow of beads, indicating that MCT had been restored (Figure 5D and 5Eiii). Under these conditions at 48 h pi, PIVCFTR restored approximately 50% of the MCT measured in parallel non-CF HAE (Figure 5D and 5Eiv). These data are the first demonstration, to our knowledge, that MCT can be restored to CF airway epithelia by delivering CFTR to ciliated cells, indicating that the cumulative effects of CFTR deficiency on mechanical innate defense can be reversed by this strategy.
Why MCT was not completely restored to normal levels, especially when ion transport processes and ASL volume regulation were fully corrected, remains to be determined. A possible explanation may relate to potentially subtle cytopathic effects of PIV infection at 48 h pi. To address this possibility, we inoculated non-CF HAE with PIVGFP or PIVCFTR, and assessed MCT 48 h later. PIVGFP decreased MCT to 40±3% (n = 5) of normal levels, whereas PIVCFTR decreased MCT to approximately 74±11% (n = 5) of normal levels. These data strongly suggest that after 48 h of PIV infection, virus-induced cytopathic events, likely linked to virus replication capacity, limit complete restoration of MCT to non-CF levels.
Duration of CFTR correction is limited by shedding of PIV-infected ciliated cells. Experiments to determine the duration of PIVCFTR-mediated bioelectric correction showed that significant functional correction was maintained for at least 1 wk with detectable, but decreased, levels of function remaining at 21 d pi (Figure 6A). Previously, we have shown that PIV-infected ciliated cells are shed from HAE 3–7 d pi by a poorly understood process of extruding infected ciliated cells from the epithelium onto the lumenal surfaces of HAE [17]. This process likely represents an innate defense function of the epithelium to rid itself of PIV-infected ciliated cells. To confirm that PIV-infected ciliated cells were being shed from the epithelium, we assessed the cellular composition of apical secretions 6 d after PIVGFP inoculation. By morphologic and immunodetection of GFP-positive cells and ciliated cells, we determined that PIV-infected ciliated cells were shed into apical secretions at a rate far exceeding that of natural ciliated cell shedding (Figure 6B). Therefore, the temporally related loss of CFTR functional activity in PIVCFTR-corrected CF HAE likely reflects shedding of infected ciliated cells and suggests that the extent of correction is directly related to the numbers of ciliated cells expressing CFTR. To further explore this relationship, we counted the numbers of PIV-positive cells present in CF HAE over time and show that the loss of PIV-positive ciliated cells paralleled the loss of Cl− transport (compare Figure 6C to 6A), suggesting that the magnitude of correction was indeed directly proportional to the number of CF ciliated cells expressing CFTR.
Interestingly, although the numbers of PIV-positive ciliated cells were similar at day 2 and 4 pi for both transgenes (CFTR or GFP), by day 8 pi, significantly more PIVCFTR-positive ciliated cells remained compared to PIVGFP. Because PIVCFTR has a lower replication capacity than PIVGFP, we speculate that ciliated cell shedding is related to the rate of virus replication and that identification or generation of PIV vectors with further attenuated replication may provide delivery vectors that prolong functional CFTR correction. Further characterization of the processes involved in PIV-induced ciliated cell shedding from HAE may also provide novel strategies to prolong the lifespan of CFTR-expressing ciliated cells.
A central question in CF gene transfer studies has been the efficiency of CFTR delivery required for clinical benefit. For chemical corrector therapies, efficiency reflects the percent increase in CFTR function per cell. With respect to gene transfer studies, when CFTR expression exceeds endogenous levels on a per cell basis, it may be speculated that efficiency reflects the number of cells within the epithelium targeted for CFTR delivery. Given the low endogenous level of CFTR expression in ciliated cells, coupled with high levels of exogenous CFTR expression generated from PIV vectors, we performed dose-effect experiments with PIV designed to ask what percentage of CF surface epithelial cells must be corrected to restore hydration and MCT to the airway surface in CF HAE.
First, we determined that inoculation of CF HAE with different concentrations of PIVCFTR (range 103–106 PFU) resulted in inoculum-dependent increases in the percentage of PIV-positive cells at 24 h pi (Figure 7A). Increasing numbers of cells expressing CFTR paralleled increasing CFTR mRNA levels, Fsk-stimulated Cl− secretion, and Amil-sensitive Na+ absorption (Figure 7B, 7C, and 7D).
We next used this approach to determine the percentage of cells required to express CFTR to restore normal ASL homeostasis and MCT to CF HAE under thin-film conditions. ASL height and MCT were measured in CF HAE 24 h after inoculation with PIVCFTR at different concentrations. PIVCFTR resulted in concurrent increases in ASL height and MCT rates that were proportional to the percentages of PIV-positive cells, with a plateau occurring at approximately 40% of PIV-positive cells (Figure 7E and 7F). As controls, PIV expressing GFP or ΔF508CFTR failed to increase ASL height or MCT rates in CF HAE (Figures 7E and 7F). By comparison of these data to ASL height and MCT measurements in non-CF HAE (Figure 7E and 7F, dashed lines), it was noted that ASL height and MCT rates both plateaued at levels similar to those measured in non-CF HAE, suggesting that normal homeostasis had been reached. We speculate that the plateau levels for ASL height and MCT reflect homeostatic feedback signals within ASL under the thin-film conditions, e.g., ATP release rates that regulate ion channel activity.
By comparison of data obtained in corrected CF HAE to that in normal non-CF HAE, we calculate that at least 25% of surface epithelial cells (30% of ciliated cells) required CFTR expression to restore ASL height regulation to non-CF HAE levels. For MCT, approximately 40% of cells (50% of ciliated cells) required CFTR expression to approach MCT rates measured in parallel non-CF HAE. These data identify the efficiency of epithelial cell CFTR delivery to restore defective MCT in CF HAE. Of note is the observation that although PIVCFTR was able to restore MCT to nearly normal levels, MCT, unlike ASL height, was not fully restored. This discrepancy is likely related to the cytotoxic effects of PIV infection on MCT, but not ASL measurements. In these experiments, performed 24 h after inoculation, MCT restoration is much improved over that measured at 48 h pi (Figure 5D), further suggesting that 48 h, but not 24 h, of infection with this prototypic PIV vector had detrimental effects on MCT. These data, taken together with those indicating that overexpression of CFTR on a per cell basis was not detrimental to ion and fluid transport processes, indicate that gene delivery vectors capable of targeting at least 25% of the surface airway epithelial cells will be sufficient to restore ASL height regulation and MCT to levels comparable to those exhibited in non-CF airway epithelia and that precise regulation of CFTR levels at least in ciliated cells is not required.
Successful gene transfer to CF airways in vivo has been principally hampered by a lack of efficacy due to the inefficiency of gene transfer to human airway epithelium that normally exhibits CFTR function [30]–[33]. Although there is clear evidence that both airway surface and submucosal gland epithelia are dysfunctional in CF, presently, the precise airway regions of the CF lung that require CFTR delivery for restoration of normal physiological function and reduction of disease symptoms are not well established. Although dependent on antibodies used, CFTR has been localized to human ciliated cells [3] and the fluid-secreting cells of the submucosal glands [4]. Previously noted physiologic characteristics of ciliated cells also indicate that ciliated cells function to maintain airway surface hydration [34]. Certainly, ciliated cells facilitate effective MCT and airway mucus clearance. Together, these properties of ciliated cells combined with the abundance of ciliated cells throughout the human airways make this cell type a logical, although not exclusive, target for CF lung gene delivery strategies.
In this study, we have shown that PIV-mediated CFTR delivery to ciliated cells is efficient and sufficient for correcting the CF airway epithelium phenotype, i.e., efficient delivery of CFTR to CF human ciliated airway epithelium corrected hallmark characteristics of CF HAE that mimic the initiating events of CF lung disease, i.e., abnormal ASL volume homeostasis and mucostasis (Figures 5 and 7). Abnormal ASL homeostasis in CF airway epithelium is due to dysregulated Na+ and Cl− ion transport [35], both consequent to the absence/dysfunction of CFTR at the apical membrane [24]. Here, we have demonstrated that delivery of CFTR to ciliated cells restores Cl− secretion and reduces the Na+ hyperabsorption characteristic of the CF airway epithelium in vitro and in vivo, providing confirmatory evidence that CFTR functions as both a Cl− channel and regulator of ENaC within ciliated cells. Further, we have demonstrated that correction of the ion channel defects of CF HAE restores the integrated physiology required for ASL regulation, which ultimately restores MCT (shown schematically in Figure 7G).
A critical variable for restoration of CFTR functional activity by PIV is the percentage of cells expressing CFTR. Using CF HAE, we demonstrate that restoration of normal ASL height and MCT required CFTR delivery to approximately 25% and approximately 40% of surface epithelial cells, respectively. We suggest that restoration of ASL height is the most predictive measure for these studies, as PIV clearly exerts cytotoxic effects on MCT, but not ASL measurements. These effects were isolated to MCT, but not ASL, suggesting that virus-mediated cytotoxicity may affect the synchrony of cilia beat, leading to modestly reduced effectiveness of ciliated cells to transport mucus.
Previous studies with gap junction–coupled polarized, but not differentiated, airway epithelial cell lines suggested that approximately 6%–10% of cells required CFTR to correct the Cl− transport defect [36], whereas almost all cells (>90%) required CFTR overexpression to correct ENaC hyperabsorption [21]. Clearly, expression of CFTR in nonciliated airway epithelial cells would be predicted to increase fluid secretion onto the apical surfaces of these cells although these previously published studies did not test this hypothesis. In our studies, we have directly shown that expression of CFTR in 60% of CF ciliated cells fully corrects the ENaC hyperabsorption defect (Figure 5A) and that CFTR expression in approximately 25% of cells (approximately 30% of ciliated cells) corrects ASL volume homeostasis in CF HAE (Figure 7E). With respect to ENaC activity after CFTR delivery, the reasons why our data differ from these previous studies are unclear, but it may be speculated that differentiated human airway epithelium models as used in this current study are more representative and relevant to the required efficiency of CFTR delivery to human airway epithelium in vivo.
Although our data indicate that CFTR delivery to CF ciliated cells is sufficient for restoring MCT to CF HAE, it is likely that delivery of CFTR to other nonciliated surface epithelial cells may provide functional CFTR activity capable of hydrating the airway surface. At present, we are not aware of gene delivery vectors capable of delivering CFTR exclusively to nonciliated cells of HAE to determine whether CFTR expression in nonciliated cells also restores MCT. In our studies, we have combined the requirement of ciliated cells for generation of MCT with CFTR targeting of CF ciliated cells to restore defective MCT. Since ciliated cells are the predominant airway surface epithelial cell type throughout the human conducting airways, vectors targeting at least ciliated cells may achieve the required efficiency of delivery for restoration of MCT. We propose that targeting at least ciliated cells provides efficient and effective CFTR function that is sufficient for restoration of MCT.
Our data using an in vitro model of human airway epithelium predict that CFTR delivery to 25% of CF airway epithelial cells will restore MCT to near normal levels. However, it remains to be determined what proportion of normal MCT rates in vivo would be beneficial to CF patients. Tracheal mucus velocities in young smokers are significantly reduced compared to young nonsmokers (3.4 mm/min versus 10.0 mm/min) but without significant differences in lung function, perhaps suggesting that MCT at rates below “normal” may be sufficient to maintain pulmonary healthy [37]. These in vivo studies measured mucus velocities only in the trachea, whereas CF lung disease likely initiates in the more vulnerable, smaller bronchiolar airway regions. If these regions respond similarly to CFTR delivery, then it is possible that delivering CFTR to fewer than 25% of CF cells may provide sufficient MCT to maintain healthy airways. Further testing of this hypothesis will require appropriate in vivo studies.
Since PIV infects ciliated airway epithelium of hamsters and human and nonhuman primates, but not those of the murine airways, testing our PIV vectors in vivo in appropriate models is difficult. Additionally, it has been recently reported that expression of human or murine CFTR in ciliated cells of CFTR−/− mice failed to correct the nasal epithelium bioelectric defect, although correction was demonstrated in neonatal, but not adult, tracheal epithelium [38]. One explanation for these results, in contrast to our data, is that murine CFTR expressed in murine ciliated cells may not function correctly. Our study highlights the need to test vectors for CFTR delivery in appropriate human models and that such data obtained from CF mouse models require cautious interpretation. The recent generation of a CF pig model [39],[40] may be beneficial for testing such vector systems, but preliminary data using ciliated airway epithelial cultures derived from porcine trachea suggest that this species is also not infected by the human viruses from which our PIV vectors are generated.
Using our prototypic PIV vector, we could not determine the lowest limit of CFTR expression on a per cell basis required for correction since this vector significantly overexpressed CFTR relative to endogenous levels. In this regard, because CFTR is critical for regulation of ASL homeostasis, there has been concern that overexpression of CFTR in CF airways would “supercorrect” Cl− transport and generate excessive fluid secretion. This concern, while reasonable, appears unwarranted based on our observations that ion transport rates and ASL heights in CF HAE after CFTR delivery did not exceed that measured in non-CF HAE (Figures 3, 5, and 7). These data agree with previously published reports in which CFTR was transgenically overexpressed in mouse airway epithelium without deleterious results in terms of cell or organ toxicity [22]. However, in this transgenic study, CFTR was overexpressed in Clara cells and alveolar type II cells of the mouse lung, and so our study represents the first demonstration, to our knowledge, of the functional safety of CFTR overexpression in human ciliated cells. Since CFTR overexpression did not supercorrect ASL regulation, we speculate that normal airway epithelium exhibits multiple apical membrane regulatory mechanisms in addition to CFTR levels that prevent excessive secretion of fluid into the airway lumen, i.e., airway flooding.
Demonstrating that PIV expresses CFTR at levels in excess of those required to restore full function to CF HAE suggests that further attenuation of PIV will be feasible while still providing sufficient CFTR for functional correction. Indeed, the lower replication capacity 10-fold reduction of PIVCFTR compared to PIVGFP, in the context of >100-fold overexpression of CFTR in individual ciliated cells, suggests that further attenuation of PIV will continue to provide sufficient CFTR for correction of the CF MCT defect and possibly further reduce the generation of inflammatory mediators and cytotoxicity associated with our PIV vector prototype. The continued effort to develop vaccines against PIV has generated a wealth of live attenuated recombinant PIV [41] that exhibit attenuated replication. It is interesting to note that PIV3 vaccine candidates have been extensively evaluated after lumenal airway delivery in adults and infants as young as 3 mo [42]; an age of CF patients in which CFTR replacement would be desirable.
The demonstration of efficacious CFTR gene delivery to human ciliated airway epithelium overcomes a major hurdle to gene transfer approaches for CF lung disease. Other strategies to improve gene delivery to the human airways are ongoing and are focused on vector development [43]–[45] and/or manipulation of the host tissue [46],[47]. However, the results so far published have not shown significant improvement in the ability to deliver transgenes to human ciliated airway epithelium. Lentiviral-based vectors pseudotyped with Ebola, influenza virus, baculovirus, Sendai, or SARS-CoV envelope proteins efficiently transduce airway epithelial cells in vitro and murine airways in vivo [48]–[53], suggesting that combining useful envelope glycoproteins with the potential longer duration of gene expression afforded by lentiviruses may provide novel vectors for lung gene transfer strategies. Similar vectors can be envisioned using the glycoproteins of PIV to target lentiviruses to human ciliated airways. However, to date, none of these vector systems have progressed to functional studies after delivery of CFTR to human ciliated airway epithelium, and no demonstration of correction of the CF phenotype (e.g., ASL height or MCT) has been reported.
Collectively, the studies reported here demonstrate the efficiency of CFTR delivery to human CF ciliated airway epithelium that is sufficient to reverse the CF phenotype of ASL dehydration and mucostasis. Our prototypic PIV vector provides a useful tool for manipulating ciliated cell function and for investigating the future potential of delivering functional CFTR to the airways of CF patients.
Recombinant hPIV3 (NC_001796) encoding human CFTR (NM_000492) or ΔF508CFTR cDNA or GFP and CFTR as separate genes was generated from the cDNA antigenome of full-length hPIV3 JS strain and described in detail in the Supplemental Methods (Text S1). After rescue, PIV replicated in HEp2 cells to a titer comparable to the JS wild-type strain (109.1 50% tissue culture infective dose [TCID50]/ml) suggesting that GFP, CFTR, or ΔF508CFTR did not adversely affect the growth capacity of PIV in producer epithelial cell lines. Virus titers generated by CF HAE were determined in duplicate by procedures previously described [17].
Human tracheobronchial tissues were obtained by the University of North Carolina (UNC) CF Center Tissue Culture Core from airways resected from CF (ΔF508/ΔF508 mutation) or non-CF patients undergoing elective surgery under UNC Institutional Review Board–approved protocols. Isolated epithelial cells were obtained and plated at a density of 250,000 cells per well on permeable Transwell-Col supports (T-Col, 12-mm diameter; Corning-Costar [13],[54]). For bioelectric measurements in Ussing chambers, cells were plated on type IV collagen-coated Snapwell supports (Corning-Costar). HAE were generated by provision of an air–liquid interface for 4–6 wk to form well-differentiated, polarized cultures that resemble in vivo pseudostratified ciliated epithelium [13]. Prior to viral inoculation, the apical surfaces of HAE were rinsed three times over 15 min and inoculated with 100 µl of 107 PFU/ml virus stocks for 2 h at 37°C (unless otherwise described). After removal of inoculum, HAE were returned to humidified incubators. Human tracheobronchial tissues with 1 cm2 of epithelial cell surface area were inoculated with 100 µl of PIVGFP (107 PFU/ml) or vehicle control for 2 h at 37°C, and then tissues were washed in medium and returned to the incubator for 24 h in minimal media volume. After fixation in 4% PFA, tissues were paraffin-embedded, and histological sections were prepared. Immunodetection of GFP was performed as described below.
To determine epithelial cell types infected by PIVGFP in vitro, HAE were fixed in 4% PFA, permeabilized with 1% Triton X-100, and ciliated cells identified with mouse primary antibodies (Ab) against acetylated α-tubulin (Zymed Laboratories) and anti-mouse IgG-Texas Red (Jackson ImmunoResearch) using confocal X-Z scanning microscopy. Ex vivo tissues were paraffin-embedded, and histological sections were prepared. GFP signal was enhanced by indirect immunofluorescence with rabbit anti-GFP polyclonal Ab (Ab-Cam) and goat anti-rabbit IgG-fluorescein (Jackson ImmunoResearch). Both in vitro and ex vivo, GFP colocalized to cells that were also positive for acetylated α-tubulin, confirming the targeting of ciliated cells in vitro and ex vivo by PIV.
Immunolocalization of PIV F protein was performed on HAE fixed in 4% PFA and immunostained either en face or on paraffin-embedded histological sections. For en face detection, the apical surfaces of HAE were incubated with PIV3 F-specific monoclonal Ab (clone 216.16), followed by goat anti-mouse IgG-AlexaFluor594 (Invitrogen). Quantitation of percentages of cells expressing PIV F protein was performed as described previously [17] by assessing four different en face fields of the HAE surface with two cultures obtained from each of three patients. For histological sections, HAE were immunostained as previously described [17]. GFP fluorescence was enhanced as described above. To immunolocalize CFTR in ciliated cells, CF and non-CF HAE inoculated with PIVGFPCFTR or PIVGFP were gently scraped with pipette tips and cell suspensions in PBS, immediately pelleted onto glass slides with Cytospin, and then air-dried. Following fixation in 4% PFA, CFTR was detected with anti-human CFTR mouse monoclonal Ab #596 (a gift from Dr. J. Riordan, University of North Carolina at Chapel Hill) and Alexafluor594-conjugated goat anti-mouse antibody. Cell nuclei were counterstained with Hoechst 33342 (Invitrogen). Images were taken with a Leica SP2 Laser Scanning Confocal Microscope, and processed with Adobe Photoshop CS2.
To assess ciliated cell shedding induced by PIV, the apical surfaces of CF HAE inoculated with PIVGFP or mock were washed in 200 µl of PBS for 30 min, harvested, and washes pelleted onto glass slides using a StatSpin Cytofuge2 (Iris Sample Processing) and then air-dried. Slides were then counterstained with Giemsa (Invitrogen) or probed with rabbit anti-GFP polyclonal Ab (Ab-Cam) with goat anti-rabbit IgG-fluorescein (Jackson ImmunoResearch) and Ab against acetylated α-tubulin (Zymed Laboratories) with anti-mouse IgG-Texas Red (Jackson ImmunoResearch). Cell nuclei were counterstained with Hoechst 33342 (Invitrogen). Fluorescent confocal images and DIC were taken with Leica SP2 Laser Scanning Confocal Microscope. Image processing and overlay were done with Adobe Photoshop CS2.
Western blot analyses of CFTR protein was performed on HAE lysed in M-PER buffer (Pierce). Equal amounts of total protein (850 µg) per sample were adjusted to 1 ml volume with lysis buffer and added to 2 µl of anti-CFTR Ab #596, followed by 50 µl of immobilized-protein G agarose bead slurry (Pierce). Proteins were released from beads with sample buffer, separated with a NuPAGE 3%–8% Tris-Acetate Gel (Invitrogen), and transferred to PVDF membranes. The membranes were then incubated with anti-CFTR Ab (#596) followed by goat anti-mouse IgG-HRP (Jackson ImmunoResearch), and CFTR were visualized with SuperSignal West Dura Substrate (Pierce).
Total RNA was isolated HAE after inoculation with either PIVCFTR, PIVGFP, or vehicle alone using acid phenol-guanidine thiocyanate followed by DNase digestion and further purification using the Qiagen RNeasy Mini Kit. RNA from three individual CF HAE per inoculation was pooled and first-strand cDNA was synthesized with oligo(dT) and SuperScript II reverse transcriptase (Invitrogen) to ensure amplification of mRNA and not viral genome RNA. Quantitative PCR was performed using a Roche LightCycler with the Roche FastStart DNA Master SYBR Green I Kit according to the manufacturer's protocols. Using the LightCycler Software version 4.0, levels of CFTR mRNA were normalized to the level of GAPDH.
Apical and basolateral samples were collected 48 h pi by applying 0.2 ml of serum-free medium to apical surfaces and harvested 30 min later. Basolateral samples were harvested from the basolateral medium. Samples were stored at −80°C before cytokine analyses using 28-plex Beadlyte Assays (Upstate) with Luminex technology (see Text S1 for details).
HAE were mounted in Ussing chambers for measurement of transepithelial resistance (Rt), transepithelial potential difference (Vt), and short-circuit current (Isc) as previously described [55]. HAE were bathed in bilateral Krebs Bicarbonate Ringer solution (KBR) gassed with 95% O2, 5% CO2, and maintained at 37°C. Vt was clamped to zero, and pulsed to ±10 mV for 0.5 s every 60 s. The electrometer output was digitized online, and Isc, Rt, and calculated Vt displayed on a video monitor. Drugs (amiloride [10−5 M], forskolin [10−6 M], and UTP [10−4 M]) were added from concentrated stock solutions to either lumenal and/or serosal surfaces (all obtained from Sigma-Aldrich). CFTR172 (10−5 M) was synthesized from a local source according to appropriate standards and used as previously described [20]. CFTR172 was added to the apical or basolateral bath 15 min before or during forskolin-activated CFTR ion transport. For the basolateral studies, CFTR172 did not affect forskolin-activated responses even when left 15 min until addition of CFTR172 to the contralateral surface. For the time course of CFTR functional activity experiments, all CF HAE were inoculated with either PIVGFP or PIVCFTR at day 0, and on specific days, cultures were mounted in Ussing chambers for analyses.
For microelectrode measurements of Vt in thin films of ASL, borosilicate glass microelectrodes (World Precision Instruments) were filled with 3 M KCl and positioned into the ASL by a motorized micromanipulator (MC1000e; SD Instruments) connected to a high-impedance electrometer (World Precision Instruments). A macroelectrode, constructed of polyethylene tubing containing 3 M KCl/4% agar, was placed in the serosal bath as the ground. To measure the contribution of basal Cl− and Na+ transport to Vt, bumetanide (10−4 M) and benzamil (10−5 M), respectively, were added to the basolateral bath 10 min prior to recording, as previously described [35]. To avoid evaporation of the thin ASL layer in low-humidity environments, 100 µl of immiscible perfluorocarbon (Fluorinert-77; 3 M Corporation) was added to the airway surface as previously described [56].
To visualize the ASL height, 25 µl of PBS containing 0.2% vol/vol Texas Red-dextran (10 kDa; Invitrogen) was added to the lumenal surfaces of HAE. This volume of PBS results in an initial ASL height of approximately 20–30 µm, as previously described [56]. Images of the Texas Red–labeled ASL are acquired by laser-scanning confocal microscopy (Zeiss Model 510) using the appropriate filters, 540 nm excitation/630 nm emission for Texas Red. Perfluorocarbon was added to the airway surface 10 min after the addition of the dye to avoid evaporation of ASL as described above. ASL height was determined by averaging the height obtained from XZ scans of five predetermined points per HAE over time [56].
HAE were rinsed three times with PBS, then placed on an inverted phase contrast microscope (TE 2000; Nikon) to record cilial movement with a 20× objective. High-speed (125 Hz) video images were captured with an eight-bit b/w camera (GS-310 Turbo; Megaplus). The analog signal was digitized via an analog-to-digital converter board (A/D; National Instruments). A digital computerized CBF analysis system was used to analyze the acquired video images, using specialized software, based on Sisson-Ammons Video Analysis [57].
HAE were removed from a well-humidified incubator and washed three times with PBS, and green fluorescent microspheres (0.02% vol/vol, 1 µm; Invitrogen) were added to apical surfaces in 20 µl of PBS and then HAE immediately returned to the incubator. The rate of microsphere displacement was measured from time-lapse fluorescent images (488 nm excitation/530 nm emission) acquired for 3 s with an inverted epifluorescence microscope (Eclipse; Nikon) and a charge-coupled device (CCD) camera (OrcaER; Hamamatsu). Angular bead transport velocity was calculated as previously described [6].
All data are expressed as means±standard error of the mean (SEM) and followed normal distribution as assessed by a standard Normality Test (Kolmogorov-Smirnov). Unpaired Student t-test was used to assess the difference between groups. One-way ANOVA was performed when more than two groups were compared with a single control, and then the differences between individual groups within the set assessed by a multiple-comparison test (Tukey) when the F was <0.05. A p-value of <0.05 was considered significant.
|
10.1371/journal.pntd.0006492 | Molecular characterization of Vibrio cholerae responsible for cholera epidemics in Uganda by PCR, MLVA and WGS | For almost 50 years sub-Saharan Africa, including Uganda, has experienced several outbreaks due to Vibrio cholerae. Our aim was to determine the genetic relatedness and spread of strains responsible for cholera outbreaks in Uganda.
Sixty-three V. cholerae isolates collected from outbreaks in Uganda between 2014 and 2016 were tested using multiplex polymerase chain reaction (PCR), multi-locus variable number of tandem repeat analysis (MLVA) and whole genome sequencing (WGS). Three closely related MLVA clonal complexes (CC) were identified: CC1, 32% (20/63); CC2, 40% (25/63) and CC3, 28% (18/63). Each CC contained isolates from a different WGS clade. These clades were contained in the third wave of the 7th cholera pandemic strain, two clades were contained in the transmission event (T)10 lineage and other in T13. Analysing the dates and genetic relatedness revealed that V. cholerae genetic lineages spread between districts within Uganda and across national borders.
The V. cholerae strains showed local and regional transmission within Uganda and the East African region. To prevent, control and eliminate cholera, these countries should implement strong cross-border collaboration and regional coordination of preventive activities.
| Cholera, an acute diarrheal disease, essentially was eliminated in the western world many decades ago, but has continued to cause many deaths in sub-Saharan Africa, South America and Asia. Cholera diagnosis in most countries in sub-Saharan Africa, including Uganda, is by stool culture, serology and biochemical methods. These testing methods are unable to establish the relatedness, virulence and spread of Vibrio cholerae in region. To determine the spread, relatedness and virulence of V. cholerae responsible for the various cholera outbreaks in Uganda, we used DNA-based testing methods. We tested 63 V. cholerae isolates from samples collected in Uganda from 2014–2016. Our results showed three distinct lineages of genetically related cholera-causing bacteria. These organisms showed internal spread in Uganda and cross-border spread to neighboring countries in East Africa. These findings provide a valuable baseline and help define the context for directing control measures and technologies for cholera prevention in East Africa.
| Vibrio cholerae remains a major cause of morbidity and mortality globally [1]. There have been seven cholera pandemics since the disease was recognized as a global threat [2]. The English record of pandemics of cholera started in 1816, but cholera as a disease goes back centuries in Indian literature [3]. The organism responsible for cholera outbreaks, V. cholerae, was cultured over 130 years ago by Robert Koch (1884) in India [4] and its epidemiology in England was described by John Snow in 1886 [5].
Over time, considerable knowledge and skills in the management of this deadly infectious disease have accumulated leading to better prevention and control of epidemics [6–8]. Industrialized countries essentially have eliminated cholera as a public health problem through improved water and sanitation [9]. Nonetheless, this enteric bacterium continues to cause deaths and suffering in many countries [10–12]. Sub-Saharan Africa bears the highest reported cholera disease burden [13]. The ongoing outbreaks in Africa and elsewhere in the world are part of the seventh pandemic caused by the V. cholerae O1, El Tor lineage [14,15]. Genetic differences among isolates allow for a greater understanding of the transmission of the bacteria within and between geographic regions and time periods [16].
Two methods, multilocus variable-number tandem-repeat analysis (MLVA) [17,18] and whole genome sequencing (WGS) [19], provide sufficient genetic differentiation to distinguish between the isolates across different places and times. Less complex methods such as culture, biochemical and serological tests to detect, confirm and describe V. cholerae [20], do not permit accurate tracking of the spread of specific genetic lineages. Yet these are the only methods available in most African countries including Uganda [21]. The goal of this study was to analyze V. cholerae isolates responsible for cholera outbreaks that occurred between 2014–2016 in Uganda using multiplex PCR, MLVA and WGS to determine the genetic relatedness and spread of V. cholerae isolates from different outbreaks in Uganda.
A cross-sectional study was conducted using all available viable V. cholerae isolates collected during cholera outbreaks in Uganda between 2014 and 2016 and kept frozen (-80°C) at the Central Public Health Laboratory (CPHL) in Kampala. In addition, aggregated epidemiological cholera surveillance data for the years 2014–2016 were reviewed and used to generate Epi-maps that contextualized the epidemic spread and transmission of cholera.
A total of 63 V. cholerae isolates for the years 2014–2016 were tested. The isolates were from 9 locations: 8 districts in Uganda and a ninth from patients who acquired their illness in Juba, South Sudan, and were treated in Uganda. All 63 isolates tested positive for ompW, toxR and ctxA indicating the presence of V. cholerae virulence genes. The isolates included both V. cholerae Inaba (63%) and Ogawa (34%) serotypes as shown in Table 1.
All 63 V. cholerae isolates were genotyped using MLVA. Three clonal complexes (CC) were identified circulating in Uganda. MLVA CC1 contained 32% (20/63); MLVA CC2, 40% (25/63); and MLVA CC3, 28% (18/63) of the isolates. The three MLVA CCs are shown in Fig 1.
The spatial distribution of MLVA CCs in Uganda reveals the presence of multiple genetic lineages within outbreaks and genetically defined connections between outbreaks (Fig 2). Two lineages were observed in 2014, when CCs 1 & 3 were isolated in Arua and Moyo districts in northwest Uganda. In 2015, CCs 1 & 3 were observed in Hoima and CCs 1 & 2 were isolated in Kasese district in southwest Uganda.
Each separate CC identified one of three genetically related series of outbreaks. First, isolates from CC3 were observed in June 2015 in individuals from Juba, South Sudan, and later in July 2015 in nearby Arua district, Uganda. Additional isolates were seen further south in September 2015 in Hoima on Lake Albert in Uganda. A second outbreak, defined by CC2, was initially identified in April 2015 in Kasese district in western Uganda, and subsequently in November 2015 in Wakiso district in central Uganda, in December 2015 in Kampala district in central Uganda and in December 2015 in Moroto district in northeastern Uganda. This outbreak persisted into January 2016 when it was found in Kampala and Mityana in central Uganda and in Mbale district in eastern Uganda. A third outbreak, defined by CC1, contained isolates collected in May and July 2015 in Kasese district, Uganda, and in June 2015 in individuals from Juba, South Sudan.
WGS genotyping of ten isolates indicated that the DNA was typical of the third wave of the seventh pandemic containing the classical allele of ctxA (S2 Table). The Ugandan DNA sequences belonged to three distinct clades. Within these distinct clades, the Ugandan sequences differed by five or fewer nucleotides (Fig 3). Two clades were contained in the transmission event (T)10 lineage and the other was contained in T13; no Ugandan isolate sequences were contained in a third African lineage T12.
The Ugandan clades were closely related to each other and to sequences from Democratic Republic of Congo and Tanzania (Fig 3). Clade 2 sequences from Kasese district in April 2015 were related most closely to sequences from Mbale district in January 2016 and secondarily to sequences from i) the Democratic Republic of Congo and ii) epidemic isolates from Dar es Salaam, Tanzania in August 2015 which spread across Tanzania during 2015. Clade 3 sequences from Arua and Moyo districts, Uganda in April and May 2014 and Clade 1 sequences from Kasese district, Uganda in April and May 2015 were related closely to sequences from an outbreak in January 2015 in Kigoma, Tanzania. The distance between the Ugandan and Tanzanian clades was nine or fewer nucleotides.
Our data are consistent with the spread of multiple genetic lineages of V. cholerae within Uganda and across its borders during 2014, 2015 and 2016. We found three CCs identified by MLVA that corresponded to the three clades of sequences by WGS. Each of these three genetic lineages displayed cross-border spread and spread within Uganda. The cross-border spread was both into and out of Uganda. These three clades circulating in East Africa belong to wave 3 of the seventh cholera pandemic, ctx carrying V. cholerae El Tor strain and belong to the T10 and T13 introductions of V. cholerae into East Africa [29].
Our data do not change the fundamental topology of the phylogenetic tree for V. cholerae. However, our WGS data revealed incidences of cross-border spread and of spread within Uganda. One example of cross-border spread was demonstrated by the close relationship between isolates (CC1, Clade 1, T10) from i) the Democratic Republic of Congo in 2014, ii) an outbreak in January 2015 in Kigoma, Tanzania, on the shores of Lake Tanganyika, iii) isolates from an outbreak in April and May 2015 in Kasese district on the western border of Uganda about 600 kilometers north of Kigoma, and iv) extended based on MLVA data to include the travelers seeking medical care in Uganda from, Juba, South Sudan. Cross-border spread between the Democratic Republic of Congo, South Sudan and Uganda was previously inferred from epidemiological evidence alone [30,31]. A second cross-border spread was revealed by the close relationship between isolates from an outbreak (CC2, Clade 2, T13) in April 2015 in Kasese district and those from Dar es Salaam, Tanzania, in August 2015 [32]. This lineage also spread from Kasese district to Mbale district in January 2016 or perhaps the seeding of these early 2016 cases came from Tanzania. The genetic distances between the various isolates was too small for the origin to be determined with certainty. Although these two incidences of cross-border spread included isolates from Kasese district in April 2015, the isolates that spread were from two distinct genetic lineages. This finding implies that the two distinct genetic lineages were present at the same time in the cholera outbreak in Kasese district similar to the cholera outbreak in Kenya in January 2009 –May 2010 in which two distinct lineages were also found [33]. A third example of cross border spread comes from MLVA CC3 (Clade 3, T10), the genetically related isolates included isolates from Kigoma, Tanzania and Kasese district, Uganda in January and April 2015. Additional isolates were collected in June 2015 among the fishing community in Hoima district on Lake Albert, Uganda indicating spread within Uganda. A fourth example of cross-border spread comes from the presence of South Sudanese refugees in Uganda in the last half of 2016 seeking health care for cholera, although no isolates were available for testing.
Examples of spread within Uganda included CC3 that was found in April and May 2014 in Arua and Moyo districts respectively, 125 kilometers apart, in northwest Uganda; and was found in July 2015 again in Arua district and in September 2015 in Hoima district, 250 kilometers to the southwest. A second example of spread within Uganda is CC2, initially identified in Kasese district in April 2015 and identified subsequently in December 2015 in Kampala and Moroto districts, in central and eastern Uganda respectively, although the latter could have come from Tanzania, as the genetic data are insufficient to distinguish between the two alternatives.
Tracking the spread of V. cholerae requires genetic identification as demonstrated by the presence of multiple genetic lineages occurring simultaneously in the same region. Multiple lineages were collected in Moyo, Kasese and Hoima districts in Uganda. Multiple lineages were found despite our analyses being limited to a small number of isolates.
Analyses of additional isolates may identify even more cases of multiple lineages in a single location. Each genetic lineage in a given location probably represents an independent introduction event to that location. The caveat to that hypothesis are the reports of multiple lineages within a single person [18], a phenomenon that has not been explored in Africa.
The spread of cholera inferred by this study is consistent with the documented movement of populations including refugees and traders affecting communities located along the great lakes, rivers, fishing villages, and trade and communication routes [30,34]. This is supported by evidence from the 2016 cholera outbreak in northern Uganda that was confined to districts hosting refugees from or bordering South Sudan.
These findings have several implications for cholera control in the region. Apart from providing a baseline for future molecular studies in Uganda, they demonstrate the need for approaches to disease prevention and control that cross national boundaries. In addition to strengthening interventions within countries, an approach similar to that taken to contain Ebola in West Africa [35,36] should be adopted. An outbreak in one country should elicit support from neighbors to ensure timely control [37]. Cross-border collaboration and joint interventions between neighboring countries should be implemented and sustained over an extended period to promote cholera elimination.
No V. cholerae isolates were collected and tested from a cholera outbreak in 2016 in northwestern Uganda that started with the influx of South Sudan refugees and was restricted to districts where the refugees settled and their immediate neighborhoods. However, since this outbreak was restricted to a few districts in northwestern Uganda with refugees, it is unlikely that this had an effect on the findings of this study.
The cholera outbreaks in Uganda were due to genetically diverse V. cholerae O1 isolates from two introductions from wave 3 of the seventh pandemic carrying the classical El Tor toxin gene. The V. cholerae strains showed local and regional transmission within Uganda and East Africa. Interventions to prevent, control, and eliminate cholera in Uganda and throughout East Africa should be strengthened with a focus on regional collaboration.
|
10.1371/journal.pbio.1001911 | Structure of a Membrane-Embedded Prenyltransferase Homologous to UBIAD1 | Membrane-embedded prenyltransferases from the UbiA family catalyze the Mg2+-dependent transfer of a hydrophobic polyprenyl chain onto a variety of acceptor molecules and are involved in the synthesis of molecules that mediate electron transport, including Vitamin K and Coenzyme Q. In humans, missense mutations to the protein UbiA prenyltransferase domain-containing 1 (UBIAD1) are responsible for Schnyder crystalline corneal dystrophy, which is a genetic disease that causes blindness. Mechanistic understanding of this family of enzymes has been hampered by a lack of three-dimensional structures. We have solved structures of a UBIAD1 homolog from Archaeoglobus fulgidus, AfUbiA, in an unliganded form and bound to Mg2+ and two different isoprenyl diphosphates. Functional assays on MenA, a UbiA family member from E. coli, verified the importance of residues involved in Mg2+ and substrate binding. The structural and functional studies led us to propose a mechanism for the prenyl transfer reaction. Disease-causing mutations in UBIAD1 are clustered around the active site in AfUbiA, suggesting the mechanism of catalysis is conserved between the two homologs.
| The biosynthesis of Vitamin K and Coenzyme Q requires the transfer of a long, hydrophobic moiety known as an isoprenyl onto an aromatic acceptor compound. This process is catalyzed by a family of proteins known as the UbiA proteins, which are embedded in the hydrophobic environment of cell membranes. To understand how the prenyltransfer reaction is carried out, we solved the three-dimensional structure of a member of the UbiA family by X-ray crystallography. This structure reveals how magnesium ions and the prenyl substrate are bound within a sealed amphipathic chamber inside the protein and suggests how the reaction intermediate may be stabilized by the protein and protected from the solvent. Functional studies carried out on another member of the UbiA family, as well as comparison to known disease-causing mutations in the human homolog UBIAD1, demonstrate that the residues involved in this process are conserved across the UbiA family.
| Vitamin K is an essential cofactor required for the posttranslational modification of proteins involved in blood-clotting and normal bone metabolism. One of the major forms of vitamin K in humans, menaquinone-4, is produced by cleaving the phytyl group from dietary phylloquinone to produce menadione, which is then modified with a polyprenyl group donated from geranylgeranyl diphosphate (Figure S1). This latter step is catalyzed by the protein UBIAD1, a member of a family of integral membrane proteins known collectively as UbiA prenyltransferases [1]–[3]. Recently, it has also been proposed that UBIAD1 is responsible for the prenylation of coenzyme Q10 in Golgi membranes [4]. Missense mutations to the UBIAD1 gene are the underlying cause of the genetic disorder Schnyder corneal dystrophy (SCD), which causes accumulation of cholesterol and phospholipids in the cornea of the eye, eventually leading to blindness [5].
Membrane-embedded prenyltransferases belonging to the UbiA family are found in every branch of life, and are involved in the biosynthesis of a highly diverse range of molecules, including respiratory lipoquinones such as ubiquinone and menaquinone [6]–[8], prenylated hemes and chlorophylls [9],[10], archaeal lipids [11], numerous prenylated plant flavonoids [12], the antibiotic aurachin [13], Vitamin E [14],[15], and bacterial cell wall precursors [16]. Although the nature of the prenyl acceptor and donor vary considerably, reactions catalyzed by UbiA homologs are Mg2+-dependent and generate pyrophosphate as a leaving group. A representative reaction, catalyzed by the eponymous E. coli protein UbiA, involves the cleavage of the C–O bond in polyprenyl diphosphates of variable length and transfer of the prenyl chain to the ortho position of the phenol 4-hydroxybenzoic acid (4HB; Figure 1A). UbiA family members are typically predicted to contain eight or nine transmembrane helices and have two characteristic conserved motifs with the consensus sequences NDXXDXXXD and DXXXD (Figure 1B), often referred to as the first and second aspartate-rich motifs, respectively.
To understand the structural basis of UbiA function, we set out to elucidate the structure of a member of the UbiA family using X-ray crystallography. The resulting structures of an archaeal homolog reveal locations of Mg2+ and polyprenyl diphosphate binding sites, a possible hydrophobic substrate tunnel allowing the protein to accommodate polyprenyls of variable length, and the location of a cluster of highly conserved residues forming a potential catalytic site.
After screening a large number of bacterial and archaeal UbiA proteins for suitability for crystallization, a homolog from the extremophile Archaeoglobus fulgidus (AfUbiA) was chosen for further study based on its stability in detergent (Figure S2A). Crystals of the selenomethionine-substituted protein diffracted to 3.2 Å, and the structure was solved by single-wavelength anomalous dispersion (Table S1). Despite the modest resolution, assignment of the sequence register was greatly facilitated by the high quality of the experimentally phased electron density maps, and by the locations of five selenomethionine residues (Figure S2B,C). The selenomethionine structure was then used as a molecular replacement search model for 2.4 and 2.5 Å datasets collected on native AfUbiA crystals, grown in lipidic cubic phase (LCP) and soaked with either geranyl diphosphate (GPP) or dimethylallyl diphosphate (DMAPP) prior to freezing (Table S1, Figure S2D). The final models for the two substrate-bound structures contain four AfUbiA molecules in the asymmetric unit, with one molecule of GPP or DMAPP and two Mg2+ per protein chain. Comparison of the detergent and LCP crystal forms shows that none of the interaction surfaces between neighboring protomers are conserved between the different lattices, and so the protein is likely a monomer in the membrane (Figure S2E).
The AfUbiA structure contains a total of nine transmembrane helices, with the N and C termini emerging on opposite sides of the membrane (Figure 1C). Based on the distribution of positive and negative charges in the soluble loops, the N terminus of the protein is probably oriented towards the cytoplasm and the C terminus to the extracellular side [17]. This assignment is also consistent with the experimentally determined orientation of E. coli UbiA [18]. The first eight helices can be grouped into two bundles of four helices each (Figure 1D). The loops connecting the transmembrane helices are short, with the exception of the cytoplasmic loops connecting TM2 and 3 (L2–3) and TM6 and 7 (L6–7), which are both over 25 and 18 residues in length, respectively, and contain short helical regions. The two conserved aspartate-rich motifs are both positioned on the cytoplasmic side of the protein, between the C-terminal ends of TM2 and TM6 and the L2–3 and L6–7 loops (Figure 1E). Interestingly, the conserved motifs as well as the large cytoplasmic loops that follow them are at equivalent positions in the two four-helix bundles. Closer examination of the two bundles reveals that they are structurally homologous and can be superposed by a twofold pseudosymmetry axis running through the center of the protein perpendicular to the bilayer (Figure 1D, Figure S3). This raises the possibility that the UbiA fold may have arisen from the duplication of an ancient four-helix, dimeric protein.
The crystal structure of another archaeal UbiA homolog from Aeropyrum pernix (ApUbiA) was recently reported with a resolution of 3.6 Å [19]. The overall fold of AfUbiA is similar to that seen in ApUbiA, but there are differences in the location and coordination of Mg2+ and substrate (detailed in the section “Differences between the AfUbiA and ApUbiA crystal structures”). Both UbiA family members resemble proteins belonging to the isoprenoid synthase superfamily [20], and in particular, those members that catalyze the synthesis of all-trans polyprenyls by repeated addition of isopentenyl pyrophosphate (IPP; Figure S1), the trans-IPPSs. These enzymes are also Mg2+-dependent and contain similar aspartate-rich motifs. Although any sequence identity between the soluble and transmembrane families is negligible, comparison of AfUbiA to the trans-IPPS farnesyl diphosphate synthase (FPPS) from E. coli (Figure S4A–B) [21] reveals that the two four-helix bundles comprising helices 1–8 in AfUbiA and 2–9 in FPPS are superposable. TM9 in AfUbiA and the first helix in the trans-IPPS fold have no equivalent in the other family. Nevertheless, despite their structural similarity, examination of the distribution of charged residues on the two enzymes clearly reveals their distinct identities as soluble and transmembrane proteins (Figure S4C–D).
The bound isoprenyl-diphosphates and Mg2+ are located in a large cavity at the interface between the four-helix bundles near the cytoplasmic side, which is partly closed off from the solvent by the L2–3 and L6–7 loops (Figure 2A). Interestingly, in the unliganded structure, a 13-residue long region of the L2–3 loop is disordered, leaving this cavity widely accessible to the solvent, whereas all but four residues become resolved in the GPP-bound structure. In the DMAPP-bound structure, the entire loop is resolved and completely occludes the cavity from the solvent (Figure S5A,B). This difference may be attributed to the lower resolution of the unliganded structure, however, as the substrate in the DMAPP-bound structure is occluded from the cytoplasm, a more likely possibility is that substrate binding induces conformational changes in the L2–3 loop and thus seals off the active site.
Near the cytoplasm, the central cavity is broad and is lined with polar and charged residues, including the aspartate-rich motifs and many of the other residues that are most conserved across the UbiA family (Figure 2B). The cavity becomes more hydrophobic and tapers into a narrow tunnel as it extends deeper into the transmembrane region of the protein. Approximately halfway into the bilayer, the tunnel bends sharply and forms a fenestration in the side of the protein that opens into the bilayer (Figure 2C). This tunnel could offer a possible explanation for how UbiA family members utilize prenyl donors of varying lengths, which range from DMAPP (C5) [22] to dodecaprenyl phosphate (C60) [16]. The latter substrate approaches 60 Å in length in a fully extended conformation; in comparison, the membrane-spanning region of AfUbiA is less than 40 Å. The hydrophobic tunnel in AfUbiA could potentially accommodate up to six prenyl units, and even longer polyprenyls could bind to the protein by extending directly into the hydrophobic core of the bilayer.
In the GPP-bound structure, the substrate is located in the central cavity with its diphosphate positioned between the two aspartate-rich motifs (Figure 3A,B). Two electron densities are also visible on either side of the diphosphate that likely correspond to Mg2+. This observation is consistent with data showing that the activity of UbiA family members is Mg2+-dependent, as well as extensive mutagenesis experiments, confirming the importance of the two motifs for activity in E. coli UbiA [23] and other homologs [13],[24]. The two Mg2+ are coordinated by N68 and D72 in the first aspartate-rich motif and D198 and D202 in the second aspartate-rich motif (Figure 3A). The conserved aspartate D76 is too far away to bind directly to Mg2+ in the first motif, but could interact indirectly by stabilizing a water molecule coordinating the ion. The diphosphate group of GPP is stabilized by Mg2+ in the first motif, which bridges two oxygens with coordination distances of 2.3 and 2.6 Å. In contrast to the first motif, the Mg2+ bound to the second motif is 3.5–4.0 Å away from the diphosphate, which is significantly farther than the expected coordination distance of 2.0–2.3 Å. Additional interactions between the protein and GPP oxygens are provided by the basic residues R22 and K146. The GPP molecule is slightly kinked after the diphosphate, so that the isoprenyl tail extends along the wall of the cavity close to highly conserved residues on TM2, 4, and 5. In particular, the C–O bond cleaved in the prenyltransfer reaction is positioned near a cluster of conserved polar residues including N68, Y139, and S140 (Figure 3B). The geometry of the Mg2+ and substrate binding sites in the DMAPP-bound structure is similar (Figure S5C,D).
We used two approaches to verify the interactions between the bound substrate, ions, and protein in our crystal structure. First, to confirm the Mg2+ binding sites, we co-crystallized the protein with Cd2+ (Table S1). Two strong electron densities consistent with Cd2+ appear coordinated by the aspartate-rich motifs, which align well to the Mg2+ locations in the GPP-bound structure (Figure S6). Second, we used isothermal titration calorimetry (ITC) to measure the Mg2+ dependence of GPP binding to AfUbiA. In the presence of 2 mM MgCl2, GPP binds to AfUbiA with a KD of 3.2±0.1 µM (Figure 3C). However, when 1 mM EDTA was added instead of MgCl2, no GPP binding was observed (Figure 3D). Residues N68, D72, D198, and D202 from the two aspartate-rich motifs, as well as the basic residues R22 and K146, were then mutated to alanine to test their contribution to Mg2+-dependent GPP binding (Figure 3E, Figure S7). Mutations to all six residues had pronounced effects on binding of GPP to AfUbiA. Four of the mutations completely abolished GPP binding, whereas the effects of the D198A and D202A mutations were comparatively mild, increasing the KD by 45- and 21-fold, respectively. This is consistent with the observation that the distance between the Mg2+ bound to the second aspartate-rich motif and the GPP diphosphate is outside of the typical coordination distance range.
There are notable differences in the shape of the substrate-binding cavity and in the organization of the active sites of the AfUbiA and ApUbiA structures. The central cavity in ApUbiA is smaller than in AfUbiA, largely because ApUbiA lacks the long hydrophobic tunnel and second opening observed in AfUbiA. For the ApUbiA structure, it was proposed that longer prenyl chains may extend out of the protein via its single entrance to the central cavity, which is closer to the membrane interface than in AfUbiA. This mechanism of accommodating long prenyl chains is not likely for the current AfUbiA structure, because although the cytoplasmic opening exists in unliganded AfUbiA, it is completely closed in the DMAPP-bound structure (Figure S5B).
In both structures, the bound Mg2+ and the diphosphate moiety are located in the central cavity between the two conserved aspartate-rich motifs; however, interactions between the protein and ligands are different (Figure 4). Although two bound Mg2+ were modeled in both structures, in the current AfUbiA structure, N68, D72, D198, and D202 directly coordinate the Mg2+, while in the ApUbiA structure all the corresponding residues are >3.4 Å away from Mg2+. These four residues were demonstrated to be important for Mg2+-dependent GPP binding according to our ITC data (Figure 3E). The differences in Mg2+ location and coordination between the two structures are likely attributable to the significantly lower resolution (3.6 Å) of the ApUbiA structure.
4HB was also modeled into the ApUbiA structure, although there is currently no direct biochemical evidence that it can act as a prenyl acceptor for ApUbiA. This proposed 4HB binding site is unlikely to hold the prenyl acceptor in the current AfUbiA structure, as its position clashes with the location of the geranyl moiety. We were also unable to detect binding of 4HB to AfUbiA by ITC.
Although two crystal structures are available now for the UbiA family of proteins, both AfUbiA and ApUbiA are from archaeal thermophiles and enzymatic activity has not been demonstrated for either of the proteins. To understand the relevance of the AfUbiA structure to other UbiA family members, we mutated a number of residues on the E. coli MenA homolog (EcMenA) that are equivalent to key active site residues in AfUbiA (Figure 5A, Figure 5B). EcMenA catalyzes the transfer of a prenyl chain onto 1,4-dihydroxy 2-naphthoic acid (DHNA) to produce the menaquinone precursor demethylmenaquinone. Two independent functional assays were used to measure the effects of mutations on EcMenA: an in vivo genetic complementation assay in which growth under anaerobic conditions was measured in an menA− E. coli strain [25] transformed with WT or mutant EcMenA (Figure 5C), and an in vitro assay measuring prenyltransferase activity with purified membranes from E. coli cells overexpressing WT or mutant EcMenA (Figure 5D, Figure S8). For both assays, mutations to the equivalents of N68 and D72 in the first aspartate-rich motif and D198 and D202 in the second motif resulted in total or near-total loss of function. Mutation of the highly conserved tyrosine (Y139 in AfUbiA) near the C–O bond also resulted in loss of function.
Functional data for eukaryotic UbiA homologs are currently scarce, but missense mutations to 19 different residues on human UBIAD1 are known to cause SCD [5],[26]–[35]. Of these, three align to insertions not present on AfUbiA (Figure 5A). As shown in Figure 5E, the remaining 16 mutated residues all map to the region around the putative active site at the cytoplasmic end of the cavity. The residues Y174 and T175 on human UBIAD1 are equivalent to Y139 and S140, which belong to the cluster of polar residues on TM2 and TM4 likely important for catalysis; the residues A97 and G98 (F63 and S64 on AfUbiA) pack into the interface between TM2 and TM4 near this site (Figure 5F). Residues N102, K181, D236, and D240 on UBIAD1 are homologous to N68, K146, D198, and D202, which form part of the Mg2+/diphosphate binding site (Figure 3A). Residues 112, 118, 119, 121, and 122 on UBIAD1 align to residues 78, 90, 91, 93, and 94 on the highly mobile L2–3 loop of AfUbiA, which changes conformation upon substrate binding in AfUbiA. Potential functions for G177, G186, and L188 (P142, D152, and I154 on AfUbiA) are less evident, but all three residues are located in close proximity to the proposed active site.
Overall, the above experiments on EcMenA and the mapping of UBIAD1 mutations onto the AfUbiA structure suggest that the fold and location of substrate-binding sites are conserved across the UbiA family, and that the mechanism of the prenyltransfer reaction is conserved as well. Although this mechanism is currently unknown, possible clues may be found by comparison to the soluble trans-IPPS proteins. In addition to sharing a fold with the UbiA homologs, the two protein families also exhibit similarities in the architecture of their active sites. The structure of E. coli FPPS bound to IPP and a thio- analog of the prenyl donor, thioDMAPP, is representative of available structures of trans-IPPS ternary complexes (Figure S9) [21]. Like AfUbiA, members of the trans-IPPS family contain two signature acidic motifs, which both contain the conserved sequence DDXXD [36] and which coordinate Mg2+ atoms that stabilize the diphosphate on the prenyl donor. In trans-IPPS proteins, the reaction is believed to proceed via a three-stage ionization–condensation–elimination mechanism [37], involving a carbocation intermediate in the allylic site that is stabilized by the liberated diphosphate as well as interactions with nearby polar side chains [21],[38]. Given the structural and functional similarity between AfUbiA and the trans-IPPS proteins, it is possible that this catalytic mechanism is shared with homologs of the UbiA family (Figure 6).
In the GPP- and DMAPP-bound structures, C-C bond formation would occur near a triad of three polar side chains: N68 from the first aspartate-rich motif, and Y139 and S140 on TM4 (Figure 3B). Interestingly, the tyrosine is the single most highly conserved residue in the UbiA family, present in 97% of the more than 10,000 UbiA sequences currently in the Pfam database [39]. Mutation of this residue in both EcMenA (Figure 5C,D) and EcUbiA [19] results in a loss of function. The near-universal conservation of the tyrosine residue implies that this site is involved in a function that is shared among all the different branches of the UbiA family, regardless of the nature of the highly variable prenyl acceptor. We therefore propose that this site could be involved in stabilizing the carbocation intermediate on the prenyl donor after cleavage of the pyrophosphate leaving group, possibly by cation–π interactions with Y139. A similar role for active site tyrosine residues has been proposed for the structurally unrelated aromatic prenyltransferases DMATS and CloQ [40]–[42].
We have described the structure of an archaeal member of the UbiA family of membrane-embedded prenyltransferases. The substrate-bound structures reveal that the diphosphate group interacts with arginine and lysine side chains as well as Mg2+ coordinated by conserved aspartate-rich motifs. Although short prenyl donors containing only one or two prenyl units were chosen for crystallization due to their higher solubility in water, the structure reveals a long, narrow cavity that opens into the membrane and could allow the protein to bind significantly longer polyprenyl chains. Due in part to its low homology with UbiA family members of known function, we were unable to identify the natural prenyl-accepting substrate of AfUbiA. Nonetheless, the EcMenA functional assays using mutations designed with the AfUbiA structure, as well as the clustering of SCD-causing mutations around conserved residues in the substrate-binding cavity of AfUbiA, suggest that the fold and key aspects of the catalytic mechanism may be conserved between these distantly related homologs and that AfUbiA is a useful structural model for understanding UbiA family prenyltransferases.
One curious feature of the substrate-bound structures is that the Mg2+ bound to the second conserved motif is out of range for coordination of the prenyl donor, and yet the functional data for EcMenA indicate that the residues coordinating this Mg2+ are critical for function. One possible explanation is that when the protein is bound to Mg2+ and the isoprenyl diphosphate only, the substrate is coordinated by only one Mg2+ in order to prevent reaction of the isoprenyl diphosphate with water in the absence of the prenyl acceptor, as interactions with one Mg2+ may not be sufficient to induce spontaneous cleavage of the C–O bond. Binding of the prenyl acceptor would then induce a conformational change to bring the Mg2+ bound to the second conserved motif within coordination distance of the diphosphate and exclude water from the cavity. Another possibility has also been proposed based on homology models of E. coli UbiA [23], that UbiA homologs stabilize the diphosphate via a single Mg2+ bound to only one of the two aspartate-rich motifs and that the other motif activates the phenolic substrate by abstracting a proton. Resolution of the issue will likely require a structure of the full ternary complex.
In the structurally related trans-IPPS FPPS, comparisons of the ternary complex and apo-structure show that the enzyme undergoes a conformational change upon binding substrate; in particular, the loops corresponding to L2–3 and L6–7 in UbiA close over the active site to occlude it from the solvent [21]. The unliganded and substrate-bound AfUbiA structures show similar behavior, in that a region of the L2–3 loop undergoes a disordered to ordered transition upon substrate binding. The prenyl acceptor could therefore enter the cavity through an opening formed by fluctuations in the L2–3 loop, which would then be stabilized in the closed conformation. Although short polyprenyl diphosphates like GPP could also enter the substrate-binding cavity via the opening to the cytoplasm, longer polyprenyl diphosphates have poor solubility in water and likely partition into the lipid bilayer, and therefore may bind to the protein laterally from within the membrane. Because it seems implausible that the negatively charged diphosphate group enters the core of the bilayer and threads into the hydrophobic tunnel, the protein may undergo a conformational change to allow the polyprenyl substrate access to the substrate binding cavity. In the crystal structure, one wall of the substrate tunnel is formed by TM9, whose removal leaves the central cavity completely exposed to the bilayer (Figure 2A). A slight movement of this loosely packed helix could potentially suffice to allow substrates to enter the binding site and allow release of the product as well.
Fifty-two bacterial and archaeal UbiA homologs were cloned and tested for expression [43]. A UbiA gene from Archaeoglobus fulgidus DSM 4304 (GenBank AAB89594.1) was identified as the most promising candidate for crystallization trials. The AfUbiA gene was cloned into a modified pET vector (Novagen) with an N-terminal polyhistidine tag. For large-scale purification of native AfUbiA, the plasmid was transformed into BL21 (DE3) cells. For expression of the native protein, the transformants were grown in Luria broth supplemented with 100 mg/l Kanamycin at 37°C until OD600 reached 1.0 and induced with 0.5 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) at 20°C for 15 h. For expression of selenomethionine-incorporated proteins, the cells were grown in minimal medium containing 32.2 mM K2HPO4, 11.7 mM KH2PO4, 6 mM (NH4)2SO4, 0.68 mM Na Citrate, 0.17 mM Mg2SO4, 32 mM glucose, 0.008% (w/v) alanine, arginine, aspartic acid, asparagine, cysteine, glutamic acid, glycine, histidine, proline, serine, tryptophan, glutamine, tyrosine, 0.02% (w/v) isoleucine, leucine, lysine, phenylalanine, threonine, valine, 25 mg/l L-selenium-methionine, 32 mg/l thiamine, and 32 mg/l thymine, and induced when OD600 reached 0.6.
Cell membranes were solubilized with 40 mM n-decyl-β-D-maltoside (DM, Anatrace), and the His-tagged protein was purified with TALON Metal Affinity Resin (Clontech Inc.). After removal of the N-terminal His-tag with TEV protease, the native protein was subjected to size exclusion chromatography with a Superdex 200 10/300 GL column (GE Health Sciences) pre-equilibrated in a buffer of 150 mM NaCl, 20 mM HEPES, pH 7.5, 5 mM β-mercaptoethanol (βME), and 40 mM n-Octyl-β-D-Glucopyranoside (OG, Affymetrix). The protein was concentrated to 10 mg/ml as approximated by ultraviolet absorbance. The selenomethionine-incorporated protein was purified by the same procedure. AfUbiA for the ITC assay was also purified with the same protocol except that 4 mM of DM was used in place of OG in the size-exclusion chromatography buffer.
Selenomethionine-incorporated AfUbiA crystals were obtained in mother liquor containing 12.5% PEG20000, 100 mM MES buffer, pH 6.7. To obtain LCP crystals, the purified AfUbiA protein was concentrated to around 35 mg/ml as approximated by ultraviolet absorbance at 280 nm and mixed with monoolein (1-oleoyl-rac-glycerol; Sigma Aldrich) at a 2∶3 ratio (protein/lipid, w/w) using the twin-syringe mixing method [44]. The protein/lipid mixture was dispensed manually in 30–50 nl drops onto 96-well glass Laminex plates (Molecular Dimensions) and overlaid with 1.7 µl precipitant solution per drop. Crystals reached full size within 2 wk at 20°C in 34% (w/v) PEG400, 0.1 M Tris-HCl pH 8.2, 0.1 M NaCl, and 0.1 M MgCl2. Before harvest, crystals were soaked in 1 mM GPP or 1 mM DMAPP. The LCP crystals were flash frozen in liquid nitrogen without additional cryoprotectant. Crystals of native AfUbiA bound to Cd2+ were obtained in 30% PEG 550 MME, 100 mM MES buffer, pH 6.6, 5 mM MgCl2, and 100 mM CdCl2. Before flash-freezing in liquid nitrogen, these crystals were cryoprotected in serial mother liquor solutions containing 5%–25% (v/v) glycerol.
X-ray data were collected at beamlines X29 at the National Synchrotron Light Source and 24ID-C and 24ID-E at the Advanced Photon Source. A data set collected on a selenomethionine crystal was processed and scaled with a 3.2 Å cutoff using HKL2000 [45]. Four selenium sites were located with phenix.hyss [46], and phases and a partial polyalanine model were obtained with phenix.autosol [47]. The locations of the selenium atoms and clear side chain densities from aromatic side chains in the experimental maps (Figure S2B,C) were used to manually assign the sequence register, and the structure was refined through iterative rounds of manual model building and automated reciprocal-space refinement using Coot [48] and phenix.refine. The final refined model has R and Rfree values of 25.1% and 28.9%, respectively, and contains residues 15–73 and 86–300 of one UbiA monomer and one molecule of OG, which was used to solubilize the protein. The native GPP-bound and DMAPP-bound structures were solved by molecular replacement using the selenomethionine structure as a search model and refined with phenix.refine using strong NCS restraints that were gradually relaxed over the course of refinement. The final structures each contained four molecules of AfUbiA, 8 Mg2+, and 4 molecules of GPP or DMAPP. The Cd2+-bound structure was solved by a similar protocol. The final Cd2+-bound structure contained two molecules of AfUbiA in the asymmetric unit and eight Cd2+ ions. In the SeMet, GPP- and DMAPP-bound structures, the putative substrate tunnel is partly occupied by a strong, tubular, nonprotein electron density (); however, the resolutions do not allow a definitive identification of this ligand. The GPP-bound, DMAPP-bound, SeMet, and Cd2+-bound structures have been deposited in the PDB under the accession codes 4TQ3, 4TQ4, 4TQ5, and 4TQ6, respectively.
Chain A in the GPP-bound structure had the highest quality 2Fo-Fc density, as well as the lowest average B-factors of the four protein chains in the asymmetric unit, and was therefore used to generate figures and for distance measurements unless otherwise noted. All structure figures were made using PyMol (Schrödinger). Sequence conservation scores in Figure 2B were calculated with the ConSurf server [49], using the seed sequences for the UbiA family from Pfam [39] for the multiple sequence alignment. The alignment of AfUbiA, EcMenA, and human UBIAD1 used for Figure 5A was generated by aligning the sequences to the Hidden Markov Model profile for the UbiA family in Pfam.
The ITC buffer comprised 20 mM HEPES (pH 7.5), 150 mM NaCl, and 4 mM n-decyl-β-D-maltopyranoside (DM). The chamber contained ITC buffer plus 50 µM AfUbiA and either 2 mM MgCl2 or 1 mM EDTA. The syringe contained ITC buffer plus 0.6 mM geranyl pyrophosphate (GPP) and either 2 mM MgCl2 or 1 mM EDTA, whichever matches the chamber condition. The buffer-alone control had no AfUbiA in the chamber. For experiments with mutant proteins, 50 µM AfUbiA mutant proteins were in ITC buffer containing 2 mM MgCl2 with either 0.6 mM GPP (for R22A, N68A, D72A, K146A) or 2 mM GPP (for D198A and D202A) in the syringe. Solutions were filtered and centrifuged at 18,000× g for 5 min prior to the experiments. All binding measurements were performed using a MicroCal iTC200 System (GE Healthcare) at a constant temperature of 25°C. For experiments with apparent binding, thermograms were processed and fit in Origin to a one-site model to obtain n (stoichiometry), K (association constant), and ΔH (enthalpy). The dissociation constant (KD) was calculated from KD = 1/K, and ΔS was calculated from ΔG = ΔH−TΔS. All experiments were performed at least three times.
The menA-deficient E. coli strain AN67 [25], which exhibits a grow defect under anaerobic conditions, was obtained from the Coli Genetic Stock Center and transformed with a pET31 plasmid containing WT and mutant EcMenA genes, or an unrelated protein (the TrkH potassium transporter from Campylobacter jejuni) as a negative control. The transformants were grown aerobically in Luria broth to an optical density of 1.0%±0.05%, supplemented with 20% glycerol, flash frozen, and stored at −80°C. The EcMenA genetic rescue experiments were carried out using a protocol adapted from Suvarna et al. [7]. The 10 mL cultures of a glycerol/trimethylamine N-oxide minimal media [50] containing 0.5 mM IPTG and 0.1 mg/mL ampicillin were inoculated with 10 µl of the glycerol stocks and then incubated in an anaerobic chamber at 37°C. At 24 h postinoculation, OD600 was measured for each culture. Values shown in Figure 5C are averages for three experiments.
Purified E. coli membranes were prepared by harvesting 1 l of cells overexpressing WT and mutant EcMenA, grown to 1 OD as described for AfUbiA. The proteins were expressed as SUMO fusion proteins to increase yield. The cell pellets were resuspended in 20 ml lysis buffer (20 mM Hepes pH 7.5, 150 mM NaCl, 2 mM βME, 5 mM MgCl2, 1 mM PMSF, 25 µg/ml DnaseI). After breaking the cells by sonication, the cell lysates were centrifuged for 30 min at 3,000× g and 4°C. The supernatant was then transferred to clean centrifuge tubes and centrifuged a second time for 60 min at 100,000× g and 4°C. The membrane pellet was resuspended in 3 ml 50 mM Tris pH 7.5, flash frozen, and stored at −80°C. Overexpression of WT and mutant SUMO-EcMenA was verified by running 0.5 µl of the membrane suspension before and after a 30 min digestion with 1 µg SUMO protease on an SDS-PAGE gel (Figure S8C). In addition to the WT and mutant EcMenA proteins, membrane fractions were prepared in the same manner for cells expressing SUMO-EcUbiA, which does not utilize DHNA as a prenyl acceptor, for the negative control.
For the enzymatic assay, 30 µl reaction mixtures were prepared with the following components: 3 µl purified membrane fractions, 2 mM DHNA, 1 mM GPP, 5 mM MgCl2, 5 mM βME, 5% acetonitrile (ACN), and 50 mM Tris pH 7.5. The reaction mixtures were incubated for 10 min at 37°C, quenched with the addition of 2% formic acid, and extracted with 10 volumes of chloroform. The chloroform was dried under air and the residue resuspended in 60 µl 65% ACN/35% 50 mM Tris pH 7.5 in dH2O. The resulting samples were then separated using reverse phase HPLC with a gradient of 65%–75% ACN for the mobile phase. Enzyme activity was quantified as the area of the product peak, normalized by the activity for the WT protein. Values shown in Figure 5D are averages for three experiments.
|
10.1371/journal.ppat.1001136 | Transforming Growth Factor-β: Activation by Neuraminidase and Role in Highly Pathogenic H5N1 Influenza Pathogenesis | Transforming growth factor-beta (TGF-β), a multifunctional cytokine regulating several immunologic processes, is expressed by virtually all cells as a biologically inactive molecule termed latent TGF-β (LTGF-β). We have previously shown that TGF-β activity increases during influenza virus infection in mice and suggested that the neuraminidase (NA) protein mediates this activation. In the current study, we determined the mechanism of activation of LTGF-β by NA from the influenza virus A/Gray Teal/Australia/2/1979 by mobility shift and enzyme inhibition assays. We also investigated whether exogenous TGF-β administered via a replication-deficient adenovirus vector provides protection from H5N1 influenza pathogenesis and whether depletion of TGF-β during virus infection increases morbidity in mice. We found that both the influenza and bacterial NA activate LTGF-β by removing sialic acid motifs from LTGF-β, each NA being specific for the sialic acid linkages cleaved. Further, NA likely activates LTGF-β primarily via its enzymatic activity, but proteases might also play a role in this process. Several influenza A virus subtypes (H1N1, H1N2, H3N2, H5N9, H6N1, and H7N3) except the highly pathogenic H5N1 strains activated LTGF-β in vitro and in vivo. Addition of exogenous TGF-β to H5N1 influenza virus–infected mice delayed mortality and reduced viral titers whereas neutralization of TGF-β during H5N1 and pandemic 2009 H1N1 infection increased morbidity. Together, these data show that microbe-associated NAs can directly activate LTGF-β and that TGF-β plays a pivotal role protecting the host from influenza pathogenesis.
| Transforming growth factor-beta (TGF-β) is a multifunctional protein that serves as a global regulator of immunity by controlling the initiation and resolution of inflammatory responses. A pathogen that can regulate TGF-β activation could promote an immune-privileged state for itself within its host. Indeed, multiple parasitic, bacterial, and fungal pathogens successfully evade immune responses by regulating TGF-β. We demonstrate that the neuraminidase proteins from influenza A viruses and Clostridium perfringens convert biologically inactive TGF-β to its active form. Importantly, modulation of TGF-β activity during influenza infection affects viral titers and disease outcome in mice, suggesting that TGF-β plays an important role in influenza pathogenesis, particularly in protecting the host during infection. These studies suggest that neuraminidases from diverse microbes may be able to directly regulate TGF-β, which may in turn play an important role in disease.
| Transforming growth factor-β1 (TGF-β) is the prototypic member of a family of multifunctional cytokines that modulate diverse cellular, developmental, and immunological processes (reviewed in [1]–[3]). TGF-β is secreted by virtually all cells as a biologically inactive molecule termed latent TGF-β (LTGF-β) [4], [5]. The latent complex consists of an N-terminal latency-associated peptide (LAP) and the mature TGF-β domain. LAP and TGF-β are products of a single gene, which after posttranslational modifications such as glycosylation and phosphorylation and cleavage by furin remain noncovalently associated, forming the small latent complex [6]. The small latent complex is secreted by cells as an inactive complex, and in some cases is linked by a disulfide bond to the latent TGF-β-binding protein to form the large latent complex.
The non-covalent association of LAP with the mature domain is critical for latency. The molecular mechanism by which LAP confers latency to mature TGF-β is largely unknown. However, recent studies suggest that amino acids 50–85, several of which are glycosylated and contain terminal sialic acid residues, are critical for proper formation and function of the LTGF-β complex [7]. Mutations in this region reduce the binding of LAP to the mature domain and significantly impair the ability of LAP to confer latency to mature TGF-β [8]. Agents that activate the latent complex can disrupt the association of LAP with the mature domain either by proteolysis or denaturing the LAP or by altering its folding [6]. Given the abundance of LTGF-β and the prevalence of high-affinity receptors on most cell types, the activation of LTGF-β is recognized as a crucial step in TGF-β function (reviewed in [9], [10]).
Chaotropic agents, heat, reactive oxygen species [11], [12], and extreme pH [13], [14] can activate LTGF-β. In vitro studies have identified proteases, which degrade the LAP (reviewed in [15]), and molecules such as thrombospondin-1, which alter the conformation of the LAP [16], [17], [18], [19], [20], as putative physiological TGF-β activators. Less is known about activation in vivo, although integrins appear to be the primary LTGF-β activators in the lung [10], [21], [22].
Little is known about the direct activation of LTGF-β by microbes. Several parasites such as Trypanosoma cruzi, Leishmania spp., and Plasmodium spp. and the bacteria Mycobacterium tuberculosis activate LTGF-β through proteolysis, using either host-derived plasmin or microbe-encoded proteases [23]–[26]. In a previous study, we have shown that influenza viruses activate TGF-β in vitro and in vivo [27]. Antibodies to the viral neuraminidase (NA) protein inhibited viral-induced LTGF-β activation, suggesting that NA plays a role in LTGF-β activation, but the precise mechanism of activation remains to be identified and the role of TGF-β in influenza disease is unknown. In this study, we determined the mechanism of activation of rLTGF-β by viral and bacterial NA. Since NA is essential for viral replication, we tested a panel of influenza virus subtypes (including two 2009 H1N1 pandemic strains) for their ability to activate LTGF-β in vitro. For strains that failed to activate LTGF-β, we used reverse genetic studies to determine whether these viruses had deficient NA activity. We also investigated whether exogenous TGF-β provides protection from H5N1 influenza pathogenesis and whether depletion of TGF-β during virus infection increases morbidity in mice.
To begin defining the mechanism of NA-mediated activation of LTGF-β, we first asked if NA purified from the virion was sufficient for activation. Thus, recombinant LTGF-β (rLTGF-β) was incubated with buffer alone, purified A/Gray Teal/Australia/2/1979 virus (N4 virus), purified Gray Teal NA (N4 NA, BEI Resources, Manassas, VA), or low-protease-content NA purified from Clostridium perfringens (Roche), which was used as a non-viral NA control (bNA). All the samples were standardized to equivalent NA enzymatic activity and rLTGF-β activation was monitored by two different assays; the PAI/L bioassay, which monitors the activation of a TGF-β-specific reporter construct expressed in a stable cell line, or a sandwich ELISA specifically recognizing an epitope on the active TGF-β protein. All of the samples activated rLTGF-β in both assays in a dose-dependent manner. In the PAI/L bioassay, the concentration of active TGF-β increased with increasing amounts of NA (Fig. 1A). However, at the highest concentration of N4 virus (180,000 RFU) there was no TGF-β activity and the cells appeared dead. In the ELISA, both the N4 virus and purified NAs had low, but detectable levels of TGF-β activity at the lowest dose tested (10,000 RFU), that increased at 30,000 RFU, and then remained steady at the higher NA concentrations (Fig. 1B). Overall, these studies demonstrate that both influenza viral and a bacterial NA can activate LTGF-β.
To determine if NA-mediated activation involved removal of the sialic acid motifs on the LAP, rLTGF-β was incubated with PBS, bNA, N4 virus, or purified N4 NA, and the size of the LAP was determined by Western blot. HCl was used as a control for non-enzymatic-mediated activation of rLTGF-β. The rLTGF-β incubated with bNA, N4 virus, and N4 NA showed a slight shift in mobility of the LAP as compared to that incubated with PBS or HCl (Fig. 2A). There was no significant difference in the mobility between the bNA, N4 virus, and N4 NA. This shift in mobility was not evident when N4 NA was incubated with rLTGF-β purified from insect cells (Fig. 2B). The rLTGF-β produced by insect cells is unsialylated, as insect cells have no detectable sialyltransferase activity [28]. Thus, the lack of size change upon incubation with N4 NA suggests that the increased mobility of LTGF-β treated with N4 virus, bNA, or N4 NA is due to removal of sialic acid moieties. To confirm this, rLTGF-β was incubated with PBS, bNA, or N4 NA, proteins separated on a reducing SDS-PAGE, and sialic acid expression monitored by Western blot analysis using digoxigenin (DIG)–labeled lectins Maackia amurensis agglutinin (MAA; recognizes α2-3 sialic acid linkages, Fig. 2C) or Sambucus nigra agglutinin (SNA; recognizes α2-6 sialic acid linkages, Fig. S1). Blots were also probed with anti-LAP to confirm that the protein analyzed was the LAP (Fig. 2D). rLTGF-β incubated with PBS was detected by both SNA and MAA, suggesting the presence of both α2-6 and α2-3–linked sialic acids on the LAP (Fig. 2C and Fig. S1). In contrast, MAA and SNA failed to recognize bNA-treated rLTGF-β. Similarly, N4 NA–treated rLTGF-β was not recognized by MAA (Fig. 2C), but was detected by SNA (Fig. S1). However, this does not imply that the N4 NA fails to cleave α2-6 linkages. Hence, the mobility shift observed when rLTGF-β was incubated with NA is likely because of the loss of LAP-associated α2,3-linked sialic acids, but the specific sialic acid linkages removed may depend on the NA, as seen in the case of bNA-treated rLTGF-β.
To determine whether the enzymatic activity of NA was required for loss of specific sialic acid motifs and TGF-β activation, N4 virus or NA was pre-incubated with the influenza-specific inhibitor (NAi) oseltamivir carboxylate (10 nM) before incubation with rLTGF-β. Pre-incubation with NAi inhibited the mobility shift in the LAP (Fig. 2A), the loss of sialic acid (Fig. 2C), and TGF-β activation (Fig. 3A and B). N4 virus and NA–induced activation was completely inhibited with 10 nM NAi in the PAI/L assay (Fig. 3A) and to a lesser degree (75–95%) in the ELISA (Fig. 3B). Increasing the concentration of NAi up to 10 µM failed to completely inhibit activation in the ELISA assay (data not shown). Because the NAi is specific for influenza NA, it did not inhibit bNA-induced activation of rLTGF-β (Fig. 3A and 3B).
Proteases are established activators of LTGF-β [15] and can be contaminants of viral preparations or even components of the viral membrane [29], [30]. To examine the role for proteases, the LAP shift and activation assays were performed in the presence of a broad-spectrum protease inhibitor (PI) cocktail. The PI cocktail used in these studies had no effect on sialidase activity of either the virus or NAs and did not inhibit active TGF-β detection in either assay (data not shown). Unlike NAi, pre-incubation with PI had no effect on the LAP mobility shift (Fig. 2A). However, the PI inhibited the N4 virus and NA-induced activation of LTGF-β in the ELISA assay (Fig. 3B) and to some extent in the PAI/L assay (Fig. 3A), although the inhibition was not as much as that seen with NAi treatment. To determine the specific class of proteases causing the inhibition, virus was pretreated with increasing concentrations of individual protease inhibitors within their effective inhibitory ranges and incubated with rLTGF-β, and TGF-β activity was determined by ELISA (Fig. 3C). None of the protease inhibitors blocked rLTGF-β activation by the N4 virus. Further, when incubated with substrate for 1 h, all reagents tested protease-free (negative for trypsin, chymotrypsin, thrombin, plasmin, elastase, subtilisin, papain, cathepsin B, thermolysin, and pepsin) in a fluorescein thiocarbamoyl-casein derivative-based assay kit (data not shown). Even when incubations were extended to 24 h, only few viral stocks were positive for proteases (Fig. S2). Together, these data suggest that NA activates LTGF-β primarily via a mechanism involving enzymatic activity. However, a role for proteases cannot be discounted especially during infection in vivo.
As NA is essential for viral replication, we hypothesized that all influenza strains could activate LTGF-β. rLTGF-β was incubated with a panel of influenza virus subtypes, including two 2009 H1N1 pandemic strains, and several highly pathogenic avian influenza viruses (H5N1 and H5N9), and TGF-β activity was measured by the PAI/L assay (Fig. 4A) or ELISA (Fig. 4B). Although most of the strains activated LTGF-β, the levels of activation differed despite having equivalent NA activity. Surprisingly, several of the H5N1 influenza viruses failed to activate rLTGF-β; only A/Hong Kong/486/1997 (HK/486) consistently activated rLTGF-β (Fig. 4A and 4B). Incubation of rLTGF-β with a representative non-activating H5N1 virus, A/Hong Kong/483/1997 (HK/483), did not cause the expected mobility shift in the LAP (Fig. 4C), suggesting that the NA from viruses that did not activate rLTGF-β may also not cleave sialic acids from the LAP.
To determine whether the failure of H5N1 viruses to activate rLTGF-β was due to an intrinsic defect in the H5 NA protein, we first examined rLTGF-β activation by A/Teal/Hong Kong/W312/97 (Teal/HK) H6N1 virus. Teal/HK NA shares 97% sequence nucleotide homology with the H5N1 NA including a 19-amino-acid deletion in the stalk region and is the proposed donor of the NA and the internal genes of the H5N1 viruses [31]. Unlike the H5N1 viruses, Teal/HK virus activated rLTGF-β in both assays (Fig. 4A and 4B) suggesting that the deletion in the NA stalk domain has no effect on TGF-β activation.
To further assess the H5N1 NA, two H1N1 influenza viruses (A/California/04/09 and A/Puerto Rico/8/34) expressing the HK/483 NA were generated (CA/09+HK/483 NA and PR8+HK/483 NA) and tested for rLTGF-β activation. Both the parental viruses and the reassortant viruses containing the HK/483 NA activated rLTGF-β in the PAI/L (Fig. 5A) and ELISA (Fig. 5B) assays. Further, activation was inhibited by NAi but not the PI cocktail (Fig. 5C), confirming that the HK/483 NA can activate rLTGF-β in a NA-dependent manner.
To construct an H5N1 virus that activated rLTGF-β, HK/483 virus expressing the HK/486 NA was generated. Unfortunately, this virus was unable to activate rLTGF-β (Fig. 5A and 5B). Further, expressing the HK/483 NA on the HK/486 virus led to reduced activation as compared to the parental HK/486 virus. To evaluate LTGF-β activation in vivo, BALB/c mice were intranasally inoculated with PBS (control, n = 8) or 104 TCID50 units of the different reassortant viruses (n = 10) and lungs collected at 2, 4, and 7 days post-infection (dpi). Active or total (determined by acid activation of the sample) levels of TGF-β in the lung homogenates were determined by ELISA (Fig. 5D). Similar to the in vitro results, only HK/486 increased TGF-β activity in the lungs of infected mice as compared to PBS-inoculated mice. Levels of active TGF-β were increased >5-fold within 2 dpi, remained elevated at 4 dpi, before returning to control levels at 7 dpi (Fig. 5D). These kinetics were similar to those observed in mice infected with PR8 virus [27] and other highly pathogenic avian influenza viruses [32].
The total TGF-β levels in the lungs remained constant for all the viruses except for a significant decrease (∼60%) in the HK/483 infected mice at 2 dpi (Fig. 5D). This decline was not seen in mice infected with the HK/483+HK/486 NA reassortant virus, which remained at control levels at 2 dpi. These findings suggest that the NA may influence the total levels of TGF-β in the lungs of infected mice through an undefined mechanism.
Because we were unable to construct an HK/483 H5N1 virus that activates TGF-β in vivo, active TGF-β1 was administered to HK/483-infected mice by using a replication-deficient adenovirus vector. Twenty-four hours after HK/483 infection (104 TCID50), 108 PFUs of control adenovirus vector (AdDL70, n = 12), TGF-β-expressing vector AdTGFβ223/225 (n = 12), or PBS (n = 12) were administered intranasally. Lung TGF-β levels were measured at 2, 4, and 7 dpi. By 2 dpi, TGF-β levels in the lung increased to approximately 450 pg/ml in mice treated with the TGF-β–expressing adenovirus and remained above control levels even at 7 dpi (Fig. 6A). Mice treated with the control virus AdDL70 showed a transient increase in lung TGF-β activity at 2 dpi (100 pg/ml), which returned to control levels by 4 dpi. By 4 dpi with the HK/483 virus, all infected mice lost approximately 15% (p<0.01) of their initial body weight, which increased to more than 20% by 7 dpi in the HK/483 and +AdDL70 groups (Fig. 6B), at which time mice either succumbed to infection or were euthanized (Fig. 6C). Mice inoculated with AdTGFβ223/225 showed delayed weight loss and prolonged survival. At 7 dpi, weight loss remained at approximately 15% (p<0.01), but increased to 25% by 9 dpi (Fig. 6B). This was associated with a significant delay in mortality: AdTGFβ223/225- infected mice survived until 10 dpi (p<0.05, Fig. 6C). These mice also had significantly lower viral titers than HK/483-infected mice (Fig. 6D). By 2 dpi (1 day post AdTGFβ223/225 inoculation), viral titers decreased from approximately 107.5 TCID50 to 105.5 TCID50 (p<0.05) in the HK/483 alone and AdDL70 groups. Similar decreases in titers were observed at 4 and 7 dpi (p<0.05) in the HK/483 alone group. However, there was no significant difference in titers between the AdDL70 and AdTGFβ223/225 groups at 4 dpi.
Given the increased survival of mice infected with AdTGFβ223/225, we tested whether pretreatment with TGF-β afforded additional protection to mice. Mice (n = 12) were administered 108 PFUs AdDL70 control or the AdTGFβ223/225 virus 48 h before HK/483 infection. Pretreatment with AdTGFβ223/225 provided no added protection; all the HK/483-infected mice succumbed to infection by 8 dpi (Fig. S3B). Both the uninfected and HK/483-infected mice pretreated with AdTGFβ223/225 lost significantly more weight by 4 dpi than mice in other groups (10% vs. 0%, p<0.01, Fig. S3A), suggesting that increased TGF-β activity before H5N1 influenza infection can be detrimental to mice.
We then examined the effect of removing TGF-β during HK/486 infection by depleting TGF-β using a pan-TGF-β neutralizing antibody. Briefly, 1D11 antibody or isotype-control IgG was administered and total TGF-β levels in the lungs were monitored by ELISA (Fig. 7A). Total TGF-β levels in HK/486-infected mice were significantly (3 times) lower (p = 0.0003) by 24 hpi than in the HK/486-infected mice receiving isotype IgG. Two doses of the neutralizing antibody decreased TGF-β levels to control levels; by 7 dpi, levels were significantly (3 times) lower (p = 0.04) than control levels. By 8 dpi, all HK/486-infected mice (105 TCID50, n = 15) lost approximately 20% of their starting weight whereas HK/486-alone mice lost only 10% (Fig. 7B, p<0.01). By 9 dpi, 40% of mice in the TGF-β–depleted group succumbed to infection, and all died by 10 dpi (Fig. 7C), whereas those in the HK/486 and isotype IgG groups began to recover and gain weight. The increased mortality in the TGF-β–depleted group was not associated with a significant increase in viral replication. HK/486-infected mice with and without isotype IgG had peak lung titers of approximately 104.6 TCID50 by 2 dpi, which decreased to 101.5 TCID50 by 10 dpi (Fig. 7D). In contrast, the TGF-β–depleted group had a slight, although not significant, increase in viral titers at 2 and 4 dpi. At 8 dpi, 1D11-treated mice had a 15-fold increase in viral titers over HK/486 and isotype IgG–treated mice (p<0.02). No virus was detected in control tissues or outside the lungs of infected mice.
To determine if these findings were specific to the highly pathogenic H5N1 influenza viruses, mice (n = 6) were pre-treated with PBS, the 1D11 antibody or isotype-control IgG as described, infected with A/California/04/09 (CA/09, 105 TCID50), and monitored for morbidity. The CA/09-infected mice treated with PBS or receiving the isotype IgG lost approximately 20–25% of their starting weight by 6 to 8 dpi before returning to day 0 weights by 12 dpi (Fig. 8A). Clinically the mice had ruffled fur and were shivering. The 1D11-treated mice followed a similar pattern but lost significantly more weight by 6 dpi (p = 0.047) and had a delayed recovery with significantly more weight loss still evident at 12 dpi (p = 0.029). The 1D11-treated mice had more significant clinical signs of infection including rear-leg paralysis and had 20% mortality by 8 dpi reaching 67% by 12 dpi (Fig. 8B). Taken together, the data suggest that TGF-β is modulated by the virus, and this modulation during infection may be important in disease outcome.
These studies establish NA as a direct activator of LTGF-β and demonstrate a role for TGF-β in protection against influenza virus pathogenesis. We have previously shown that purified influenza virus activates TGF-β [27] and that antibodies to the viral NA but not the HA inhibit viral-mediated LTGF-β activation. In this study, we demonstrate that purified NA alone can convert the biologically latent form of TGF-β to its active form and that TGF-β plays an important role in protection against influenza virus pathogenesis. Activation of LTGF-β by viral NA involves removal of sialic acid moieties to release the active TGF-β molecule from the latent complex or expose other residues for cell surface interactions. To our knowledge, these are the first studies demonstrating that microbe-associated sialidases can directly activate LTGF-β.
Our study also shows that NA-mediated LTGF-β activation is not specific to influenza virus. As the topology of the NA catalytic domain is well conserved and the active sites share many structural features [33], NAs from diverse pathogens may activate LTGF-β. In our study, Clostridium perfringens–derived bNA also activated LTGF-β, which is consistent with previous studies showing LTGF-β activation by bNA, although sialidase activity was not explicitly identified as the means of activation [34], [35]. Paramyxoviruses, which also have a functional NA protein, directly activate LTGF-β (unpublished data). A study by Zou and Sun demonstrated that LTGF-β2 and LTGF-β3 were also activated by NA [36], suggesting that NA may be a biological activator of numerous types of LTGF-β.
Despite the functional conservation among NAs, some highly pathogenic avian (HPAI) H5N1 influenza viruses failed to activate LTGF-β in vitro and in vivo [32], [37]. The NA from these viruses has a 19-amino-acid deletion in the stalk [31], [38] that could contribute to the decreased ability to activate LTGF-β. To test this possibility, we assessed the activation of rLTGF-β by the Teal/HK H6N1 virus. Hoffmann et al., proposed that this virus may have donated the NA gene to the H5N1 viruses given the high degree of nucleotide homology [31]. In spite of the stalk deletion, Teal/HK activated rLTGF-β unlike the H5N1 viruses. Further, expressing the HK/483 NA on either PR8 or CA/09 virus had no effect on rLTGF-β activation suggesting that there is no intrinsic defect in the NA.
However, expressing the HK/483 NA on the H5 HK/486 virus led to an inability to activate LTGF-β in vitro and in vivo. In addition the HK/486 NA failed to rescue the activation phenotype with the HK/483 virus. A recent study by Matsuoka et al demonstrated that short-stalk NAs from the H5N1 viruses are more virulent in mice and chickens. Intriguingly, the NA-mediated virulence can be affected by HA glycosylation. Virulence in mice conferred by a short stalk NA was most evident when the HA had no glycosylation [38]. Although virulence in vivo is much more complicated than LTGF-β activation, studies are underway to examine the role of HA in LTGF-β. We hypothesize that although HA will not be directly involved in activation, it may influence the ability of NA to activate.
The question remains whether NA-mediated activation has an important biological role in TGF-β activation during influenza infection in vivo. At this time we can't definitively answer that question. What is intriguing is that the NA may influence the total levels of lung TGF-β during infection. Mice infected with HK/483 had a dramatic decrease in total LTGF-β levels by 2 dpi (from ∼2000 pg/ml to ∼1000 pg/ml). This phenotype was reversed with the HK/483 virus expressing the HK/486 NA. A similar trend was seen when HK/483 NA was expressed on HK/486; a significant decrease in total TGF-β levels. Studies are on-going to determine if this is due to a change in the cells in the lung associated with TGF-β secretion or if the viruses differentially regulate the known physiologic activators.
There are numerous physiologic TGF-β activators in the lung: the infected epithelium could release thrombospondin-1, proteases and matrix metalloproteases, or even reactive oxygen species (reviewed in [15], [21], [39]). Virus-induced injury to the epithelium can directly activate LTGF-β through the induction of apoptosis [40] or the upregulation of integrins [41]–[44], and immune cells have high levels of active TGF-β [45], [46]. Proteases may play a role in influenza virus-induced LTGF-β activation, especially during influenza infection in vivo, wherein cellular proteases are essential for influenza virus replication (reviewed in [47]–[50]). Proteases can be contaminants of viral preparations or even components of the viral membrane [29], [30]. Thus, only protease-free reagents were used in our assays, and a broad-spectrum PI cocktail partially blocked influenza virus and NA-mediated LTGF-β activation. Further studies confirmed that the broad-spectrum PI cocktail had no effect on either sialidase activity (as measured in the MUNANA assay) or directly on TGF-β detection in either assay (data not shown). However, we have not been able to identify the specific class of proteases or a potential cleavage site within LTGF-β by mass spectrometry (data not shown).
Since our initial attempts to construct H5 viruses that can activate TGF-β in vivo were unsuccessful, we evaluated the role of TGF-β in influenza pathogenesis by using a neutralizing antibody and administering exogenous TGF-β via an adenovirus vector. Mice administered TGF-β neutralizing antibody during HK/486 H5N1 infection had higher morbidity and mortality than mice treated with control virus or isotype IgG. This finding was not specific to H5N1 influenza viruses; inhibiting TGF-β activity during 2009 H1N1 infection also increased morbidity. Although mice exhibited clinical signs of illness, administration of exogenous TGF-β to HK/483-infected mice once at 24 hpi delayed morbidity and mortality. Administration of exogenous TGF-β 48 h pre-infection did not affect survival, and TGF-β–treated infected and non-infected mice had increased weight loss by 4 dpi, suggesting that the timing of TGF-β activation may be important.
Although we are still investigating the specific protective role of TGF-β during influenza infection, we did find that mice administered exogenous TGF-β had significantly lower titers within 2 dpi than untreated infected and AdDL70-treated mice. This decrease in viral load may contribute to the delayed morbidity observed, but was not sufficient to protect mice from severe infection. In contrast, depleting TGF-β during HK/486 infection had little to no significant effect on viral load until 8 dpi. Thus, mechanisms other than reduction of viral load may be involved in TGF-β–mediated modulation of influenza pathogenesis.
TGF-β serves as a global regulator of immunity by controlling the initiation and resolution of inflammatory responses (reviewed in [39], [46]). Thus, a pathogen that can regulate TGF-β activation could promote an immune-privileged state for itself within its host, as has been seen in the case of multiple parasitic, bacterial, and fungal pathogens (reviewed in [46], [51]–[55]). We postulate that failure of certain H5N1 influenza viruses to activate TGF-β [56], [57] may result in improper immune stimulation and resolution, contributing to exacerbated immunopathology for the host. Further investigation into specific immune cell activities and cytokine profiles during TGF-β modulation is required to fully elucidate the mechanisms of TGF-β regulation of influenza virus replication.
All procedures involving animals were approved by the Southeast Poultry Research Laboratory (USDA-ARS), University of Wisconsin-Madison School of Medicine and Public Health, and the St. Jude Children's Research Hospital IACUCs and were in compliance with the Guide for the Care and Use of Laboratory Animals. These guidelines were established by the Institute of Laboratory Animal Resources and approved by the Governing Board of the U.S. National Research Council.
All experiments in which H5N1 viruses were used were conducted in a Biosafety level 3 enhanced containment laboratory [58]. Investigators were required to wear appropriate respirator equipment (RACAL, Health and Safety Inc., Frederick, MD). Mice were housed in HEPA-filtered, negative pressure, vented isolation containers (M.I.C.E. ®, Animal Care Systems, Littleton, CO).
A/Turkey/Wisconsin/68 (Tk/WI, H5N9), A/Gray Teal/Australia/2/79 (N4, H4N4), A/Swine/Nebraska/2/92 (Sw/NB, H1N1), A/Turkey/England/69 (Tk/Eng, H3N2), A/Turkey/Oregon/71 (Tk/Oreg, H7N3), A/Turkey/Ontario/6528/67 (Tk/Ont, H5N9), A/Mallard/Wisconsin/8/76 (Mal/WI, H1N1), A/Teal/Hong Kong/W312/97(Teal/HK, H6N1), and the 2009 H1N1 A/California/04/09 (CA/09) and A/Wisconsin/054/09 viruses were propagated in the allantoic cavity of 10-day-old specific pathogen-free embryonated chicken eggs (Sunnyside Farms, Beaver Dam, WI) at 37°C. Allantoic fluid was harvested, clarified by centrifugation and stored at −70°C. A/Puerto Rico/8 (PR8, H1N1), A/WSN/33 (WSN, H1N1), A/New Caledonia/20/99 (New Cal, H1N1), A/Hawaii/10/2002 (Hawaii, H1N2), A/Wyoming/3/2003 (Wyom, H3N2), A/Korea/770/2002 (Korea, H3N2), A/Aichi/2/68 (Aichi, H3N2), and the H5N1 viruses A/Hong Kong/156/97 (HK/156), A/Hong Kong/486/1997 (HK/486), A/Hong Kong/483/1997 (HK/483), A/Vietnam/1203/2004 (VN/1203), and A/Vietnam/1194/2004 (VN/1194) were propagated in Madin-Darby canine kidney (MDCK) cells as described previously [59]. Culture supernatants were harvested, clarified by centrifugation, and stored at −70°C. All viral titers were determined by 50% tissue culture infective dose (TCID50) analysis in MDCK cells and evaluated by the method of Reed and Muench [60]. MDCK cells were cultured in Eagle's minimal essential medium (MEM) supplemented with 2 mM glutamine (Mediatech, Manassas, VA), and 10% fetal bovine serum (FBS, Gemini Bio-Products, West Sacramento, CA). Mink lung epithelial cells stably transfected with the TGF-β-sensitive plasminogen activator inhibitor reporter construct (Mv1Lu-PAI cells, generous gift of Dr. Daniel Rifkin, New York University) were propagated in Dulbecco's modified Eagle's medium (DMEM) supplemented with 2 mM glutamine, 7% FBS, and 400 µg/ml Geneticin (G418, Calbiochem, La Jolla, CA).
Gray Teal influenza virus was purified by sucrose gradient ultracentrifugation [61]. Purified Gray Teal NA was obtained through the NIH Biodefense and Emerging Infections Research Resources Repository, NIAID, NIH: N4 Neuraminidase (NA) Protein from Influenza Virus, A/grey teal/Australia/2/79 (H4N4), Recombinant from baculovirus, NR-656 (BEI Resources, Manassas, VA). Briefly, it was expressed in Sf9 cells using a baculovirus expression vector system and purified using conventional chromatographic techniques.
NA enzymatic activity was determined by the MUNANA (2-(4-methylumbelliferyl)-α-d-N-acetylneuraminic acid) assay as described previously [62]. NA inhibition was assayed with purified NA, and virus was standardized to equivalent NA enzyme activity and incubated for 1 h at 37°C with oseltamivir carboxylate (0–1000 nM, generous gift of Hoffman La-Roche, Inc., Nutley, NJ).
TGF-β activity was assessed by the plasminogen activator inhibitor-luciferase (PAI/L) bioassay or by a TGF-β-specific ELISA following manufacturer's instructions (R&D Systems, Minneapolis, MN). The ELISA is a quantitative sandwich immunoassay where a monoclonal antibody specific for the active region of TGF-β1 is coated onto a microplate and any bound TGF-β1 is detected with an enzyme-linked polyclonal antibody specific for TGF-β1. It will not detect the latent form of TGF-β1. The PAI/L bioassay was performed as previously described [63], with several modifications. Briefly, 2×104 Mv1Lu-PAI cells per well of 96-well plates were incubated overnight, washed with PBS, and incubated with 100 µl/well test sample for 5 h at 37°C, 5% CO2. After washing, cells were lysed and luciferase activity measured by using a luciferase assay substrate (Promega, Madison, WI) on a Turner Biosystems 20/20n luminometer (Turner Biosystems Instruments, Sunnyvale, CA). Test samples included 10 ng/ml recombinant LTGF-β1 (rLTGF-β1, R&D Systems) incubated with different concentrations of low-protease-content Clostridium perfringens-purified NA (Roche Applied Sciences, Indianapolis, IN), purified NA, or influenza virus in serum-free DMEM containing 0.1% BSA for 1 h at 37°C. To generate TGF-β1 standard curves, 2-fold dilutions of active TGF-β1 (0–1000 pg/ml, R&D Systems) in DMEM containing 0.1% BSA were added to Mv1Lu-PAI cells. To determine the role of NA activity or proteases, rLTGF-β1 was pre-incubated for 1 h at 37°C with different NA activities of bNA, purified influenza virus, or viral NA (as noted in figures and figure legends) in the presence of oseltamivir carboxylate (10 nM) or 1× EDTA-free PI cocktail (Pierce, Rockford, IL).
To examine the role of individual proteases, 106 TCID50 units/ml of purified Tk/WI influenza virus was pre-incubated for 1 h at 37°C with bestatin (20–1000 nM; Sigma), leupeptin (10–100 µM; Sigma), or GM 1489 (1–500 nM; Calbiochem), followed by incubation with 10 ng/ml rLTGF-β1 for 1 h at 37°C. LTGF-β1 activation was determined by the PAI/L assay. All protease inhibitors were used within their effective inhibitory concentrations, as determined by the manufacturer. To test for the presence of proteases in experimental reagents, including purified virus, proteins, and inhibitors, a thiocarbamoyl-casein derivative-based assay was used as per manufacturer's instructions (Calbiochem) in the presence or absence of 1× protease inhibitor cocktail (Pierce).
rLTGF-β (0.4 µg) or rLAP (0.5 µg, R&D Systems) was incubated with PBS, HCl (final pH of 2), purified Gray Teal virus (2 µg), purified Gray Teal NA (0.5 µg), or bNA either alone or pre-incubated with NAi (1 µM) or 1× PI for 1 h at 37°C as described previously. Samples were then separated on a 5%–20% SDS-PAGE gel under reducing conditions. After transferring to nitrocellulose, blots were blocked in 2% non-fat dry milk in Tris-buffered saline plus 1% Tween 20 (TTBS) for 1 h at room temperature and probed for LAP with mouse-anti-LAP (1∶500, R&D Systems) in TTBS for 1 h at room temperature. Blots were washed and incubated with goat anti-mouse-HRP (1∶5000, Jackson Laboratories, Bar Harbor, ME). To examine the sialic acids present on LAP, TGF-β samples were prepared as described above and blots were probed with DIG-labeled MAA (1∶200), which recognizes α2,3-linked sialic acids, or SNA (1∶1000), which recognizes α2,6-linked sialic acids, for 1 h at room temperature (DIG glycan differentiation kit, Roche Applied Science). Lectins were visualized by staining with anti-DIG-AP. After detection, blots were analyzed by Western blotting for LAP, as described above.
H5N1 reverse genetic viruses were generated by the RNA polymerase I reverse genetics system [64]. The PR8 and CA/09 virus expressing HK/483 NA was constructed by using the eight-plasmid system as previously described [65]. P1 viral stocks were generated, the NA genes sequenced to ensure that no spurious mutations arose during viral propagation, stiters determined by TCID50 analysis in MDCK cells, and NA activity quantitated by the MUNANA assay as described above. To determine TGF-β levels with the reassortant viruses, BALB/c mice were lightly anesthetized and inoculated with 104 TCID50 units of the different reassortant viruses or PBS alone, as described below.
To deplete TGF-β activity during HK/486 and CA/09 virus infection, 4- to 6-week-old BALB/c mice (Charles River Laboratories, Wilmington, MA) were intraperitoneally (i.p.) inoculated with PBS, anti-TGF-β neutralizing antibody 1D11, or isotype-matched mouse IgG (0.5 mg per mouse, Sigma) in PBS 6 to 48 h pre-infection and then every 48 hpi. Mice were then intranasally (i.n.) inoculated with 25 µl PBS or 105 TCID50 virus. 1D11 was either purchased (R&D Systems) or purified from the 1D11 hybridoma (ATCC# 1D11.16.8). 1D11 produces IgG1 antibodies that neutralize all 3 mammalian TGF-β isoforms (β1, β2, β3) [66]. IgG was purified from cell culture supernatants by T-Gel (Pierce), potential endotoxin contaminations removed by Detoxi Gel Endotoxin Removing Gel (Pierce), and purified IgG concentrated and buffer exchanged with Amicon Ultra-15 concentrators (Millipore, Bedford, MA). The endotoxin levels were less than 0.2 EU/mg as measured by the Biowhittaker QCL-1000 assay (Biowhittaker, Walkersville, MD).
Exogenous active TGF-β1 was administered to mice by infection with a replication-defective adenovirus expressing active TGF-β1 (AdTGFβ223/225) or the control vector (AdDL70) as described previously [67], [68]. Briefly, full-length porcine TGFββ1 cDNA (differing from murine TGF-β1 by 1 amino acid) was mutated at serine 223 and 225 (TGF-β223/225) to render the protein constitutively active and expressed in a recombinant, replication-deficient type-5 adenovirus. The replication-deficient virus (AdTGFβ223/225) was purified by cesium chloride (CsCl) gradient centrifugation and concentrated by using a Sephadex PB-10 chromatography column. Mice were i.n. inoculated with 108 PFUs of active AdTGFβ223/225 or AdDL70 either 48 h pre- or 24 h post-infection with 104 TCID50 units of HK/483 virus. Mice were monitored daily and weighed every 48 h post-infection. At different time points post-infection, 2 mice from the control group and 3 mice from the experimental group were euthanized and lungs collected. Tissues were homogenized in cold PBS, and clarified tissue homogenates were tested for TGF-β levels, using a mouse-specific ELISA (R&D Systems) or viral titers by TCID50 analysis on MDCK cells.
Statistical significance of data was determined by using analysis of variance (ANOVA) or Student's t- test on GraphPad Prism (San Diego, CA). All assays were run in triplicate and are representative of at least 2 separate experiments. Error bars represent standard deviation, and statistical significance was defined as a p value of less than 0.05.
|
10.1371/journal.pbio.1000616 | Structural Insight into the Rotational Switching Mechanism of the
Bacterial Flagellar Motor | The bacterial flagellar motor can rotate either clockwise (CW) or
counterclockwise (CCW). Three flagellar proteins, FliG, FliM, and FliN, are
required for rapid switching between the CW and CCW directions. Switching is
achieved by a conformational change in FliG induced by the binding of a
chemotaxis signaling protein, phospho-CheY, to FliM and FliN. FliG consists of
three domains, FliGN, FliGM, and FliGC, and
forms a ring on the cytoplasmic face of the MS ring of the flagellar basal body.
Crystal structures have been reported for the FliGMC domains of
Thermotoga maritima, which consist of the FliGM
and FliGC domains and a helix E that connects these two domains, and
full-length FliG of Aquifex aeolicus. However, the basis for
the switching mechanism is based only on previously obtained genetic data and is
hence rather indirect. We characterized a CW-biased mutant
(fliG(ΔPAA)) of Salmonella enterica by
direct observation of rotation of a single motor at high temporal and spatial
resolution. We also determined the crystal structure of the FliGMC
domains of an equivalent deletion mutant variant of T. maritima
(fliG(ΔPEV)). The FliG(ΔPAA) motor produced torque
at wild-type levels under a wide range of external load conditions. The
wild-type motors rotated exclusively in the CCW direction under our experimental
conditions, whereas the mutant motors rotated only in the CW direction. This
result suggests that wild-type FliG is more stable in the CCW state than in the
CW state, whereas FliG(ΔPAA) is more stable in the CW state than in the CCW
state. The structure of the TM-FliGMC(ΔPEV) revealed that
extremely CW-biased rotation was caused by a conformational change in helix E.
Although the arrangement of FliGC relative to FliGM in a
single molecule was different among the three crystals, a conserved
FliGM-FliGC unit was observed in all three of them. We
suggest that the conserved FliGM-FliGC unit is the basic
functional element in the rotor ring and that the PAA deletion induces a
conformational change in a hinge-loop between FliGM and helix E to
achieve the CW state of the FliG ring. We also propose a novel model for the
arrangement of FliG subunits within the motor. The model is in agreement with
the previous mutational and cross-linking experiments and explains the
cooperative switching mechanism of the flagellar motor.
| The bacterial flagellum is a rotating organelle that governs cell motility. At
the base of each flagellum is a motor powered by the electrochemical potential
difference of specific ions across the cytoplasmic membrane. In response to
environmental stimuli, rotation of the motor switches between counterclockwise
and clockwise, with a corresponding effect on the swimming direction of the
cell. Switching is triggered by the binding of the signaling protein
phospho-CheY to FliM and FliN, and achieved by conformational changes in the
rotor protein FliG. The actual switching mechanism, however, remains unclear. In
this study, we characterized a fliG mutant of
Salmonella that shows an extreme clockwise-biased rotation,
and determined the structure of a fragment of FliG (FliGMC) of the
equivalent mutant variant of Thermotoga maritima.
FliGMC is composed of two domains and covers the regions
essential for torque generation and FliM binding. We showed that the mutant
structure has a conformational change in the helix connecting the two domains,
leading to a domain orientation distinct from that of the wild-type FliG. On the
basis of this structure, we propose a new model for the arrangement of FliG
subunits in the rotor that is consistent with the previous mutational studies
and explains how cooperative switching occurs in the motor.
| Bacteria such as Escherichia coli and Salmonella
enterica swim by rotating multiple flagella, which arise randomly over
the cell surface. Each flagellum is a huge protein complex made up of about 30
different proteins and can be divided into three distinct parts: the basal body, the
hook, and the filament. The basal body is embedded in the cell envelope and acts as
a reversible motor powered by a proton motive force across the cytoplasmic membrane.
The hook and the filament extend outwards in the cell exterior. The filament is a
helical propeller that propels the cell body. The hook connects the basal body with
the filament and functions as a universal joint to transmit torque produced by the
motor to the filament. The flagellar motor can exist in either a counterclockwise
(CCW) or clockwise (CW) rotational state. CCW rotation causes the cell to swim
smoothly in what is termed a run, whereas brief CW rotation of one or more flagella
causes a tumble. The direction of motor rotation is controlled by environmental
signals that are processed by a sensory signal transduction pathway to generate
chemotaxis behavior [1]–[3].
Five flagellar proteins, MotA, MotB, FliG, FliM, and FliN, are involved in torque
generation. Two integral membrane proteins, MotA and MotB, form the stator, which
converts an inwardly directed flux of H+ ions through a
proton-conducting channel into the mechanical work required for motor rotation. The
FliG, FliM, and FliN proteins form the C ring on the cytoplasmic side of the MS
ring, which is assembled from 26 subunits of a single protein, FliF, and this
complex acts as the rotor of the flagellar motor [1]–[3]. An electrostatic interaction
between the cytoplasmic loop of MotA and FliG is thought to be involved in torque
generation [4],[5] and in stator assembly around the rotor [6]. The
protonation-deprotonation cycle of a highly conserved aspartic acid residue in MotB
is coupled to the movement of the MotA cytoplasmic loop to generate torque [7]–[9].
Because FliG, FliM, and FliN are also responsible for switching the direction of
motor rotation, their assembly is called the switch complex [10]. Binding of a chemotactic
signaling protein CheY-phosphate (CheY-P) to FliM and FliN is presumed to induce
conformational changes in FliG that result in a conformational rearrangement of the
rotor-stator interface, allowing the motor to spin in the CW direction [11],[12]. The switching
probability is also affected by motor torque, suggesting that the switch complex
senses the stator-rotor interaction as well as the concentration of CheY-P [13],[14]. Recently,
turnover of FliM and heterogeneity in the number of FliM subunits within functioning
motors have been reported [15],[16]. The turnover rate is increased by the presence of
CheY-P, implying that turnover of FliM may be directly involved in the switching
process [15].
FliG forms a ring on the cytoplasmic face of the MS ring with 26-fold rotational
symmetry [17],[18]. FliG consists of three domains, FliGN,
FliGM, and FliGC. FliGN is responsible for
association with the cytoplasmic face of the MS ring [17],[19], and FliGM and
FliGC are required for an interaction with FliM [20]. The FliGM domains of
adjacent subunits are fairly close to each other in the FliG ring [21]. The crystal
structure of FliGMC of Thermotaoga martima
(Tm-FliGMC) shows that FliGM and FliGC are
connected by an extended α-helical linker (helix E) [22]. The linker contains two
well-conserved Gly residues and hence might be flexible [22]. This finding is supported by
genetic analyses of FliG and a computer-generated prediction of its secondary
structure [23],[24]. Critical charged residues, which are responsible for an
interaction with MotA [4]–[6], are clustered together along a prominent ridge on
FliGC
[25]. It has been
shown that the elementary process of torque generation by the stator-rotor
interaction is symmetric in CCW and CW rotation [26], although the torque-speed
curves are distinct between them [27].
A recent report on the full-length FliG structure of Aquifex
aeolicus has shown two distinct conformational differences between the
full-length FliG and FliGMC structures [28]. The helix E linker is held in a
closed conformation by packing tightly against an α-helix (helix n), which
connects FliGN to FliGM in a way similar as helix E connects
FliGM and FliGC in the full-length FliG structure. Helix E
is dissociated from FliGM in the Tm-FliGMC structure,
resulting in its being in an open conformation. The conformation of FliGC
is also different in these two structures. Combined with the previous genetic data,
it has been proposed that the closed conformation represents FliG during CCW
rotation and that switching to CW rotation may be accompanied by the dissociation of
helix E from FliGM to form an open conformation.
The S. enterica FliG(ΔPAA) mutant protein has three-amino-acid
deletion at positions 169 to 171. Motors containing this protein are extremely CW
biased [29]. The
mutant motors remain in CW rotation even in the presence of a cheY
deletion, indicating that the motor is locked in the CW state [29]. Therefore, it is likely
that binding of CheY-P to FliM may introduce a conformational change in FliG similar
to the one introduced by the in-frame PAA deletion. To elucidate the switching
mechanism, we crystallized a fragment of a T. maritima FliG mutant
variant, FliGMC(ΔPEV), which contains a deletion equivalent to
S. enterica FliGMC(ΔPAA), and determined its
structure at 2.3 Å resolution. Based on the structural difference among
full-length A. aeolicus FliG, wild-type Tm-FliGMC, and
its deletion variant, we suggest that a reorientation of helix E relative to
FliGM is important for switching and propose a new model for the
arrangement of FliG subunits in the motor.
The motors of the fliG(ΔPAA) mutant rotated only CW (Figure
S1A), whereas wild-type motors rotated exclusively CCW under our
experimental conditions. The motors of the deletion mutant produced normal
torque under a wide range of external-load conditions, indicating that the
deletion does not affect the torque generation step (Figure
S1B). Introduction of a cheA-Z deletion, which causes
wild-type motors to spin exclusively CCW [30], into the
fliG(ΔPAA) mutant did not change the CW-locked
behavior. These results are in good agreement with a previous report [29].
Switching between the CW and CCW states is highly cooperative [31]–[34]. The switching
mechanism can be explained by a conformational spread model, in which a
switching event is mediated by conformational changes in a ring of subunits that
spread from subunit to subunit via nearest-neighbor interactions [34],[35]. Therefore we
investigated rotation of a single motor composed of wild-type and mutant FliG
subunits at different ratios. FliG(ΔPAA) inhibited expansion of wild-type
colonies in semi-solid agar (Figure
1A), even when its expression level was ca. 5-fold lower than the
level of wild-type FliG expressed from the chromosome (Figure 1B). Bead assays revealed that the
decrease in colony expansion results from an increase in both switching
frequency and prolonged pausing (Figure 1C). In addition, a low level expression of FliG(ΔPAA)
partially increased the colony expansion of the ΔcheA-Z
smooth-swimming mutant, presumably because switching now occurred (Figure 1D, upper and middle
panels). These results suggest that even a small fraction of FliG(ΔPAA) in a
motor can affect the CW-CCW switching.
The CW-CCW transition, which is very fast in wild-type motors, became
significantly longer in mixed motors (Figure 1), suggesting that, as proposed
previously [24], the motor can exist in multiple states. A much
higher expression of FliG(ΔPAA) completely inhibited wild-type motility
(Figure 1D) and did not
increase the colony size of the ΔcheA-Z mutant in
semi-solid agar plates because of the extreme CW-biased rotation of its flagella
(Figure 1C and D, lower
panel), in agreement with data showing that a higher expression level of
wild-type FliG is required for complementation of the
fliG(ΔPAA) mutant (Figure S2). Therefore, we conclude that
wild-type FliG is more stable in the CCW state than in the CW state, whereas
FliG(ΔPAA) is more stable in the CW state than in the CCW state.
To identify structural differences between the CW and CCW states of FliG, we
carried out limited trypsin proteolysis of the wild-type and mutant FliG
proteins and analyzed the products by matrix-assisted laser desorption
ionization time-of-flight (MALDI-TOF) mass spectrometry and N-terminal
amino-acid sequencing (Figure
2). Both the wild-type and mutant FliG proteins were cleaved between
helix E and FliGC, producing the T1 and T2a fragments. This indicates
that there is a flexible region between them. The T1 fragment derived from
FliG(ΔPAA) was less stable than the T1 fragment from wild-type FliG,
suggesting that the deletion causes a conformational change in FliGM
and helix E. In contrast, the T2a fragment was more stable in FliG(ΔPAA)
than in the wild-type. The T2a fragment derived from the wild-type FliG protein
was detected by MALDI-TOF but not on SDS-PAGE gels, indicating that the
wild-type T2a fragment is rapidly converted into the T2 fragment. These results
suggest that the deletion also influences the conformation in the region between
helix E and FliGC.
We tried crystallizing both wild-type FliG and FliG(ΔPAA) from S.
enterica but did not succeed in obtaining crystals. It has been
reported that the crystal structure of a fragment (residues 104–335) of
T. martima FliG (Tm-FliGMC) consists of
FliGM, FliGC, and helix E connecting the two domains
([22]; PDB
ID, 1lkv). FliGC can be further divided into two sub-domains
(FliGCN and FliGCC). Therefore, we introduced the
deletion (ΔPEV), equivalent to ΔPAA, into Tm-FliGMC
(Tm-FliGMC(ΔPEV)) and determined its structure at 2.3 Å
resolution by X-ray crystallography (Figure 3).
FliGM, FliGCN, and FliGCC are composed of five
(n, A–D), three (F–H), and six (I–N) helices, respectively
(Figure 3). Since the
residues between G186 and V195 are invisible in the crystal, there are two
possible ways to connect FliGM with FliGCN: one is to
connect FliGM with its adjacent FliGCN (G186 to V195 in
Figure 3A upper panel
and Figure
S3A), and the other is with a distant FliGCN (G186 to
V195' in Figure 3A
upper panel and Figure S3A). The Cα distance between G186 and V195, and G186 and
V195' is 16.9 Å and 27.9 Å, respectively. Therefore, to connect
with the distant FliGCN, the invisible chain would have a fully
extended conformation. We thus conclude that the connection with the adjacent
FliGCN is more plausible.
Compared with the structure of wild-type Tm-FliGMC, FliG(ΔPEV)
showed a significant conformational change in the hinge between helix E and
FliGM, leading to a very different orientation of helix E
relative to FliGM (Figure 3A and B, and Figure 4A and C). As a result, some of the residues in
FliGM are exposed to solvent in the
Tm-FliGMC(ΔPEV) structure. This result is in good agreement with
the data obtained by limited proteolysis (Figure 2). Thus, the conformational
difference in the FliGM-helix E hinge between the wild-type and
mutant structures may represent the conformational switch between the CW and CCW
states of the motor.
The C-terminal half of helix E is disordered and protrudes into the solvent
channel in the Tm-FliGMC(ΔPEV) crystal (Figure
S3A). In contrast, helix E in the wild-type crystal is stabilized by
forming an anti-parallel four-helix bundle structure with the E helices of three
adjacent subunits related by crystallographic symmetry (Figure S3B)
[22].
Therefore, the orientation of FliGC relative to FliGM is
different between the wild-type and the deletion variants (Figure 3A and B upper panel). Because the
disordered region of helix E is far from the PEV deletion, we conclude that
helix E has a highly flexible nature, which may be responsible for the switching
mechanism, as suggested before [23],[24].
Tm-FliGMC(ΔPEV) also showed a conformational difference in the
H–I loop, resulting in a rigid body movement of FliGCC relative
to FliGCN (Figure 3A and
B middle and lower panels, and Figure 4A). This movement is consistent with
the limited proteolysis data because, in the Tm-FliGMC(ΔPEV)
structure, FliGCC almost covers D199, which is the residue
corresponding to R198 in S. enterica FliG. It is, however,
unclear how the deletion affects the conformation of the H–I loop, because
neither direct contact between FliGCC and helix E nor significant
structural difference in FliGCN is observed.
The crystal structure of full-length A. aeolicus FliG (Aa-FliG)
showed that the conformation of helix E and the orientation of FliGCN
relative to FliGCC are quite distinct from those of wild-type
Tm-FliGMC
[28]. We compared
the Aa-FliG structure with the Tm-FliGMC(ΔPEV) structure and
found that the conformation of helix E and the relative conformation of
FliGCC to FliGCN are also different in those two
structures (Figure 3A and C,
and Figure 4B and C). The
conformational differences are greater than those between Tm- FliGMC
and Tm-FliGMC(ΔPEV). The conformation of helix E in Aa-FliG seems
to be stabilized by interactions of helix E with FliGM and helix n in
the crystal (Figure S3C). As mentioned earlier, the conformation of helix E and
the orientation of FliGCC to FliGCN are also different
between the wild-type and mutant Tm-FliGMC structures. Therefore,
these conformational differences among the three structures strongly suggest
that both helix E and the linker connecting FliGCN to
FliGCC are highly flexible.
The interaction between FliGM and FliGCN, which share the
armadillo repeat motif [36] that is often responsible for protein-protein
interaction, is very tight in the Tm-FliGMC(ΔPEV) crystal, in
agreement with a previous report [28]. FliGM and FliGCN can be
identified as a single domain, although it is unclear whether the two domains
belong to the same molecule or not because the residues between Gly-186 and
Val-195 are invisible in the crystal (Figures 3A and S3A). The
interaction surface between FliGM and FliGCN is formed by
the C-terminal portion of αB, αC, and αD of FliGM, and
αF, αG, and the N-terminal portion of αH of FliGCN,
respectively (Figure 5A and
B). The interface is highly hydrophobic. Ala-143, Ala-144, Leu-147,
Leu-156, Leu-159, Ile-162, and Ala163 of FliGM, and Ile-204, Met-205,
Leu-208, Ile-216, Leu-220, Leu-227, and Ile-231 of FliGCN are mainly
involved in the tight domain interaction. Leu-159 is located at the center of
the hydrophobic interface (Figure
5C). Around the hydrophobic core, hydrophilic interactions between
Arg-167 and Glu-230, and Gln-155 and Thr-212, also contribute to the domain
interaction (Figure 5C).
These interactions are also conserved in the wild-type Tm-FliGMC and
Aa-FliG crystals, in which FliGM interacts with FliGCN of
an adjacent molecule related by crystallographic symmetry (Figures 3 and S3B). The
FliGM-FliGCN unit in the wild-type
Tm-FliGMC structure can be superimposed onto that in
Tm-FliGMC(ΔPEV) with root mean square deviation of 0.46
Å for corresponding Cα atoms (Figure 4A and C), and that in Aa-FliG with
0.79 Å (Figure 4B and
C). These observations support the idea that the
FliGM-FliGCN unit is a functionally relevant structure
[28]. This is
in good agreement with the previous mutational study showing that most of the
known point mutations that affect FliM-binding [37] are located either on the
bottom surface of the FliGM-FliGCN unit or on the
interaction surface between FliGM and FliGCN (Figure 6A and C).
The default direction of the wild-type flagellar motor of Salmonella
enterica is CCW, and the binding of CheY-P to FliM and FliN increases
the probability of CW rotation. CheY-P binding induces conformational changes in
FliM and FliN that are presumably transmitted to FliG, which directly interacts with
MotA to produce torque [1],[2]. Mutations located in and around helix E FliG, which
connects the FliGM and FliGC domains, generate a diversity of
phenotype, including motors that are strongly CW biased, infrequent switchers, rapid
switchers, and transiently or permanently paused, suggesting that helix E is
directly involved in the switching of the flagellar motor [24]. However, it remains unclear
how helix E affects the switch.
To investigate the switching mechanism, we characterized an extreme CW-biased
S. enterica mutant in which an in-frame deletion of three
residues, Pro-169, Ala-170, and Ala-171, in FliG caused an extreme CW-biased
rotation even in the absence of CheY. Motors containing the FliG(ΔPAA) protein
showed normal torque generation under a wide range of external-load conditions
(Figure 1 and Figure 1S). Thus, the
conformational change in FliG induced by ΔPAA is presumably similar to one
induced by CheY-P binding to FliM and FliN. Limited proteolysis revealed that
ΔPAA induces conformational changes in the hinge between FliGM and
helix E (Figure 2). This result
is in agreement with the crystal structure of Tm-FliGMC(ΔPEV), which
shows that the orientation of helix E relative to FliGM has changed
significantly compared to wild-type FliG (Figure 3).
FliG forms a ring on the cytoplasmic face of the MS ring [17],[18]. In vivo disulfide
cross-linking experiments using Cys-substituted FliG proteins have suggested that
helix A is close to the D–E loop of the adjacent FliG molecule in the FliG
ring [21]. Both a
conserved EHPQR motif in FliGM and a conserved surface-exposed
hydrophobic patch of FliGCN are important for the interactions with FliM
[21]. Because
the conserved charged residues on helix M in FliGCC are responsible for
its interaction with MotA [4],[5],[25], which is embedded in the cytoplasmic membrane, helix M
must lie on top of FliGCC
[21],[28]. Considering those
facts in light of the crystal structure of Tm-FliGMC(ΔPEV) described
here, we propose a new model for arrangement of FliG subunits in the motor (Figures 6 and 7).
In the proposed model, the conserved charged residues on helix M are located on the
top of the FliGM-FliGC unit and the EHPQR motif is present at
the bottom of the unit (Figure 6B and
C). The conserved hydrophobic patch, and most of the point mutation sites
involved in the interaction with FliM, is localized at the bottom of the
FliGMFliGCN units around the EHPQR motif or on the
interface between the FliGM and FliGCN. The D–E loop and
helix E interact with the FliGM domain in the neighboring subunit, in
agreement with data of in vivo cross-linking experiments, which show that residues
117 and 120 (118 and 121 in T. martima) on helix A of one subunit
lie close to residues 166 and170 (167 and 171 in T. martima) on the
D–E loop of the neighboring subunit [21]. In fact, these residues are
very close to each other in our model in positions in which disulfide-crosslinking
should occur. Moreover, the position of Cys residues that do not participate in
disulfide cross-linking are far from each other in the model (Figure 6D).
Our model can also explain the results of mutational studies of CW and CCW-biased
fliG mutants [37],[38]. The mutation sites are widely distributed from helix A
to the H–I loop. Most of them are localized in three regions in our model
(Figure 6A and B). In the
first region, the CCW-biased mutations, which are located on helix A, affect
residues close to residues targeted by CW-biased mutations, which are on a segment
between helix D and E of the adjacent subunit (Figure 6A and B, 1). Because these residues are distributed on the
interaction surface between the neighboring subunits, they presumably affect
cooperative changes in subunit conformation. A second cluster of residues targeted
by CW-biased mutations is located on the C-terminal half of helix B and the
E–F loop (Figure 6A and B,
2). These mutations may
change the orientation of the E–F loop and probably alter the orientation of
helix E, resulting in unusual switching behavior. The third cluster of residues
affected by mutations causing a CW switching bias is located near the loop between
helices H and I (Figure 6A and
B, 3). This region
determines the relative orientation of FliGCC to the
FliGM-FliGCN unit, and therefore the mutations may change the
orientation of FliGCC to cause anomalous switching behavior.
Helix E is directly involved in the switching mechanism, but how does the structure
of helix E affect the orientation of the FliGM-FliGC unit?
Since the D–E loop and helix E interact with FliGM in the
neighboring subunit, we propose that a hinge motion of helix E may directly change
the orientation of the neighboring FliGM domain (Figure 7A). This mechanism could explain the
cooperative switching of the motor. The conformational changes of FliM induced by
association or dissociation of CheY-P may trigger conformational changes in the
FliGM-FliGC unit that it contacts, leading to a large
change in the interaction between FliGCC and MotA. The conformational
change in one unit is probably accompanied by a conformational change in the loop
between FliGM and helix E. This change could influence the orientation of
the neighboring subunit through the interaction between helix E and FliGM
of the neighbor, thereby propagating the conformational change to the neighboring
subunit (Figure 7A).
If helix E actually contacts the more-distant FliGCN in the crystal
structure, an alternative interaction could be responsible for the cooperative
switching (Figure 7B). However,
the same general mechanism involving changes in the conformation of helix E would
still be responsible for the cooperative switching.
Recently, Lee et al. have proposed a model for FliG arrangement and switching based
on the structural differences in Aa-FliG and Tm-FliGMC
[28]. In the crystal
structure of Aa-FliG, the hydrophobic patch in FliGM is covered by the
N-terminal hydrophobic residues of helix E (closed conformation), whereas the patch
is exposed in Tm-FliGMC (open conformation). Because mutations that may
disturb the hydrophobic interaction result in strong CW-bias in motor rotation [38], the
structures of Aa-FliG and Tm-FliGMC are proposed to be in the CCW and CW
states, respectively [28]. The hydrophobic patch is also exposed in the
Tm-FliGMC(ΔPEV) structure, although the conformation of helix E
is different from that of Tm-FliGMC. Since ΔPAA in S.
enterica FliG (ΔPEV in T. maritima) caused an
extreme CW-bias, it is possible that the dissociation of helix E from
FliGM leads to CW rotation. In our model, however, the hydrophobic
patch of the FliGM is covered by the hydrophobic residues in the
C-terminal half of helix E of the adjacent subunit. This arrangement raises the
possibility that the closed conformation of helix E found in the Aa-FliG structure
is an artifact of crystal packing.
Lee et al. assume that the FliGM-FliGC unit is present in the
rotor ring, and hence is in agreement with the results of most of mutational
studies. However, the arrangement of the subunits and the mechanism of switching are
different than in our model. In their model, dynamic motion of helix E and helix n
induces a large conformational change of the FliGM-FliGC unit,
including the rotation of FliGM-FliGCN unit and relative to
the FliGCC to the unit, leading to a change in the arrangement of the
charged residues on helix M (Figure
7C) [28].
Cooperative switching is explained by the strong interaction between
FliGCN of one subunit and FliGCC of the adjacent subunit.
However, helix A of one subunit and the D–E loop of the adjacent subunit are
always at a considerable distance in both the CW and CCW states. Hence, their model
cannot explain the in vivo disulfide cross-linking experiments (Figure 7C) [21]. Since our new model can
explain the cross-linking data, it appears to be more plausible than the model
proposed by Lee et al. [28].
Although our model is consistent with most of the previous experimental data, it
still contains ambiguity. The available density map of the basal body obtained by
electron cryo-microscopy is not high enough to allow fitting of the atomic model.
Thus, a higher-resolution rotor-ring structure will be required to build a more
precise model to explain the molecular mechanism of directional switching.
S. enterica strains and plasmids used in this study are listed
in Table 1. L-broth, soft
agar plates, and motility media were prepared as described [39],[40]. Ampicillin was added to
a final concentration of 100 µg/ml.
Fresh colonies were inoculated on soft tryptone agar plates and incubated at
30°C.
Bead assays were carried out using polystyrene beads with diameters of 0.8, 1.0,
and 1.5 mm (Invitrogen), as described before [8]. Torque calculation was
carried out as described [8].
Cultures of S. enterica cells grown at 30°C were centrifuged
to obtain cell pellets. The cell pellets were resuspended in SDS-loading buffer,
normalized in cell density to give a constant amount of cells. Immunoblotting
with polyclonal anti-FliG antibody was carried out as described [41].
His-FliG and His-FliG(ΔPAA) were purified by Ni-NTA affinity chromatography
as described before [39]. His-FliG and its mutant variant (0.5 mg/ml) were
incubated with trypsin (Roche Diagnostics) at a protein to protease ratio of
300∶1 (w/w) in 50 mM
K2HPO4-NaH2PO4 pH 7.4 at room
temperature. Aliquots were collected at 0, 5, 15, 30, 60, 90, and 120 min and
trichloroacetic acid was added to a final concentration of 10%. Molecular
mass of proteolytic cleavage products was analyzed by a mass spectrometer
(Voyager DE/PRO, Applied Biosystems) as described [42]. N-terminal amino acid
sequence was done as described before [42].
Tm-FliGMC(ΔPEV) was purified as described previously [23]. Crystals of
Tm-FliGMC(ΔPEV) were grown at 4°C using the hanging-drop
vapor-diffusion method by mixing 1 µl of protein solution with 1 µl
of reservoir solution containing 0.1 M sodium phosphate-citrate buffer pH
4.2–4.4, 36%–50% PEG200, and 200 mM NaCl. Initially,
we tried to solve the structure by the molecular replacement method using
Tm-FliGMC structure (PDB ID: 1 lkv) as a search model. However,
no significant solution was obtained, even though individual domains were used
as search models. Therefore, we prepared heavy-atom derivative crystals and
determined the structure using the anomalous diffraction data from the
derivatives.
Derivative crystals were prepared by soaking in a reservoir solution containing
K2OsCl6 at 50% (v/v) saturation for one day.
Crystals of Tm-FliGMC(ΔPEV) and its Os derivatives were soaked in
a solution containing 90%(v/v) of the reservoir solution and
10%(v/v) 2-Methyl-2,4-pentanediol for a few seconds, then immediately
transferred into liquid nitrogen for freezing. All the X-ray diffraction data
were collected at 100 K under nitrogen gas flow at the synchrotron beamline
BL41XU of SPring-8 (Harima, Japan), with the approval of the Japan Synchrotron
Radiation Research Institute (JASRI) (Proposal No. 2007B2049). The data were
processed with MOSFLM [43] and scaled with SCALA [44]. Phase calculation
was performed with SOLVE [45] using the anomalous diffraction data from
Os-derivative crystals. The best electron-density map was obtained from MAD
phases followed by density modification with DM [44]. The model was
constructed with Coot [46] and was refined against the native crystal data to
2.3 Å using the program CNS [47]. About 5% of the
data were excluded from the data for the R-free calculation. During the
refinement process, iterative manual modifications were performed using
“omit map.” Data collection and refinement statistics are summarized
in Tables
S1 and S2, respectively.
|
10.1371/journal.pgen.1005855 | Synergistic Control of Kinetochore Protein Levels by Psh1 and Ubr2 | The accurate segregation of chromosomes during cell division is achieved by attachment of chromosomes to the mitotic spindle via the kinetochore, a large multi-protein complex that assembles on centromeres. The budding yeast kinetochore comprises more than 60 different proteins. Although the structure and function of many of these proteins has been investigated, we have little understanding of the steady state regulation of kinetochores. The primary model of kinetochore homeostasis suggests that kinetochores assemble hierarchically from the centromeric DNA via the inclusion of a centromere-specific histone into chromatin. We tested this model by trying to perturb kinetochore protein levels by overexpressing an outer kinetochore gene, MTW1. This increase in protein failed to change protein recruitment, consistent with the hierarchical assembly model. However, we find that deletion of Psh1, a key ubiquitin ligase that is known to restrict inner kinetochore protein loading, does not increase levels of outer kinetochore proteins, thus breaking the normal kinetochore stoichiometry. This perturbation leads to chromosome segregation defects, which can be partially suppressed by mutation of Ubr2, a second ubiquitin ligase that normally restricts protein levels at the outer kinetochore. Together these data show that Psh1 and Ubr2 synergistically control the amount of proteins at the kinetochore.
| As cells divide, their replicated chromosomes must be correctly allocated to the two nascent daughter cells. This is achieved by the kinetochore, which provides a physical link between the chromosomes and the microtubules that drive their movement. If chromosome separation fails, the resulting cells have an abnormal number of chromosomes. This state is called aneuploidy and is a hallmark of cancer cells. The regulation of the kinetochore is therefore of critical importance in maintaining genome integrity. Since a number of cancer cells have over-active kinetochore genes, it has been proposed that an excess of kinetochore proteins can disrupt the normal assembly or maintenance of kinetochores. We tested this idea in yeast by increasing the amount of a specific kinetochore protein, but found no effect upon the normal loading of kinetochore proteins. Instead, we find that two ubiquitin ligases play a role in maintaining the normal balance of the different kinetochore proteins and that this correlates with correct segregation of the chromosomes.
| Accurate chromosome segregation is necessary for the equal distribution of genetic material between daughter cells during cell division and is achieved by kinetochores which link chromosomes to spindle microtubules [1]. Perturbations of kinetochore function result in aneuploidy, i.e. changes in chromosome number, and genome instability [2, 3]. Thus kinetochore regulation is of critical importance in replicating cells. A number of different cancers overexpress kinetochore genes [4, 5] leading to the notion that disrupting kinetochore stoichiometry and regulation may be a driver of aneuploidy and genomic instability.
Budding yeast is a key model to study kinetochore composition and assembly because of its comparatively simple structure; there is only one microtubule attachment per chromosome and per kinetochore [6, 7]. Kinetochores are composed of more than 60 proteins organized into various sub-complexes that are thought to assemble hierarchically initiating at the centromeres [1]. The inner part of the kinetochore mediates centromere binding whereas the outer part mediates microtubule binding. Kinetochore structure and composition is remarkably well conserved from yeast to humans [8].
In budding yeast the position of the centromeres is sequence specific. Cbf1 and the CBF3 complex associate to centromere DNA elements (CDE), CDEI and CDEIII, respectively [9–13]. The CDEII region wraps around the centromeric nucleosome that contains the centromeric histone H3 variant CENP-A (Cse4 in budding yeast) [14–17]. Mif2 (CENP-C) and the COMA complex mediate the association between centromere and outer kinetochore. Mif2 binds to both the Cse4 nucleosome and the outer kinetochore MIND complex [18–20]. The COMA complex proteins Okp1 and Ame1 form a dimer that binds directly to DNA and the MIND complex [20, 21].
The outer kinetochore mediates interactions with microtubules emanating from opposite spindle pole bodies. The yeast homologues of the KNL1/ MIS12/ NDC80 network (KNM) are the essential complexes SPC105, MIND and NDC80, respectively [1]. The MIND complex is composed of two heterodimers: Mtw1-Nnf1, which associates with both Mif2 and the COMA complex, and Dsn1-Nsl1, which associates with the NDC80 complex [21, 22]. Both the NDC80 complex and the yeast-specific DAM-DASH complex, which may play an orthologous function to the human SKA proteins [23], bind to microtubules in a cooperative process [24, 25].
Although the centromeric DNA sequence (CEN) is essential to assemble kinetochores, protein degradation has been shown to be important to control cellular levels of various kinetochore proteins. The E3 ubiquitin ligase Psh1 restricts the localization of Cse4 to centromeres [26]. Psh1 localizes to centromeres throughout the cell cycle, and its destabilizing role is opposed by the Cse4 chaperone Scm3 [27, 28]. Levels of Cse4 are increased in psh1Δ cells [26] and these cells have a chromosomal instability phenotype [29]. More recently, the E3 ubiquitin ligase Ubr2 has been shown to control levels of the MIND complex protein Dsn1 [30]. Thus kinetochore assembly may be regulated differently from steady state homeostasis. Surprisingly, yeast kinetochores can assemble in reverse from the microtubule interface back to the inner kinetochore as shown via artificial recruitment of proteins to DNA [31]. In this situation, the conserved yeast centromere is not necessary, although inner kinetochore proteins are required [32]. These data point to a kinetochore with more flexibility in its assembly and stoichiometry than was previously assumed.
Numerous studies in budding yeast have revealed the stoichiometry of the various protein sub-complexes forming the kinetochore [20, 21, 33–37]. It is thought that the kinetochore assembles hierarchically from the centromere [37]. However, little is known about how these sub-complexes assemble to form the kinetochore in vivo and how much flexibility exists in kinetochore composition. To investigate this, we tested how increased levels of kinetochore proteins affect kinetochore composition. We used fluorescence microscopy to quantify the levels of proteins at kinetochore foci. We found that Mtw1 levels at the kinetochore correlate with chromosome number and they are not transcriptionally controlled. Moreover, we found that psh1Δ mutants, in addition to the elevated Cse4 protein, have increased levels of inner kinetochore proteins but not outer kinetochore proteins. However, the levels of outer kinetochore proteins are increased in the psh1Δ ubr2Δ double mutant, in which both Cse4 and Dsn1 are unconstrained. Finally, we found that ubr2Δ suppresses psh1Δ mitotic and meiotic defects. These findings are consistent with multiple regulatory pathways acting independently on the different kinetochore complexes.
To investigate whether we could perturb kinetochore homeostasis by overexpression of kinetochore genes, we chose to study MTW1. Mtw1 forms part of the essential MIND complex [21, 38] and the levels of one of these proteins, Dsn1, is controlled via phosphorylation status and subsequent ubiquitylation by the E3 ligase, Ubr2 [30]. We used an ectopically-expressed plasmid-encoded version of Mtw1 to elevate the levels of Mtw1 within the cell and assessed the recruitment of Mtw1 to kinetochores by fluorescence imaging. The plasmid is a single copy CEN plasmid and its MTW1 gene is driven by a constitutively-active copper promoter (CUP1) [39]. We used differential fluorescence tagging of endogenously-encoded and plasmid-encoded Mtw1 to differentiate between and quantitate the proteins loaded into kinetochores (Fig 1A, 1B and 1C). The MTW1 plasmid produced significant ectopic expression as judged by loading of plasmid-encoded Mtw1 at the kinetochore (Fig 1A). We quantified the levels of fluorescence at kinetochores using Volocity image analysis software. In brief, the mean fluorescence within a 3-dimensional spherical region around each kinetochore was assessed and a background region around each kinetochore was also measured by dilating each kinetochore selection (Fig 1E). Each background measurement was subtracted from each kinetochore measurement to produce a relative value representing the levels of fluorescence signal from the kinetochore. When we expressed an ectopic MTW1-CFP gene in cells containing MTW1-YFP at the endogenous locus, we found that the resulting fluorescence at kinetochores was approximately 50% of the haploid CFP signal and 50% of the haploid YFP signal (Fig 1B). This is consistent with an approximately equal contribution of the two proteins to the kinetochore, but not consistent with an elevation of Mtw1 loading at the kinetochore. To determine whether one fluorescent tag is preferred over the other, we then performed the same analysis but with the tags reversed i.e. ectopic MTW1-YFP and endogenous MTW1-CFP. In this case the levels of the plasmid encoded Mtw1-YFP at the kinetochore are somewhat higher than the CFP signal, although both still contribute to the kinetochore signal (Fig 1B). Again, no increase in total kinetochore fluorescence was measured. We also examined the effect of deleting the endogenous MTW1 gene in cells containing an MTW1-YFP plasmid. The level of YFP fluorescence in this stain is the same as an endogenously-encoded MTW1-YFP strain, (Fig 1B). Finally, we transformed the MTW1-YFP plasmid into an untagged strain. We find that the Mtw1-YFP level of fluorescence is equivalent to the strain with both endogenously and ectopically-encoded Mtw1, approximately 50% (Fig 1B). We also assessed whether changes in the background levels of fluorescence in the cells over-expressing kinetochore proteins were increased, resulting in an artificially low kinetochore signal. However, we find that changes to background fluorescence do not mask an effect of MTW1 expression on kinetochore protein levels (S1A and S1B Fig). Thus, these quantitative data support the notion that the fluorescently tagged proteins compete for inclusion into the kinetochore and that the total levels of kinetochore Mtw1 remain constant. There are two likely reasons for this homeostasis of Mtw1 at the kinetochore. First, an uncharacterised negative feedback mechanism could limit transcription, translation or protein stability of the endogenous Mtw1, thus maintaining a steady state level of Mtw1 protein within the cell. Second, the loading of Mtw1 onto the kinetochores is limiting, such that there is a strong affinity to load Mtw1 as part of the MIND complex but once the protein reaches a threshold level (perhaps through stoichiometric interaction with other kinetochore components), no more Mtw1 is loaded. To discriminate between these two ideas we used western blotting to assess the total cellular levels of Mtw1. We find that the ectopic expression of MTW1 causes an increase in the levels of Mtw1 protein in the cell (Fig 1F). Thus, we exclude the possibility that total Mtw1 protein levels are tightly regulated by translation or protein stability.
Our results are also consistent with the notion of hierarchical assembly of the kinetochore building up from inner kinetochore components such as Cse4. To test this notion we compared the loading of Mtw1 in diploid strains with MTW1-YFP at either one or two of the endogenous MTW1 alleles. We find that diploid kinetochore Mtw1 levels are approximately double that of haploids and heterozygous mtw1Δ/MTW1-YFP strains compensate by loading equivalent Mtw1 as diploid strains (Fig 1D). We note here that these heterozygous mtw1Δ/MTW1-YFP strains are haplo-sufficient in that they do not show sensitivity to microtubule poison drug benomyl (S2B Fig). We also confirmed that overexpression of MTW1 does not render cells sensitive to benomyl (S2C Fig), nor does it affect cell cycle progression (S3A Fig), plasmid loss (S3B Fig), or chromosome segregation (S3C and S3D Fig). We also checked whether MTW1 overexpression resulted in changes to the levels of other kinetochore proteins and consistent with the levels of Mtw1, we find no change in Dsn1 or Ndc80 (S3E and S3F Fig). In order to test more generally the effects of high levels of kinetochore proteins, we expressed various inner and outer kinetochore proteins from a CEN plasmid under the control of a CUP1 promoter. Only NDC10 overexpression showed a reduced growth in the presence of benomyl (S4 Fig) We then tested whether Mtw1 kinetochore levels were affected by the deletion of genes encoding several inner kinetochore components: the DNA-binding protein Cbf1, the Monopolin complex components Mam1 and Csm1, and the COMA complex component Ctf19. We found no change in Mtw1 levels in any of these mutants (S5A and S5B Fig), consistent with Mtw1 loading hierarchically based upon the number of centromeres present in the cell.
The hierarchical loading model is consistent with the hypothesis that the loading of inner kinetochore proteins is critical for determining kinetochore stoichiometry as a whole. To test this idea we decided to attempt to manipulate the levels of an inner kinetochore protein to test whether the MIND complex is regulated in parallel.
The levels of the inner kinetochore protein Cse4 are controlled in part by degradation via an ubiquitylation-dependent degradation pathway. Psh1 was identified as the E3 ubiquitin ligase responsible for restricting Cse4 levels at the kinetochore [26, 27]. In a psh1Δ strain Cse4 levels are elevated and furthermore overexpression of the CSE4 is lethal in psh1Δ cells, consistent with a failure to constrain Cse4 loading [26, 27]. We used the same fluorescence quantitation method described above to compare endogenous kinetochore protein levels of wild-type cells with those of psh1Δ cells. Consistent with previous studies we find that psh1Δ cells have elevated levels of Cse4 at kinetochore foci, although with considerable heterogeneity between cells (Fig 2A). We found no change in the protein levels of the inner kinetochore protein Ndc10 (Fig 2B). In addition, we find that Mif2, the ortholog of human CENP-C, (Fig 2C) and members of the Ctf19/COMA complex are also elevated in the psh1Δ (Fig 2D, 2E and 2F). However, contrary to our expectation Mtw1 kinetochore levels are unchanged in a psh1Δ strain compared with wild type (Fig 2G). We therefore examined whether other outer-kinetochore complexes are affected by deletion of PSH1. Like Mtw1, the kinetochore levels of Ndc80 and Ask1 (a member of the decameric DAM1/DASH complex) are both unaffected in psh1Δ cells (Fig 2H and 2I). These data show that although Cse4 levels may influence the inner kinetochore, the protein levels of the entire kinetochore are not affected. This result shows that for the fluorescence focus that is widely considered to represent the structural kinetochore the stoichiometry is not fixed.
One possible reason for the non-stoichiometric increase in kinetochore protein levels in psh1Δ cells is that the increased Cse4, Ctf19 etc. are not part of the canonical kinetochore structure, but rather represent a pericentromeric ‘cloud’ of protein. There is precedent for this from fluorescence studies of Cse4 [40, 41]. We therefore re-analysed our images to evaluate the size each of the fluorescence foci. The rationale is that pericentric protein recruitment will result in a larger area of fluorescence, which can be measured by fitting a Gaussian distribution to the kinetochore foci (Fig 3A). We find that psh1Δ Cse4 foci are considerably larger than WT, consistent with the notion of a cloud of pericentric Cse4 and this is rescued by overexpressing PSH1 (Fig 3B and 3C). However, the other kinetochore proteins had psh1Δ foci comparable in size to WT cells (Fig 3C–3K). We cannot say for sure that protein that is located in a comparably-sized focus is part of a structural complex, it is possible that for certain proteins the kinetochore can accommodate additional proteins within the confines of the WT diffraction limited region.
We next asked whether the effect of Psh1 upon kinetochore protein levels would function in synergy with the Mub1/Ubr2 ubiquitylation pathway. The MIND complex member Dsn1 is ubiquitylated by the E3 ubiquitin ligase Ubr2 [30]. Dsn1 contains two AuroraB (Ipl1) phosphorylation sites (serines 240 and 250) and versions of Dsn1 that cannot be phosphorylated at these residues are ubiquitylated and degraded [30, 42]. Such a mechanism may restrict the levels of MIND proteins even in the presence of excess inner kinetochore proteins. Since psh1Δ, ubr2Δ and the double mutant cells are all viable we were able to assess their relative contribution to the kinetochore focus fluorescence levels. We find that UBR2 deletion has no effect upon inner kinetochore protein levels of Cse4 or Ndc10. Cse4 levels are elevated by PSH1 deletion, but not further affected by the additional deletion of UBR2 (Fig 4A). Also addition of ubr2Δ mutation did not further increase the size of Cse4-GFP foci (S6A Fig). Ndc10 is unaffected by either of these mutants (Fig 4B). Mif2 is elevated in a psh1Δ mutant, but unaffected by further deletion of UBR2 (Fig 4C). The MIND complex shows little change in either of the single mutants but both Mtw1 and Dsn1 are modestly elevated in the double psh1Δ ubr2Δ strain (Fig 4D and 4E). The size of Mif2 and Dsn1 foci was unaffected in the ubr2Δ and in the double psh1Δ ubr2Δ cells (S6B and S6C Fig). Another MIND complex protein Nnf1 is also elevated in psh1Δ ubr2Δ cells (Fig 4F). Other outer kinetochore proteins Spc105, Spc24, from NDC80 complex, and Ask1 were unaffected by either of these mutants (Fig 4G, 4H and 4I). The degradation of Dsn1 is controlled by phosphorylation/ dephosphorylation of serines 240 and 250. The double dsn1-S240A,S250A mutant is inviable, but can be rescued by either its overexpression or by deleting UBR2 [30]. We reasoned that if increased Dsn1 was responsible for the MIND phenotype, this should be epistatic with a dsn1-S240D,S250D mutant, which would be hyper-stable. However, we find that the elevated levels of Mtw1 in a psh1Δ ubr2Δ mutant are increased further when the two Dsn1 serines are changed to aspartic acid (Fig 5A and 5B). Furthermore, we examined cellular levels of both Mtw1 and Dsn1 in psh1Δ, ubr2Δ and the psh1Δ ubr2Δ mutants and find that these are comparable with wild-type cells (S6D and S6E Fig) These data suggest that Ubr2 plays additional, potentially indirect, roles in regulating the levels of kinetochore components in addition to its function on dephosphorylated Dsn1 or that there are other mechanisms to remove dephosphorylated Dsn1 from kinetochores. These data also strengthen our observation that the stoichiometry of the various kinetochore sub-complexes is not fixed in these mutants.
Although these ubiquitin ligase mutants affect kinetochore protein levels, they are all viable and the cells appear to grow normally [26, 30]. Since there is considerable interest in the possibility that altered kinetochore protein levels would lead to kinetochore dysfunction and the resulting aneuploidy [4, 5, 43], we asked whether the psh1Δ and ubr2Δ mutants affected the mitotic or meiotic phenotype of yeast. We did not find strong defects in cell cycle progression, although S-phase was slightly faster in ubr2Δ and psh1Δ ubr2Δ mutants (S7 Fig). It has previously been reported that ubr2Δ mutants have an enhanced sporulation phenotype [44]. Consistent with this we found that the sporulation of homozygous ubr2Δ mutants is enhanced compared with wild-type diploids (Fig 6A). Addition of the psh1Δ mutant did not modify this phenotype. In all cases spore viability was similar (Fig 6B). We tested whether the increase in Mtw1 kinetochore levels in psh1Δ ubr2Δ mitotic cells (Fig 3D) was recapitulated in meiosis. Diploid cells were induced to sporulate and arrested in pachytene, prior to the two meiotic divisions by depletion of the Ndt80 transcription factor. Then, meiosis I was triggered by induction of NDT80 expression from the GAL1-10 promoter [45] (see Materials and Methods for details). We found elevated Mtw1 kinetochore levels in psh1Δ ubr2Δ in meiosis I, and to a lesser extent in meiosis II (Fig 6C and 6D).
As Psh1 is known to have a role in maintaining chromosome stability [29], we used an assay for homozygosity of chromosome III [2, 3, 29] to analyse the rate of chromosomal instability (CIN) in diploids cells, and we also tested the rate of loss of a CEN plasmid. Consistent with previous reports, we find that psh1Δ cells show elevated rates of both chromosome III loss (Fig 7A) and CEN plasmid loss (Fig 7B), whereas ubr2Δ cells are unaffected. Surprisingly, we found that the addition of ubr2Δ to a psh1Δ mutant leads to a reduction of these CIN phenotypes (Fig 7A and 7B). To investigate the effect of the ubiquitin ligases Psh1 and Ubr2 on checkpoint function, we assessed the synthetic effects of combining mutations in these genes with those of checkpoint genes. We deleted the MAD1 gene, which encodes a protein required for the activation of Mad2 [46] and also MAD3, which encodes a key member of the mitotic checkpoint complex [47]. These mutants were combined with psh1Δ, ubr2Δ or the double mutant. The resulting strains were all viable (Fig 8), so to test their checkpoint proficiency we grew them in the microtubule poison benomyl. We found that deletion of psh1Δ decreases the ability of both mad1Δ and mad3Δ to grow in the presences of benomyl (Fig 8). Moreover, deletion of ubr2Δ partially rescued the ability of mad1Δ and mad3Δ to grow on benomyl. Finally, we also found that ubr2Δ partially rescues the benomyl sensitivity of mad1Δ psh1Δ and mad3Δ psh1Δ double mutants (Fig 8). We then tested if increased Dsn1 levels could explain the rescue of ubr2Δ. However, we found that DSN1 over-expression from a CUP1 promoter did not rescue benomyl sensitivity (S8 Fig)
A number of studies have shown correlation between the overexpression of kinetochore genes and tumorigenic status [4, 5, 43]. These observations raise the possibility that increased levels of kinetochore proteins result in aberrant kinetochore function, which then leads to chromosomal instability. We wished to test the idea that overexpression of kinetochore genes would affect kinetochore protein loading. We overexpressed the kinetochore gene, MTW1 that encodes a core member of the outer kinetochore MIND complex. The MIND complex plays an essential role in linking the inner kinetochore and the outer kinetochore [48, 49]. Using quantitative fluorescence imaging we find that although overexpression of MTW1 does lead to increased Mtw1 protein in the cell, the loading of Mtw1 onto the kinetochores is unaffected (Fig 1). Our data supports the idea that kinetochores are assembled hierarchically from the inner kinetochore, likely directed by Cse4 inclusion into centromeric nucleosomes [37]. Similarly, Aravamudhan and colleagues found that the levels of Cse4 at the kinetochore did not change after increasing total Cse4 cellular levels in budding yeast [50]. The effects of kinetochore gene overexpression may be subtle and/or different in mammalian cells, however, our data do not support the idea that kinetochore gene overexpression would, a priori, lead to a kinetochore defect (Figs 1, S2–S4). On the contrary, our data also support the idea that the kinetochore focus represents the structural assembly of kinetochore proteins loaded onto centromeres [37, 51] and that kinetochore protein levels scale with centromere number (Fig 1) [52]. However, recent work using synthetic kinetochores has demonstrated that a functional kinetochore can assemble backwards from the microtubule associated DAM1/DASH complex [31, 32]. Recruitment of outer kinetochore proteins to a non-centromere sequence is sufficient to generate an artificial kinetochore that no longer requires a specific CEN sequence but does require inner kinetochore proteins. These observations challenge the hierarchical assembly model, albeit in an artificially tethered system and suggest that the kinetochore structure may be more adaptable than previously imagined.
In an effort to perturb the kinetochore structure we examined kinetochores in mutants of two ubiquitin ligases that are known to affect the degradation of kinetochore proteins, Psh1 and Ubr2. The Psh1 ubiquitin ligase regulates the levels of Cse4 protein at the kinetochore focus [26, 27]. We confirmed that the levels of Cse4 are increased in psh1Δ cells, and additionally found that the levels of inner kinetochore proteins Mif2, Okp1, Ame1 and Ctf19 also increase (Fig 2). The increase in kinetochore-loaded Cse4 was higher than the other inner kinetochore proteins, suggesting that some of the excess Cse4 is not able to recruit these additional proteins and maybe part of a local ‘cloud’ of Cse4 adjacent to the kinetochore [40] or that it is in a form that is unable to recruit the other components. Consistent with the former notion, we find that the increased Cse4 in a psh1Δ mutant is spread over a larger area, although this is not true for all kinetochore proteins that are elevated in psh1Δ cells (Fig 3). This may explain why a large increase in Cse4 levels results in only a modest increase in, for example, members of the COMA complex. Surprisingly, we found that outer kinetochore protein levels are unaffected in psh1Δ cells (Fig 2). These data support the idea that in these mutants the stoichiometry of the kinetochore is flexible. We found that mutating both PSH1 and UBR2 is sufficient to modestly increase the levels of members of the MIND complex (Fig 4). In budding yeast, if we assume two Cse4 molecules per centromere, there are about 6–7 MIND complexes per kinetochore in anaphase [7, 53]. In the psh1Δ ubr2Δ double mutants, the ~ 30% increase of Mtw1 and Dsn1 would correspond to ~2 additional MIND complexes per kinetochore. It is unlikely that the chromosome instability phenotype found in psh1Δ and psh1Δ ubr2Δ (Fig 7) accounts for the difference in kinetochore protein levels (Fig 2 and Fig 4). If these mutant cells would have a higher number of chromosomes (due to their CIN phenotype), we would expect all kinetochore components to be similarly increased. Instead, we find no change in Ndc10 protein levels in the absence of Psh1, Ubr2 or both (Fig 2 and Fig 4), and we also did not find an increase in the outer kinetochore proteins in psh1Δ cells. It is possible that the additional proteins at the kinetochore focus in psh1Δ and psh1Δ ubr2Δ are not part of the structural kinetochore assembly. However, the magnitude of the increase of Mtw1 and Dsn1 in the psh1Δ ubr2Δ double mutant (Fig 4) is similar to the increase in Mif2 and COMA complex proteins in the psh1Δ mutant (Fig 2). This suggests that the amount of MIND complex binding to the kinetochore is still limited by the amount of inner kinetochore components, consistent with a hierarchical kinetochore assembly. The double psh1Δ ubr2Δ mutant does suppress some characteristics of the psh1Δ phenotype; including meiotic sporulation defects (Fig 6) and mitotic genome instability (Fig 7). It is possible that partially restoring the stoichiometry between inner and outer kinetochore proteins contributes to this phenotypic suppression. However, it is important to note that there is no evidence that the increased Cse4 levels at the kinetochore in psh1Δ cells cause their CIN phenotype. Collectively our data show that inclusion of kinetochore proteins into the kinetochore focus is flexible in mutant backgrounds. Furthermore, that the genomic instability of psh1Δ cells, which may result from increased Cse4 loading, is suppressed by second mutation, ubr2Δ, that also increases the levels of MIND complex members.
In psh1Δ cells, Cse4 is increased at kinetochore foci (Fig 2) and also deposited ectopically in non-centromeric regions [26, 27]. Both kinetochore and non-kinetochore ectopic pools of Cse4 could contribute to psh1Δ chromosomal instability phenotype [29] (Fig 6). The negative interaction of psh1Δ with spindle assembly checkpoint components mad1Δ and mad3Δ in the presence of microtubule poison (Fig 8) suggests a decreased kinetochore function in psh1Δ. Surprisingly, ubr2Δ partially rescued benomyl sensitivity of both mad1Δ and mad3Δ also in combination with psh1Δ (Fig 8). This ubr2Δ suppressor effect was not recapitulated by DNS1 overexpression (S8 Fig), suggesting an additional role of Ubr2. It is possible that the upregulation of other Ubr2/Mub1 complex targets, such as Rpn4 [54] and Sml1 [55], contribute to the suppression of mitotic and meiotic phenotypes of ubr2Δ.
Ubr2 has been previously shown to reduce Dsn1 protein stability by ubiquitylation [30], but the impact of Ubr2 in kinetochore composition was not known. Ipl1 phosphorylation on Dsn1 promotes the interactions of the MIND complex with the inner kinetochore proteins [42]. However, the presence of dsn1-S240D/S250D did not increase Mtw1 kinetochore levels in wild type or psh1Δ cells, but only in psh1Δ ubr2Δ double mutant and slightly in ubr2Δ (Fig 5). Our data suggest an important role of Ubr2 on limiting outer kinetochore loading by restricting MIND complex availability (Figs 4 and 5). From our data, we cannot be sure whether the changes in kinetochore protein levels are a direct result of changes in ubiquitylation status of kinetochore proteins, the effects may be indirect. We note that the artificial recruitment of Ubr2 and Mub1 to kinetochores does not cause a growth defect [56]. Our data also show that Ubr2 is upstream of Ipl1 in the regulation of outer kinetochore assembly (Fig 5).
Regardless of the mechanism of action of Psh1 and Ubr2, the flexibility of kinetochore stoichiometry may have some functional significance. Kinetochore components are remarkably well conserved from S. cerevisiae to H. sapiens although the centromeres to which they bind are highly divergent both in length and sequence. It is hard to imagine that an inflexible kinetochore structure would be sufficient to support the rapid evolution that is typically seen for centromere sequences [57, 58]. Our data in yeast show that overexpression of the kinetochore gene MTW1 is not sufficient to disrupt kinetochore function, however this may not be true for all kinetochore genes or in nascent tumor cells. This is further supported by the observation that overexpression of CSE4 is not lethal without further perturbations to the kinetochore [26, 27, 59].
Yeast strains used in this study are either W303 or S288C background, as indicated in S1 Table. For plasmid construction (see S2 Table), the SPC42-RFP sequence containing 200 bp of the SPC42 promoter was cloned into pX29 plasmid (CEN6, LEU2, CUP1 promoter). Then, YFP (pHT5), CFP (pHT222), MTW1-YFP (pHT15) or MTW1-CFP (pHT223) were cloned downstream of the CUP1 promoter by gap repair. A sequence encoding four alanine residues was used as a linker between MTW1 and the fluorescent tags, and between SPC42 and RFP. Plasmids were transformed into appropriate strains by lithium acetate transformation and continuously selected in synthetic media lacking leucine.
MTW1, PSH1 and UBR2 genes were disrupted by transforming with PCR products containing either MX6-KAN or MX6-NAT selection cassettes flanked with ~250 bp of sequences upstream and downstream the corresponding coding regions. Gene deletions were confirmed by PCR. Since MTW1 is an essential gene, it was disrupted in a haploid strain harbouring CUP1-pMTW1-YFP::LEU2 plasmid (pHT15). Transformants were selected in synthetic media lacking leucine and containing G418 and confirmed by PCR. Diploid strain MTW1-YFP/MTW1-CFP (PT11) was transformed using mtw1Δ::KANMX PCR to obtain heterozygous diploids MTW1-YFP/mtw1Δ::KANMX (PT69 and PT70). Loss of CFP or YFP kinetochore foci was tested by fluorescence microscopy and insertion of the KANMX cassette at one of the MTW1 locus was confirmed by PCR.
For microscopy and western blot analysis cells were grown in synthetic complete (SC) or lacking leucine SC–LEU media supplemented with 100mg/ml of adenine (+ADE, 100 mg/mL). Yeast strains were grown overnight at 23°C. Cultures were diluted in fresh media to ≈ OD600 0.3 and grown for 3 hours before imaging or protein extraction.
Cells from log-phase cultures were mounted on microscope slides with 0.7% LMP agarose in SC +ADE or SC-LEU +ADE, and covered with 0.17 mm glass coverslips. Our microscope system uses a Zeiss AxioImager Z2 microscope, 63X Plan Apo, 1.4NA, oil immersion objective and a Hamamatsu CCD ORCAII camera (2X2 binning and maximum analog gain). The resulting pixel size was 0.205 μm. Excitation light was provided by LED Colibri system (excitation band-pass filter): CFP 445 nm (445/25), YFP 505 nm (510/15), GFP 470 nm (474/28) and RFP 590 nm (585/35). Emission band-pass filters were as follows: CFP 47HE (480/40), YFP 46HE (535/30), GFP 38HE (525/50), and RFP 63HE (629/62). Exposure times were optimized for each fluorescent protein and ranged from 100 to 250ms. Z stacks consisted of 17 vertically separated slices with 0.4 μm spacing. The theoretical dynamic range of our system is ~3000 levels of brightness, however, in practice this will be somewhat lower.
A custom-made protocol in Volocity software was used to quantify fluorescence intensity at kinetochore foci. The protocol finds the brightest spots in the image. Spots within 3 pixels from x,y,z edges of the image were removed from the analysis. A 3D box was drawn concentric to the brightest pixels (1.36 μm3). The background region was 2 pixels separated from the kinetochore box (23.51 μm3). Average intensity of the background was subtracted from average kinetochore intensity to obtain the final fluorescence value. Finally, fluorescence values were normalized to the average of wild type or control populations. For quantitation, only post-anaphase kinetochores of dividing cells were selected.
To measure the size of individual kinetochore foci we fit two Gaussian distributions to each kinetochore. A five pixel square box was selected for each kinetochore and a local background subtracted. The pixel values in each column and each row were summed and for both the rows and columns and then we used ImageJ’s fitDoFit function to fit a Gaussian curve to the values, separately both the rows and columns (Fig 3A). The two values for the full width at half maximum (FWHM), vertical and horizontal Gaussian fits, were averaged to give a mean FWHM measurement for each focus. The mean FWHM measurements for each experiment were normalized relative to the level in WT cells.
Cell were harvested by centrifugation and resuspended in 1.5X Laemmli buffer with protease inhibitors (Roche) and transferred to a fresh tube containing 0.5 mm glass beads. Cells were disrupted with a cell homogenizer. Cells extracts were harvested into a fresh tube and boiled for 5 minutes. Cells debris was pelleted and 20 μL of the protein extracts were loaded in a 12% acrylamide gel (Biorad). Proteins were transferred into a PVDF blotting membrane (GE Healthcare Amersham). The western blot was performed with monoclonal anti-GFP antibody (Roche), anti-PGK1 (Invitrogen), goat anti-mouse HRP antibody (Abcam), and ECL kit (GE Healthcare Amersham).
Yeast strains were grown o/n at 30°C in YPD or selective media. Cultures were adjusted to OD600 = 1, serially diluted and spotted into YPD or selective media plates with 0.2% DMSO and 10–15 μg/ml benomyl. For testing effects of overexpression increasing concentrations of CuS04 were added to the media as indicated. Plates were incubated for 2 days at 30°C before images were captured.
Diploid strains were grown in YPD at 23°C for 24 hours. Then, cultures were diluted 100X in YEPA media and grown at 23°C until OD600 reached 0.6 (2X107 cells/ml). Cultures were washed once with water, resuspended in SPO media and incubated at 23°C for 3 days. Four independent cultures were tested for each genotype. To test spore viability, 22 tetrads per genotype were dissected in YPD and grown for 2 days at 30°C.
Diploid strains were grown in YPD for 24 hours at 30°C. Cultures were diluted to OD600 0.3 in YPA (1% yeast extract, 2% Bacto-peptone, 1% potassium acetate) and grown for 12–15 h at 30°C. Cells were then resuspended in sporulation media (1% potassium acetate pH7) at 23°C for 12 hours. Finally, 1μM β-estradiol (Sigma) was added to induce NDT80 expression. Cells were imaged every hour to follow meiotic divisions.
MATa strains lacking the Bar1 protein were used to facilitate α-factor G1 synchronization. Strains were grown overnight at 30°C, diluted to OD600 = 0.3 and grown for 1 hour. The asynchronous sample was collected at this time, then α-factor was added and cells were incubated for additional for 2.5 hours. G1 arrest was confirmed by the presence of the characteristic ‘shmoo’ morphology. Cells were washed twice with water and resuspended in YPD with Pronase E. Samples were taken every 30 minutes until 180 minutes. Cells were prepared for flow cytometry as in [60]. Briefly, cells were fixed overnight in 70% ethanol at 4°C, washed once with water, resuspended in RNAase solution and incubated at 37°C for 2 hours. Cells were then washed once with water and resuspended in protease solution for 30 minutes. For FACS analysis, cells were resuspended in 1μM SYTOX solution (Invitrogen). Cell cycle profiles were generated in a BD Canto Flow cytometer using the GFP filter. G1, S and G2/M populations were calculated using FCS Express (De Novo Software). For S3A Fig, cell cycle progression was scored by fluorescence microscopy. Cells containing a single Mtw1-YFP (kinetochore) and Spc42-RFP (spindle pole body, SPB) foci and without bud were scored as G1 cells. Budding cells with a single kinetochore and SPB were scored as S phase. Cells with one kinetochore and two SPB or two kinetochores and two SPBs were scored as G2/M (Metaphase to Telophase).
Diploid his3-/HIS1 strains were streaked on fresh YPD plates and grown for 2 days at 30°C. Five colonies of each strain were resuspended in YPD. 3x106 cells were mixed with 3x107 cells of log-phase cultures of haploid mating tester strains (HIS3/his1-). Cells were concentrated by gentle centrifugation and incubated overnight at 23°C. The next day these cells were plated on synthetic dropout plates and incubated for 3 days at 30°C to select for HIS+ mating products. For each colony, mating products originating from both mating type MATa and MATα tester strains were summed. For each strain, the median number of colonies from the 5 colonies was calculated.
Strains with a tetracycline operator array, inserted at the URA3 locus of chromosome V and a tetracycline repressor linked to mRFP, were grown overnight in synthetic media at 23°C. The day after the culture was diluted and further grown until log phase. Cells were imaged as explain above. In each image, cells showing aberrant chromosome segregation were identified as containing two TetR-mFRP foci in G1 or S-M
Strains were transformed with a CEN plasmid with a selectable marker and grown for two days. 9 colonies were grown overnight in YPD and then plated in either YPD or selective media. The percentage of plasmid loss was calculated by subtracting the amount of cells growing in the selective media to the number of cells growing in YPD. The data is presented as the median of percentage plasmid loss of 9 colonies.
|
10.1371/journal.ppat.1006057 | The Mouse Cytomegalovirus Gene m42 Targets Surface Expression of the Protein Tyrosine Phosphatase CD45 in Infected Macrophages | The receptor-like protein tyrosine phosphatase CD45 is expressed on the surface of cells of hematopoietic origin and has a pivotal role for the function of these cells in the immune response. Here we report that following infection of macrophages with mouse cytomegalovirus (MCMV) the cell surface expression of CD45 is drastically diminished. Screening of a set of MCMV deletion mutants allowed us to identify the viral gene m42 of being responsible for CD45 down-modulation. Moreover, expression of m42 independent of viral infection upon retroviral transduction of the RAW264.7 macrophage cell line led to comparable regulation of CD45 expression. In immunocompetent mice infected with an m42 deletion mutant lower viral titers were observed in all tissues examined when compared to wildtype MCMV, indicating an important role of m42 for viral replication in vivo. The m42 gene product was identified as an 18 kDa protein expressed with early kinetics and is predicted to be a tail-anchored membrane protein. Tracking of surface-resident CD45 molecules revealed that m42 induces internalization and degradation of CD45. The observation that the amounts of the E3 ubiquitin ligases Itch and Nedd4 were diminished in cells expressing m42 and that disruption of a PY motif in the N-terminal part of m42 resulted in loss of function, suggest that m42 acts as an activator or adaptor for these Nedd4-like ubiquitin ligases, which mark CD45 for lysosomal degradation. In conclusion, the down-modulation of CD45 expression in MCMV-infected myeloid cells represents a novel pathway of virus-host interaction.
| Human cytomegalovirus (HCMV) is a tenacious pathogen, which can be life-threatening for immunocompromised patients and immunologically immature newborns. The pathogenicity of HCMV is owed to a plethora of immunomodulatory functions that interfere with host defense mechanisms. Such viral functions can teach us about viral pathogenesis mechanisms, and also about the functioning of immune cells. In this study we report that the mouse cytomegalovirus (MCMV)–a close relative of HCMV–influences surface expression of the cellular protein CD45 on macrophages and we identified the viral gene m42 mediating this effect. CD45 has long been known to be essential for the functioning of lymphocytes, however, its role in macrophages is less well understood. Growth analysis of a viral mutant indicated that the m42 gene confers a replication advantage to MCMV in vivo. We found that the m42 protein induces internalization of CD45 from the plasma membrane and degradation in lysosomes—most likely triggered by interaction of m42 with a ubiquitin ligase. In our study we detected a new element in the complex interaction of cytomegaloviruses with host cells, and further investigation into this mechanism may provide us with new insights into the functions of CD45 in myeloid cells.
| Cytomegaloviruses (CMVs) possess the largest genomes among the herpesviruses [1–3] and dedicate a substantial portion of their genomic coding capacity to accessory functions that are not directly needed for replication of the DNA genome or as structural components for the assembly of progeny virions [4,5]. Among the accessory genes in turn many encode products that interfere with various pathways of the host immune response (reviewed in [6–9]). The fact that CMVs encode so many immunomodulatory genes may be owed to the close relationship that these viruses established with myeloid cells such as macrophages and dendritic cells [10,11]. These cells are central for both initiating the innate immune response as well as for shaping adaptive immunity against infectious pathogens. Despite these adverse conditions CMVs managed to employ myeloid cells as targets for lytic replication [12–14], and moreover as safe harbor for latent infection and a site of reactivation [10,15–21]. Therefore, the development of numerous immunomodulatory functions was probably a necessity for CMV and may be the result of an extensive evolutionary virus-host arms race. In healthy individuals infection with human CMV (HCMV) is nevertheless usually well controlled by the immune system, particularly by NK cells and CD8 T cells, resulting in unapparent infection [22]. However, the balance between host defense and viral countermeasures is delicate, and consequently CMV infection can lead to severe and sometimes life-threatening manifestations in patients with a weakened immune system or in immunologically immature fetuses or newborns [22].
The competition with the host response forced the CMVs to dampen those host defense mechanisms that threaten to eliminate the viruses. Viral immunomodulatory proteins may therefore teach us about yet unknown aspects of immune defense pathways and can point to Achilles’ heels of the host defense, thereby defining targets for potential therapeutic intervention. Although the best studied CMV members–mouse and human CMV (MCMV and HCMV)–underwent separate co-evolution with their respective hosts over almost 100 million years, they frequently developed strategies to interfere with similar processes of the immune system of mouse and man. Interestingly, gene products of MCMV and HCMV that target the same host pathway often display little sequence similarity and sometimes even their mode of action is different, indicating convergent evolution. Thus, by investigating immunomodulatory proteins of MCMV and HCMV and by comparing their function and working mechanism, we may learn about new host defense processes.
We have recently discovered that the HCMV UL11 protein can bind to the host protein CD45 [23], a protein tyrosine phosphatase expressed on the surface of most cells of hematopoietic origin, including T and B lymphocytes, NK cells, dendritic cells and macrophages [24]. CD45 is a positive regulator of antigen receptor signaling, which has been intensively studied and found to be central for the development and activation of T cells, B cells and NK cells (for review see refs. [25,26]). Consequently, disruption of the CD45 gene results in severe combined immunodeficiency in man and mouse [27–32]. Major substrates of the CD45 phosphatase are Src family kinases (SFKs), e.g. Lck in T cells; the removal of a phosphate group from an inhibitory tyrosine residue in the C-terminal domain leads to a primed state of SFKs, enabling them to transduce signals received by receptors to which they are associated. Dephosphorylation of a tyrosine in the autocatalytic domain abrogates the activity of SFKs and explains the negative regulatory role of CD45 observed under certain conditions. Studies with myeloid cells from CD45-deficient mice point to an involvement of CD45 with a number of different functions in these cells, ranging from proliferation upon stimulation with growth factors over adhesion and migration [33,34] to Toll-like receptor (TLR) signaling and production of type I interferons and other cytokines [35–39]. While some of these functions can be explained by the known axis between CD45 and SFKs, the mechanistic basis for other functions, e.g., the impact on TLR signaling, remains elusive [26,40].
Since the functional consequences of the interaction between the HCMV protein UL11 and CD45 are difficult to assess for HCMV infection [41], we wondered whether MCMV has evolved a similar mechanism aiming at CD45. Most viral immunomodulatory proteins exert their function directly within infected cells, and therefore we focused in this study on target cells of MCMV infection that express CD45, in particular on macrophages. We report that MCMV leads to diminished CD45 surface expression on these cells. Furthermore, we found that the viral gene m42 mediates this effect and analyzed the mechanisms of CD45 down-modulation. Infection experiments with an MCMV Δm42 mutant revealed that already an early step of the replication process in vivo is affected.
During our previous studies when we investigated the immune response against MCMV in lungs of neonatal mice [42,43], we noticed that infected macrophages displayed less staining with CD45 antibodies than non-infected macrophages. To investigate the putative interference of MCMV with CD45 expression in more detail, we infected RAW264.7 macrophages with a GFP-expressing MCMV strain (MCMVgfp) and examined the cells 24 h post infection (p.i.) by flow cytometry. In infected cells the amount of CD45 present at the cell surface was substantially reduced (Fig 1A and S1A Fig). Inspection of infected cells by fluorescence microscopy confirmed that only residual amounts of CD45 remained at the plasma membrane (Fig 1B). Comparable results were obtained upon infection of the dendritic cell line DC2.4 (S1D Fig) and bone-marrow-derived macrophages, and also when wildtype MCMV (MCMVwt; devoid of the GFP marker) was used for infection. Treatment of RAW264.7 cells with UV-inactivated virus did not affect CD45 expression (S1C Fig). We therefore supposed that an MCMV-encoded factor mediates down-regulation of CD45 in infected macrophages and other antigen-presenting cells.
In order to identify the viral gene responsible for the observed phenotype, we made use of a set of MCMV deletion mutants (Fig 1C) that lack various parts of the viral genome, covering most genes with accessory functions non-essential for viral replication in cell culture [44,45]. Following infection of RAW264.7 macrophages with the different mutants, CD45 levels were examined by flow cytometry one day later. The results obtained with selected mutants are depicted in Fig 1D. Except of the deletion mutant lacking ORFs m42 and M43, all other mutants led to strong down-modulation of CD45 expression. To assign the function to one of the two ORFs missing in the MCMVgfp-Δm42-M43 mutant, additional mutants were generated with a deletion in either ORF m42 or M43 only (Fig 2A). Infection experiments with these mutants revealed that only the MCMVgfp-Δm42 mutant displayed a loss-of-function phenotype (Fig 2B), strongly suggesting that a gene product encoded by the m42 ORF is involved in the regulation of CD45 surface expression. However, since several transcripts spanning this region have been reported [46,47], a contribution of neighboring ORFs could not be excluded. Therefore, the MCMVgfp-m42STOP mutant was generated that carries only a short DNA cassette containing stop codons within ORF m42, preventing synthesis of a functional protein. Moreover, a rescuant with a restored m42 ORF was constructed based on the genome of the MCMVgfp-m42STOP mutant, to exclude that an accidental mutation elsewhere in the viral genome led to the phenotype. In line with the hypothesis that m42 is the candidate gene, the MCMVgfp-m42STOP mutant did not diminish CD45 surface expression, whereas the rescuant MCMVgfp-m42rev regained this function (Fig 2B).
Next, we examined the growth characteristics of the MCMV-m42STOP mutant. In murine embryonic fibroblasts the growth kinetics of the mutant, of MCMVwt and the rescuant were indistinguishable (Fig 3A), demonstrating that the m42 gene is not essential for replication and production of viral progeny in these cells. However, in bone marrow-derived macrophages–cells that express the CD45 protein targeted by m42 –growth of the m42 mutant was slightly impaired when compared to the control viruses. To learn whether a soluble factor released from the macrophages infected with the m42 mutant might be responsible for the growth phenotype, we took supernatants from cultures of bone marrow derived macrophages infected with the m42 mutant, the revertant virus or from untreated cultures at days 0, 3, 6 and 9 p.i. and transferred the filtered supernatants to new cultures that were then infected with MCMVgfp. Compared to medium from uninfected cultures, supernatants taken from infected cultures at days 3, 6 or 9 p.i. impaired virus growth resulting in lower titers (S2A Fig). However, there was no difference whether the supernatants were taken from cultures infected with the m42 mutant or the revertant MCMV. This experiment does not suggest that additional soluble factors (or increased amounts of a factor with antiviral activity) are released from m42-infected cells and are responsible for the reduced replication capacity observed for the m42 mutant.
Myeloid cells have been implicated in defense against MCMV as well as in its dissemination, and thus we asked whether the replication of the m42 mutant is impaired in vivo as well. BALB/c mice were infected intraperitoneally with 2 × 105 PFU of MCMVwt, MCMV-m42STOP and MCMV-m42rev, respectively, and viral titers in liver, spleen, lungs and salivary glands were determined on day 3, 7 and 21 p.i. Already on day 3 p.i. titers of the MCMV-m42STOP mutant were markedly reduced, with the median titers ~0.5 to one order of magnitude lower than those of MCMVwt and the rescuant MCMV-m42rev (Fig 3B). Differences of viral titers in the organs were also observed on day 7 p.i. although less pronounced (S2 Fig) and were again obvious when titers were measured on day 21 p.i. in salivary glands (Fig 3B, right panel)–a site where MCMV persists for a prolonged time period.
In an independent experiment, BALB/c mice were infected via the footpad route with the same dose (2 × 105 PFU) of the MCMV-m42STOP and MCMV-m42rev viruses and viral DNA loads in different organs and tissues were determined 3 and 5 days p.i. by quantitative PCR (Fig 3C). In all tissues examined viral loads of the m42STOP mutant were reduced at both time points, confirming the results of the previous experiment. Analysis of transcript levels for viral genes of all three temporal classes (IE, early and late) in the draining lymph node and spleen indicated that the m42STOP mutant was able to establish productive infection and that the viral DNA loads did not simply reflect the inoculated viruses (S2C Fig). The transcript levels for the m42 mutant were lower than those for the rescuant (in spleens for some transcripts below detection limit), again pointing to attenuation as a consequence of the missing m42 gene. Taken together, these results indicate that the m42 gene confers a replication advantage to MCMV in vivo, which becomes manifest already on day 3 p.i.
The polypeptide encoded by ORF m42 is predicted to be a tail-anchored membrane protein with a molecular mass of 17.6 kDa (Fig 4A). In order to detect the protein, a monoclonal antibody (mAb) was generated using a bacterially expressed recombinant protein. In macrophages infected with MCMVgfp or MCMVgfp-m42rev a protein of ~18 kDa specifically reacting with the mAb was observed by immunoblotting, which was absent in cells infected with MCMVgfp-m42STOP (Fig 4B and S1F Fig). In most experiments an additional band with a molecular mass of ~23 kDa was detected, which was often fainter than the 18 kDa band with some variability in abundance (S1B Fig). Consistent with the expected function of m42, the amount of the CD45 protein was decreased in lysates of RAW264.7 cells infected either with MCMVgfp or the rescuant compared to non-infected cells or cells infected with the m42 deletion mutant (Fig 4B). The m42 protein was first detected at 6 h p.i. and expression was strongest between 8 and 16 h p.i., followed by slight decline at later time points (Fig 4C, second panel). The onset of m42 expression was similar to that of the single-strand DNA binding protein M57, indicating that m42 belongs to the early class of viral proteins. When analyzed by immunoblotting, lower amounts of CD45 were observed in the MCMVgfp infected macrophages at 16 h p.i., and the CD45 levels were further decreased at 20 and 24 h p.i. (Fig 4C, top panel). Flow cytometric analysis revealed that the reduction of CD45 surface expression in infected RAW267.4 cells started around 12 h p.i. and was accomplished in the majority of the cells at 16 h p.i., with further progression until 20 h p.i. (Fig 4D). As expected, in cells infected with the m42STOP mutant the CD45 amounts were not diminished during the infection cycle (S1F Fig). Taken together, m42 expression starts early in infection and is followed by CD45 downregulation with a certain temporal delay.
The latter result raised the question whether modulation of CD45 surface expression is mediated solely by m42 or whether additional viral proteins are required. To address this point, RAW264.7 cells were transduced with a retroviral vector and a cell line stably expressing m42 was generated (referred to as RAW_m42). The parental RAW264.7 cells and a cell line obtained by transduction with the empty retroviral vector served as controls. Immunoblot analysis confirmed the presence of the 18 kDa as well as the 23 kDa m42 protein in the RAW_m42 cell line, and revealed lower CD45 levels than in control cells (Fig 4E, right panel). In agreement with this result, cell surface expression of CD45 in the RAW_m42 cells was one order of magnitude lower than in the parental RAW264.7 and in RAW_ctrl cells (Fig 4E, left panel), displaying reduction of the CD45 level to a similar extent as in MCMV-infected cells. Thus, m42 expression alone is sufficient for CD45 down-modulation.
Possible mechanisms that could explain the m42-mediated CD45 down-regulation are more rapid turn-over of surface-resident CD45 and degradation; however, interference with synthesis, maturation and transport of new CD45 molecules could not be excluded. Metabolic 35S-labeling of newly synthesized CD45 protein (pulse-chase experiment) and analysis of the glycosylation pattern of CD45 by treatment with endoglycosidase H did not provide hints that m42 impairs CD45 maturation or transport within the secretory pathway (S3A Fig). This brought the surface-resident CD45 molecules into the focus of our interest and we decided to track them by utilizing a FACS-based internalization assay. After labeling with a CD45 mAb, the RAW_m42 and the parental RAW264.7 cells were kept at 37°C for another 3 h. At various time points CD45 molecules remaining at the plasma membrane were visualized with a PE-conjugated secondary Ab and cells were analyzed by flow cytometry. In normal RAW264.7 cells the amount of labelled CD45 molecules was reduced by 20% within 1 h, but later remained rather constant, possibly indicating internalization and recycling of a fraction of CD45 back to the plasma membrane (Fig 5A). Conversely, in RAW_m42 cells a 40% reduction of the antibody-labelled CD45 molecules at the surface was observed after 1 h and afterwards levels continued to decline. Turnover of the transferrin receptor (CD71) at the cell surface–a protein known to cycle between the plasma membrane and endosomal vesicles–was comparable for both cell lines (Fig 5B), indicating that stronger downregulation of CD45 in RAW_m42 cells was not a general effect applying to other surface proteins. A likely explanation for the observed result is more rapid internalization of CD45 in m42-expressing macrophages. To further substantiate this finding, CD45 present on the surface of the macrophage cell lines was labelled as described above and cells were subsequently fixed and permeabilized at different time points followed by treatment with the fluorescent secondary Ab. Confocal microscopy allowed to detect potentially internalized CD45 molecules as well as those still present at the cell surface. Already 30 min after labelling a fraction of CD45 was found in dot-like structures inside the RAW_m42 cells (Fig 5C). At 2 h the dots appeared more pronounced, and few of the labeled CD45 molecules were still present at the cell surface. This became particularly obvious when compared to the images of the parental RAW264.7 cells, which revealed prominent CD45 surface staining at the time points analyzed and no accumulation of CD45 within the cytoplasm. When examined 6 h post labelling almost no Ab-loaded CD45 was detectable in m42-expressing cells, whereas substantial surface staining was still visible in RAW264.7 macrophages. Similar results were obtained for MCMV-infected cells (S3B Fig). We conclude from these data that in the presence of m42 surface-resident CD45 is rapidly internalized and subsequently degraded, whereas in the parental RAW264.7 cells CD45 is remarkably stable.
To learn which pathway is involved in degradation of CD45, RAW264.7 and RAW_m42 cells were treated with different inhibitors of the proteasome and of lysosomal proteases and cell lysates were subsequently analyzed by immunoblotting for CD45 and m42. In RAW264.7 macrophages CD45 amounts did not change upon treatment with the various inhibitors (Fig 6A), suggesting that CD45 has a long half-life in this cell line. As expected, in RAW_m42 cells CD45 expression was lower than in the parental RAW264.7 cells and was not influenced by treatment with proteasome inhibitors (MG123, epoxomycin). In contrast, when RAW_m42 were exposed to inhibitors affecting the function of lysosomal proteases, CD45 amounts increased (Fig 6A). This result strongly suggests that m42 directs CD45 for degradation via the lysosomal pathway. We also noted that treatment with the proteasome inhibitors increased the abundance of the larger 23 kDa m42 form, whereas the amount of the smaller 18 kDa form was diminished.
Since internalization and degradation of surface proteins often involves ubiquitination [48,49], we searched for hints whether m42 could promote such a mechanism. Analysis of the m42 amino acid sequence did not reveal typical features of an ubiquitin ligase, e.g., a RING domain. Although the m42 protein shares little sequence similarity to the protein encoded by the positional homolog pUL42 of HCMV, the two proteins seem to have some structural properties in common. Both are predicted to be type II transmembrane proteins and possess a short luminal tail. It has recently been reported that the HCMV UL42 protein interacts with Itch [50], a member of the Nedd4-like E3 ubiquitin ligases of the HECT domain family (homologous to the E6AP carboxyl terminus) [51–53]. To examine whether the MCMV m42 protein may employ such a ubiquitin ligase for its activity we performed transfection experiments with HEK 293T cells and a HEK 293T cell line stably expressing m42. In this cell line expression of the HECT E3 ligase Itch was strongly diminished (Fig 6B) and this applied also to Nedd4, but not to Smurf2 (Fig 6C), another member of the Nedd4-like ubiquitin ligases. When the cells were transfected with an expression vector for murine CD45, lower amounts of CD45 were detected in the HEK 293T cells expressing m42 compared to the parental cells (Fig 6B), indicating that the effect of m42 on CD45 expression can be reproduced in this cell line. To test whether there is direct interaction of m42 with CD45, we transfected HEK 293T cells with expression plasmids for m42 and CD45. In this way, higher expression levels for m42 and CD45 could be achieved than in RAW264.7 or HEK 293T cells stably expressing m42 and/or CD45, which should facilitate the detection of a putative interaction. Following immunoprecipitation of CD45 we could not detect co-precipitation of m42 (Fig 6D). Thus, our data do not suggest that there is a direct interaction between m42 and CD45.
The m42 protein like HCMV pUL42 carries two putative PY motifs (LPTY and PPSY) in the N-terminal part (Fig 4A). Such motifs were found to be relevant for the recruitment of Nedd4-like ubiquitin ligases by HCMV pUL42 [50] and by other related herpesvirus proteins [54]. Since the PPSY motif in m42 is most similar to the respective motifs in HCMV UL42, we introduced two different mutations (AASY, PPSA) into the m42 protein, to learn whether this motif is required for modulation of CD45 expression by m42. Immunoblot and flow cytometric analysis of cell lines stably expressing the m42 variants revealed that CD45 expression was not reduced as in the cells expressing the wildtype m42 protein, and CD45 amounts were comparable to those seen in RAW267.4 cells or in control cells transduced with the empty retroviral vector (Fig 7A and 7B). As expected, surface levels of CD71 were comparable for all cell lines (Fig 7B, right panel). The abundance of the m42 form with the higher molecular mass was markedly increased in the cell lines expressing the proteins with the disrupted PPSY motif (Fig 7A). From this result we concluded that the PPSY motif in m42 is needed for the function of the viral protein to down-modulate CD45, further supporting the notion that a HECT E3 ubiquitin ligase is involved in this process.
In this study we report on the modulation of the surface expression of the cellular protein tyrosine phosphatase CD45 in MCMV-infected macrophages. To our knowledge, such a phenotype has not been described for other viruses or pathogens before, and thus adds a novel element to the arsenal of known virus-host interactions. Here, we focused on the investigation into this phenotype, identified the viral m42 gene as being responsible for mediating the effect, and discovered that CD45 is internalized and degraded in lysosomes, which is most likely induced by m42-mediated activation of Nedd4-like ubiquitin ligases.
The host protein CD45 is expressed abundantly in leukocytes and is essential for keeping signaling pathways of these cells in a primed state [25]. In view of this important function, it may not be surprising that viruses have developed strategies to dampen the activity of CD45 (reviewed in ref. [55]). So far, the only viral proteins reported to interact with CD45, are the HCMV UL11 protein [23] and the E3/49K protein of the species D adenovirus (AdV) type 19a [56]. More recently, it has become clear that many if not all E3/49K homologs of species D AdV possess this property [57] and inhibit the function of T and NK cells via binding to CD45 [56]. Whereas these viral CD45-binding proteins are either secreted or expressed on the surface of infected cells, and are thought to modulate the activity of neighboring immune cells that attack infected cells and threat to eliminate the viruses, the way how MCMV affects CD45 is different in that it acts directly within the infected myeloid cells and thus represents a novel mode of interference with CD45. Compared to lymphocytes, the role of CD45 in myeloid cells is less well analyzed, with only a few publications reporting on an altered cytokine response and adhesion properties of dendritic cells and macrophages lacking CD45 [33,35,37,39,58]. In macrophages completely devoid of CD45 a higher activity of Src family kinases (SFK) such as Hck and Lyn has been observed [33], implying that CD45 negatively regulates these kinases. Analysis of the activity of different SFKs in MCMV-infected macrophages and macrophage cell lines expressing m42 is therefore one of the objectives for further studies. One has to point out, however, that MCMV infection does not lead to a complete loss of CD45 and–according to studies in T cells [59]–variation of CD45 amounts over a wide range may result in only moderate functional consequences.
One effect we observed was the slightly impaired growth kinetics of the m42STOP mutant in bone marrow-derived macrophages, a phenotype which was not visible in murine fibroblasts, pointing to a possible cell-type specific role of m42. We did not find hints that a secreted factor inhibits the growth of the m42 mutant, although in view of the reported CD45-dependent modulation of cytokine production in myeloid cells [35,37,39] this hypothesis was standing to reason. For further analyses a two-pronged approach will be applied; on the one hand pathways downstream of CD45 will be examined, e.g. adhesion of macrophages and cytokine production, and on the other hand—by considering the targeting of Nedd4-like E3 ubiquitin ligases by m42—processes known to be regulated by these ubiquitin ligases will be investigated [51,60].
In vivo the m42STOP mutant grew to lower titers in all examined organs and tissues of infected mice. The fact that the replication deficit of the mutant becomes manifest already on day 3 in infected mice is compatible with the idea that m42 interferes with an early defense mechanism. The early innate response has a drastic influence on the amount of viral progeny produced in the first round of replication, setting the stage for the further course of viral infection [61], and such an early effect could therefore also explain the reduced viral titers observed at later time points (e.g., in salivary glands at 21 days p.i.). To which extent the in vivo phenotype of the m42STOP mutant is connected with the regulation of CD45 expression remains to be elucidated. Due to the multiple important functions of CD45, particularly in the adaptive immune response, this question cannot easily be addressed by using CD45 knock-out mice. Lack of T cell immunity due to a mutated CD45 gene has for instance been identified as a factor increasing the susceptibility to HSV-1 infection and the risk of herpes simplex encephalitis [62]. The T cell defect in CD45 knock-out mice would mask other, possibly more subtle effects of CD45 in other cell types such as macrophages. Further investigation into the in vivo phenotype of the m42 mutant has to await a more comprehensive understanding of the m42-mediated effects at the molecular level.
After identifying the m42 gene we asked by which mechanism the CD45 down-modulation is mediated. We did not detect altered maturation of CD45 in infected macrophages and synthesis of CD45 transcripts was rather increased than reduced (S1E Fig). Although the CD45 protein has been intensively studied over decades, little seems to be known about the turn-over of this molecule. A single publication dating from 1992 [63] reported that interference with glycosylation in the erythroleukemic cell line K562 leads to rapid degradation of newly synthesized CD45 molecules–probably by a pathway that is now known as endoplasmic-reticulum-associated protein degradation [64]. Data from the very same publication suggested that mature CD45 molecules have a long half-life, which is in agreement with our finding that CD45 is quite stable in macrophages as well (Figs 5 and 6A). The internalization assays revealed that in m42-expressing macrophages the internalization rate of CD45 molecules is increased, whereas there was no difference in the internalization of a control protein (transferrin receptor CD71). Moreover, treatment with substances inhibiting the function of lysosomes led to an increase of CD45 levels, indicating that following internalization CD45 molecules undergo lysosomal degradation.
Endocytosis of functionally important surface molecules has been observed upon infection of cells with various viruses, and often involves ubiquitination of the target molecules [65–69]. For the UL42 protein of HCMV it has recently been reported that it interacts with the Nedd4-like E3 ubiquitin ligase Itch and induces its degradation [50]. Although there is little amino acid sequence similarity between pUL42 and the MCMV m42 protein, both of them harbor so-called PY motifs, which are known to mediate the interaction with WW domains, present for instance in E3 ubiquitin ligases of the Nedd4 family [53]. Similar to the findings for HCMV pUL42, we detected here that the amounts of Itch and of the related Nedd4 ubiquitin ligase are severely diminished in cells expressing m42. Interestingly, another member of the Nedd4-like ubiquitin ligases, Smurf2, was not affected, indicating that m42 displays selectivity for certain proteins of this ubiquitin ligase family. Disruption of one of the putative two PY motifs found in m42 led to loss of the ability to regulate CD45 surface expression in macrophages. Based on these data we propose the following model to explain m42-induced internalization and degradation of CD45 (Fig 7C). As has first been suggested for the HSV-2 UL56 protein [70], m42 may act in a similar manner as Nedd4 family-interacting proteins (NDFIPs) that are known to induce conformational changes, thereby releasing Nedd4-like ubiquitin ligases from an auto-inhibitory state. m42 and NDFIPs have indeed structural features in common, including PY-motifs, a transmembrane region and a short C-terminal tail. Once activated, Nedd4 family members can ubiquitinate various substrates [71]. As described for HCMV pUL42 and Itch, interaction triggers degradation of Itch itself [50], and the diminished amounts of Itch and Nedd4 found in m42-expressing cells suggest the same mechanism. It might be that the m42 protein is a substrate for these ubiquitin ligases as well and is turned over together with Itch and Nedd4, because when the PPSY motif is mutated and interaction with the Nedd4-like proteins presumably is disrupted, the abundance of the m42 variants, especially the 23 kDa protein species, was increased. We assume that particularly in this case the 18 kDa m42 version is modified, perhaps by phosphorylation as has been reported for the related UL56 proteins of HSV-2 and equine herpesvirus type-1 (EHV-1) [70,72].
This leaves the question how CD45 could become a substrate for Nedd4-like ubiquitin ligases and whether it is their only target in MCMV-infected cells. Nedd4 family members have been implicated in trafficking of plasma membrane proteins, and substrate specificity is mediated by binding of WW domains to PY motifs as well as other phosphorylated serine and threonine residues in target proteins, or via adapter molecules [73,74]. We could not detect an interaction between CD45 and m42, which argues against an adapter function of m42; however, we cannot exclude that due to the limited detection sensitivity of the m42 antibody this escaped our notion. Interestingly, a potential PY motif (LYSP) is present in the C-terminal intracellular part of murine CD45, which is conserved in human and rat CD45. If an interaction via this PY motif could be experimentally confirmed, this would point to a role of Nedd4-like ubiquitin ligases in the natural turnover of CD45. However, as has been shown for Itch, the interaction between one WW domain and one PY motif is not sufficient for activation [71]. Accordingly, the additional binding of m42 (or of a cellular regulatory protein) may be needed to activate Itch, resulting in the internalization of CD45.
By testing the surface expression of some other proteins (MHC-I, CD40, CD86 and CD71) in macrophages expressing m42 (S4 Fig), we could exclude that m42 increases endocytosis of surface-resident proteins in general. In view of the many known substrate proteins of Nedd4-like ubiquitin ligases [60] it is nevertheless likely that besides CD45, m42 expression affects other specific molecules located at the plasma membrane and also inside the cell, and such substrate proteins are prime candidates for testing in MCMV-infected cells.
In summary, we identified an MCMV gene that interferes with the expression of the host molecule CD45 in myeloid cells, pointing to a new pathway of virus-host interaction. We believe that the mouse model of CMV infection as a natural virus-host system is ideal to further unravel the relevance of the viral down-modulation of CD45 expression. The m42 gene will therefore not only be a powerful tool to learn how MCMV manipulates host pathways to its own advantage, but could also provide us with novel insights into the functions of CD45 in myeloid cells.
The macrophage cell line RAW264.7 (ATCC TIB-71) and the human epithelial cell line HEK 293T (ATCC CRL-3216) were grown in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% FCS, 2 mM glutamine, 100 U/ml penicillin and 100 μg/ml streptomycin. For generation of bone marrow-derived macrophages (BMDM), bone marrow cells were isolated from femurs and tibias of male C57BL/6 mice (purchased from Charles River, Sulzfeld, Germany) as described previously [75] and were kept for 7 days in a mixture of 75% DMEM containing 10% FCS and 25% conditioned medium of L929 cells (ATCC CCL-1) as a source of macrophage stimulating factor [76]. Differentiated cells were routinely tested by flow cytometry and >99% of the cells expressed the macrophage-specific marker F4/80 (Ab BM8; BioLegend). The dendritic cell line DC2.4 [77] was grown in RPMI 1640 medium supplemented with 10% FCS, 2 mM glutamine, 100 U/ml penicillin and 100 μg/ml streptomycin. Murine embryonic fibroblasts (MEF) were prepared from embryos of BALB/c mice (purchased from Charles River) on day 17 of gestation following a published protocol [78] and were grown in the same medium as RAW264.7 cells. Treatment of cells with inhibitors was performed for 4 h with the following substances: MG132 (1.25 μM), epoxomycin (2.5 μM), NH4Cl (20 mM) Bafilomycin A1 (50 nM) or Leupeptin (200 μM).
The MCMV strain herein referred to as MCMVwt is derived from the BAC-cloned MCMV genome pSM3fr-MCK-2fl [79]. The strain expressing the enhanced green fluorescent protein (MCMVgfp) and the deletion mutants used in the screen were described previously [44,80]. The genomes lacking the ORFs m42 or M43 were generated in E.coli using the BAC pSM3fr-GFP and by following the mutagenesis procedure described in Loewendorf et al., 2004. Genomes giving rise to the m42STOP mutant and the m42rev rescuant were once generated for the BAC pSM3fr-MCK-2fl and a second time for a pSM3fr-GFP-derived BAC, in which a frameshift mutation in the m129 locus was repaired as described in Jordan et al. (2011). En passant mutagenesis [81] was performed using plasmid pEPkanS2 as template and the following primers to amplify the kanamycin resistance (KanR) cassette: m42STOP_fwd, 5’-ttc cga acc gga gca ccg ttt gcc tac tta tct gga agc ggg cta gtt aac tag ccg tcg gtg aga acg ctc gct agg atg acg acg ata agt agg g-3’; m42STOP_rev, 5’-cgt tga gaa aca cag ttt cat agc gag cgt tct cac cga cgg cta gtt aac tag ccc gct tcc aga taa gta ggc aca acc aat taa cca att ctg att ag-3’. m42rev_fwd, 5’-ttc cga acc gga gca ccg ttt gcc tac tta tct gga agc ggt cgg tga gaa cgc tcg cta gga tga cga cga taa gta ggg-3’; m42rev_rev, 5’-cgt tga gaa aca cag ttt cat agc gag cgt tct cac cga ccg ctt cca gat aag tag gca caa cca att aac caa ttc tga tta g-3’. Viruses based on pSM3fr-MCK-2fl (without GFP gene) were used for in vivo experiments. To reconstitute virus mutants, MEF were transfected with the respective BAC-DNA using the jetPEI transfection reagent (Polyplus-transfection, Illkirch, France). Virus stocks were prepared by ultracentrifugation, and viral titers were determined by plaque assay on MEF following established protocols [82]. RAW264.7 macrophages were infected at a multiplicity of infection (MOI) of 3 (if not indicated otherwise) followed by centrifugal enhancement for 30 min at 280 × g.
The retroviral vector pLHCX (Clontech) carrying a hygromycin B resistance gene was used to clone the m42 ORF downstream of the HCMV major IE promotor. The PPSY motif was disrupted by site-directed mutagenesis using the retroviral plasmid pLHCXm42 as template and the following primers: AAxY_fwd: 5’-gcc gcat cct acg aga gtc tc-3’; AAxY_rev: 5’-ggg atc gtc gtc cgc-3’ and PPxA_fwd: 5’-gcc gag agt ctc ttt ggt-3’; PPxA_rev: 5’-gga cgg tgg ggg atc-3’. For generation of stable cell lines, retroviruses were produced by transfection of HEK 293T cells with the respective retroviral vector together with plasmids encoding the VSV-G envelope protein and the gag and pol functions, respectively. RAW264.7 cells or HEK 293T cells were transduced with the retroviruses, and 24 h later cells were selected for hygromycin resistance by adding 200 μg/ml Hygromycin B gold (InvivoGen). Cell lines were propagated in medium with 100 μg hygromycin/ml.
Single cell suspensions were incubated for 5 min in 5% horse serum (to block Fcγ-receptors) and subsequently for 20 min at 4°C with antibodies specific for CD45 (clone C363.16A [for experiment shown in Fig 1D] or clone 30-F11 [for remaining experiments]; both eBioscience) or CD71 (R17217; eBioscience). For exclusion of dead cells staining was performed with 7-AAD (10 μg/ml). Subsequently, cells were washed with PBS/2 mM EDTA and fixed with 1.5% PFA. Samples were measured with the Cytomics FC 500 or CytoFLEX Flow Cytometer (both Beckman Coulter) and data were analyzed using Kaluza 1.5 software (Beckman Coulter).
Cells were lysed in NP40 lysis buffer (1% NP40, 25 mM Tris-HCl (pH 8.0), 150 mM NaCl, 5 mM EDTA). Following the determination of protein concentrations by Bradford assay (Bio-Rad), the equivalent of 20 μg of total protein (or 60 μg for analysis of m42) was loaded per well on 6–8% SDS-polyacrylamide gels (or 13% for analysis of m42). Antibodies used for probing the blots were CD45 (69/CD45; BD); GAPDH (14C10; Cell Signaling); m42 (m42.02); IE1 (Croma101); E1 (Croma103), M57 (M57.02). Antibodies directed against MCMV proteins were generated and provided by CAPRI (Rijeka, Croatia). Horseradish peroxidase-coupled secondary antibodies (Dako) were used at a 1:5000 dilution. Signals were detected with an LAS-3000 imager following treatment with the ECL Select substrate (GE Healthcare). Images were processed using Adobe Photoshop CS4.
Cells were grown on coverslips and either mock-infected or infected with an MOI of 1. At the time points indicated cells were fixed with 3% PFA and permeabilized with 0.2% Triton X-100. Upon blocking with 0.2% gelatin, cells were labelled with the CD45-specific antibody (30-F11; eBioscience) and Alexa Fluor 647-coupled goat anti-rat IgG. Nuclei were counterstained with Hoechst 33342 dye (Cell Signaling). Confocal images were acquired with a Leica DM IRB microscope with a TCS SP2 AOBS scan head and processed using ImageJ and Adobe Photoshop CS4.
For the FACS-based internalization assay, cells were labelled with primary antibodies specific for CD45 (30-F11; eBioscience) or CD71 (R17217; eBioscience) for 20 min on ice. After washing cells were incubated at 37°C until samples were cooled on ice at the indicated time points and stained with PE-coupled secondary antibodies, followed by flow cytometric analysis.
For the microscopy-based internalization assay, cells were grown on cover slips. To label surface resident CD45, cells were first incubated on ice for 15 min and then incubated with the 30-F11 antibody for 30 min on ice, followed by washing. Cells were incubated at 37°C and at the time points indicated fixed with 3% PFA. Further treatment was done as described for immunofluorescence microscopy.
Proteins were metabolically labeled by incubating the cells with medium containing (35S)-methionine and -cysteine (4–6 MBq per sample) for 30 min. Thereafter, cells were either lysed immediately or further incubated for the indicated time periods in the presence or absence of MG132 (5 μM). Precipitation of CD45 was performed with a specific antibody (30-F11, eBioscience) in combination with protein G-sepharose beads; precipitated proteins were optionally treated with Endoglycosidase H (Roche) and then separated by SDS PAGE. For autoradiography dried gels were exposed to a phosphor screen for 2 weeks and developed using an Optimax machine.
Viral replication in vivo was analyzed in 7 week-old BALB/c mice upon intraperitoneal infection with 2 × 105 PFU of tissue culture-derived virus stock. At time points indicated mice were sacrificed and viral titers in organ homogenates were determined by plaque assay.
Quantitation of MCMV genomes from organ homogenates were performed as described [83]. Briefly, DNA was isolated using the DNeasy Blood and Tissue Kit (Qiagen) and viral and cellular genomes were quantitated in absolute number by M55-specifc and pthrp-specific qPCR normalized to an log10-titration of standard-plasmid pDrive_gB_PTHrP_Tdy [84]. Viral transcripts in the draining lymph node were quantified by RT-qPCR specific for m123/IE1, M112/E1 and M86/MCP monitoring all kinetic stages of viral replication as described in greater detail previously [83]. Total RNA was isolated with the RNeasy Mini Kit (Qiagen) and absolute quantification of viral transcripts was performed with graded numbers of the specific in vitro generated transcripts as standard. For normalization cellular ß-actin transcripts were quantitated in parallel. For relative quantification of CD45 mRNA expression in RAW264.7 cells total RNA was isolated with the RNeasy Mini Kit and reverse transcribed into cDNA. qPCR was performed using CD45 specific primer (5´-aga aga gag atc cac cca gtg acc-3´ and 5´-gct cac tct ctt tgc tca tct cca-3´) and CD45 mRNA levels were normalized to expression of β-actin.
Differences between two data sets were evaluated by Student’s t test (unpaired, two-tailed) after log-transformation with Welch’s correction using Graphpad Prism version 5.0 (GraphPad Software, San Diego, CA). P values <0.05 were considered as statistically significant.
All animal experiments were performed according to the recommendations and guidelines of the Federation of European Laboratory Animal Science Associations (FELASA) and the Society of Laboratory Animals (GV-SOLAS) and animal research protocols were approved by the Niedersächsische Landesamt für Verbraucherschutz und Lebensmittelsicherheit (permission no AZ33.12-42502-04-12/1042), according to German Federal Law ³8 Abs. 1 TierSchG (animal protection law).
|
10.1371/journal.ppat.1002549 | Transcriptional Activation of the Adenoviral Genome Is Mediated by Capsid Protein VI | Gene expression of DNA viruses requires nuclear import of the viral genome. Human Adenoviruses (Ads), like most DNA viruses, encode factors within early transcription units promoting their own gene expression and counteracting cellular antiviral defense mechanisms. The cellular transcriptional repressor Daxx prevents viral gene expression through the assembly of repressive chromatin remodeling complexes targeting incoming viral genomes. However, it has remained unclear how initial transcriptional activation of the adenoviral genome is achieved. Here we show that Daxx mediated repression of the immediate early Ad E1A promoter is efficiently counteracted by the capsid protein VI. This requires a conserved PPxY motif in protein VI. Capsid proteins from other DNA viruses were also shown to activate the Ad E1A promoter independent of Ad gene expression and support virus replication. Our results show how Ad entry is connected to transcriptional activation of their genome in the nucleus. Our data further suggest a common principle for genome activation of DNA viruses by counteracting Daxx related repressive mechanisms through virion proteins.
| To initiate infection, DNA viruses deliver their genome to the nucleus and express viral genes required for genome replication. Efficient transport is achieved by packing the viral genome as a condensed, transcriptionally inactive nucleo-protein complex. However, for most DNA viruses, including Adenoviruses (Ads), it remains unclear how the viral genome is decondensed and how transcription is initiated inside the nucleus. Cells control unwanted gene expression by chromatin modification mediated through transcriptionally repressive complexes. A key factor in repressive complex assemblies is the transcriptional repressor Daxx. The Ad structural capsid protein VI is required for endosomal escape and nuclear transport. Here we show that protein VI also activates the Ad E1A promoter to initiate Ad gene expression. This is achieved through the removal of Daxx repression from the E1A promoter, which requires a conserved ubiquitin ligase interacting motif (PPxY-motif) in protein VI. We further show that capsid proteins from other unrelated DNA viruses also activate the Ad E1A promoter and support Ad replication by counteracting Daxx repression, functionally replacing protein VI. Our data suggest that reversal of Daxx repression by virion proteins is a widespread mechanism among DNA viruses that is not restricted to a single virus family.
| DNA viruses require the transport of their genome into the nucleus to initiate replication. Cells perceive the introduction of foreign nucleic acids or unscheduled replication as danger signals and activate a DNA damage response that leads to cell cycle arrest and/or apoptosis. To ensure proper replication, DNA viruses express ‘early’ viral genes to degrade or displace key regulators of cellular antiviral machinery. In return, cells repress incoming viral genomes through a network of transcriptional repressors and activators that normally control cellular homeostasis [reviewed in 1], [2].
The nuclear domains thought to be responsible for repressing viral genomes are ND10 or promyelocytic nuclear bodies [PML]-[NBs; reviewed in 3,4] named after the scaffolding PML protein. PML-NBs are interferon inducible, dot-like nuclear structures associated with proteins with transcriptional repressive functions. These include HP-1, Sp100, ATRX and Daxx [summarized in 4], [5]. Daxx (death domain associated protein) was first described as a modulator of Fas-induced apoptotic signaling [6]. When chromatin-bound, Daxx inhibits basal gene expression from various promoters by binding to transcription factors (e.g. p53/p73, NF-kappaB, E2F1, Pax3, Smad4 or ETS1), ATRX, histone deacetylases and core histones to form a repressive chromatin environment [7]–[13]. In contrast, Daxx localization to PML-NBs reduces its repressive capacity and facilitates apoptosis through p53 family members [5], [7], [14].
PML-NBs are found in close proximity to replication centers of DNA viruses (e.g. adenoviruses (Ads), herpes simplex virus (HSV-1), human cytomegalovirus (HCMV) and human papillomavirus [HPV]; [ 15], [16]–[18]. Gene expression from these viruses is repressed via the PML-NBs, suggesting a role in antiviral defense [19]–[22].
To counteract genome repression, viral genome activation involves PML-NB disruption or degradation of Daxx, Sp100 and/or PML via different mechanisms. HCMV gene expression is initiated by proteasomal degradation of Daxx via tegument protein pp71 of the incoming particle [23]. Early HSV-1 gene expression requires PML degradation, mediated by the virus encoded ubiquitin ligase ICP0. Furthermore, in order to activate viral gene expression, transcriptional repression by Daxx and ATRX needs to be relieved [3], [24], [25]. HPV early gene expression is supported by reorganization of PML-NBs through the minor capsid protein L2 [26].
At the beginning of infection, Ads express the immediate early protein E1A from the E1A promoter. E1A binds and displaces the transcriptional repressor Rb from E2F transcription factors. This results in the auto-stimulation of E1A expression and the activation of the downstream viral expression units E1B, E2, E3 and E4 as well as promoting cellular gene expression. The early E1B-55K protein forms a SCF-like E3-ubiquitin ligase complex with the viral E4orf6 and several cellular factors. This complex degrades factors (for example, factors of the DNA damage response) to ensure progression of the replication cycle [summarized in 1], [2], [27]. E1B-55K protein complex also targets Daxx for proteasomal degradation counteracting its repressive effect [21]. In contrast to HSV-1, PML is not degraded by Ads but relocalized into track-like structures through the E4orf3 protein [28], [29].
Despite the well-characterized mechanism of E1A dependent transactivation of early Ad genes, it is unclear how the E1A transcription is efficiently initiated before other viral genes are expressed. The genome enters the cell as a transcriptionally inactive nucleoprotein complex, which is highly condensed by the histone-like viral protein VII inside the capsid shell. Partial disassembly of the endocytosed capsid releases the endosomolytic internal capsid protein VI, permitting endosomal membrane penetration [30], [31] and transport towards the nucleus. After import through the nuclear pore complex, Ad genomes associate with PML-NBs and replication centers are established [30], [31], [reviewed in 32], [33]–[35]. Endosomal escape and subsequent transport are facilitated by Nedd4 ubiquitin ligases, which are recruited through a conserved PPxY motif in protein VI. Ads with mutated PPxY motif do not bind Nedd4 ligases and have reduced infectivity, showing the importance of this interaction for the onset of gene expression from the viral genome [36].
Here we report that Ad capsid proteins and cytoplasmic entry steps are linked to initiation of the adenoviral E1A expression by counteracting Daxx mediated transcriptional repression. Using the Ad system, we further show that capsid proteins from several other DNA viruses share and complement this function. This suggests a conserved mechanism among DNA viruses and provides insights into the very early virus-host interactions required to establish an optimal cellular environment for productive infection.
The capsid protein VI participates in two crucial steps in the nuclear delivery of the Ad genome. Firstly, protein VI is required for lysis of endosomal membranes. Secondly, it is needed for efficient post-endosomolytic transport, mediated by the cellular ubiquitin ligase Nedd4 that binds to a conserved PPxY motif in protein VI. Mutating the PPxY motif interferes with capsid transport toward the nucleus and efficient viral gene expression [30], [36].
To investigate the role of protein VI during post-endosomolytic steps required for the onset of viral replication, we constructed replication competent Ads containing the E1 region with either wildtype (wt) protein VI (HH-Ad5-VI-wt, depicted in the Figure S1) or mutant “M1” protein VI in which the PPSY motif was mutated to PGAA that abolished Nedd4 interaction [HH-Ad5-M1; Fig. S1; [36]]. Following infection of U2OS cells, we observed that M1 virus replication was attenuated compared to wt (Figure 1A and S1B). This is in agreement with our previous observations showing reduced infectivity of an E1-deleted M1 Ad vector compared to the corresponding E1-deleted wt Ad vector [36]. To distinguish between capsid transport and possible more downstream effects, we infected cells with different amounts of replication competent wt and M1 viruses. Then, we determined the genome copy numbers in nuclear and cytoplasmic fractions by qPCR and the efficiency of the initiation of virus replication by quantification of E2A stained replication centers (detailed in Figure S2). Compared to wt, fewer M1 virus genomes accumulated in the nucleus associated fraction, independent of the amount of input virus. In contrast, initiation of virus replication for M1 genomes was reduced for low, but not at high physical particle per cell ratios (Figure S2) suggesting defects downstream of virus nuclear transport.
Therefore, the expression of the early viral proteins E1A, E1B-55K and E2A in wt and M1 infected cells was analyzed by western blot, starting 8 h post infection (p.i.) and throughout the whole replication cycle (Figure 1B, left panel). We observed that expression of E1A in M1 virus infected cells was reduced compared to wt (Figure 1B, right panel) and accordingly, all other gene products were expressed with a delayed kinetic. This observation can be explained by the initial lower levels of E1A expression, because E1A is required for full activity of Ad downstream promoters [37]. Thus, we next investigated if the reduced E1A protein expression in M1-infected cells was due to reduced transcriptional activation of the E1A promoter following infection. We isolated and quantified newly synthesized E1A mRNA from cells infected with wt and M1 virus starting as early as 1–2 h p.i. (Figure 1C). The results confirmed that, at 1–4 h p.i., M1-infected cells showed reduced levels of newly synthesized E1A mRNA compared to wt-infected cells. Interestingly this reduction was gradually compensated throughout the first hours of infection (Figure 1C, compare 1–2 h, 3–4 h and 5–6 h) suggesting that low levels of initially made E1A were sufficient to compensate for the M1-defect in E1A transcription.
The high particle per cell ratio requirement for transcriptional activation and the reduced levels of E1A mRNA and E1A protein expression for the M1 virus indicated that the PPxY motif in protein VI not only affects transport towards the nucleus, but also early viral gene expression, presumably through separate mechanisms.
We previously showed that protein VI contains nucleo-cytoplasmic transport signals [38]. To test if protein VI could play a direct role in the initial activation of the viral genome, we first analyzed whether protein VI from incoming Ad capsids is imported into the nucleus. Using nucleo-cytoplasmic fractionation, we observed rapid protein VI accumulation in the nuclear fraction after infection (Figure 2A).
Fractionation does not discriminate between nuclear (inside) or nucleus-associated (outside) accumulation of protein VI (e.g. capsid-associated at the microtubule organizing center). Thus, we investigated the subcellular localization of protein VI derived from entering viral particles by confocal microscopy in synchronous infected cells. Within one hour, we observed protein VI specific signals in dot-like structures inside the nucleus for wt- and the M1-virus. Using antibodies (Ab) against PML, we showed some protein VI associated with PML-NBs (Figure 2B).
We confirmed the association of some protein VI with PML-NBs in a virus free system by transfecting protein VI-mRFP alone or together with EGFP-PML expressing plasmids into U2OS cells. Transfected proteins were detected via the mRFP and EGFP signal or with specific Ab for endogenous PML (“endogenous” highlighted throughout the text and in figures by the suffix “e”, e.g. ePML). The results show that protein VI was able to independently associate with PML-NBs (Figure 2C). Using a serie of protein VI mutants, we mapped the region of protein VI required for PML-NB association (Figure S3). This analysis revealed that the N-terminal amphipathic helix was required for efficient PML-NB targeting, because a mutant (VI-delta54) deleted of the amphipathic helix showed a diffuse nuclear distribution (Figure S3). We repeatedly observed the clustering of PML in transfected cells, suggesting PML-NB structure modulation resulting from protein VI expression. In summary, these data showed that some protein VI from incoming Ad particles is targeted into the nucleus, where some of it consistently localizes adjacent to PML-NBs, suggesting an involvement in additional intranuclear steps.
It was recently reported by some of the co-authors of this work that the transient PML-NBs resident factor Daxx suppressed Ad replication and was degraded late in the infection cycle [21]. The observation that some protein VI was associated with PML-NBs prompted us to investigate whether PML itself, or PML-NB-associated factors such as Daxx, interact with protein VI. These interactions could provide an explanation for the reduced transcription of the E1A promoter observed for the M1 virus. Cells were infected with HH-Ad5-VI-wt or -VI-M1 and harvested after 24 h. Lysates were subjected to immunoprecipitation (IP) using PML or Daxx specific Ab and analyzed by western blot (Figure 3A). The data showed that protein VI could be precipitated from both wt and M1 infected cells using either PML or Daxx specific Ab. In contrast to virus infected cells, we did not detect co-precipitated protein VI following cotransfection and IP with different PML isoforms, suggesting an indirect association of PML and protein VI, presumably bridged by other viral or infection induced factors (Figure 3B). In contrast, co-IP of protein VI with Daxx also occurred after isolated transfection of protein VI-wt as well as protein VI-M1 suggesting that the interaction is independent of other viral factors (Figure 3C). We next asked whether Daxx interaction with protein VI could explain the reduced replication of HH-Ad5-VI-M1. For these assays, we used the hepatoma derived cell line HepaRG, because of its close resemblance to primary cells [39], and HepaRG cells depleted of Daxx (HAD, Daxx was depleted with shRNA expressing lentiviral vectors [20]). We infected Daxx-depleted HAD and HepaRG parental cells with HH-Ad5-VI-wt and HH-Ad5-VI-M1 and determined virus yields and gene expression at 12, 24 and 72 h p.i. (Figure 3). The M1 virus was more strongly attenuated in HepaRG cells than in U2OS cells (compare to Figure 1), while Daxx depletion strongly enhanced virus production for both viruses and nearly restored the M1 virus yields to wt levels (Figure 3D). This improvement of Ad permissivity was confirmed by an increase of expression of all analyzed viral genes, including gene products from the E1A and E1B promoters (Figure 3E).
The data showed that Daxx depletion was sufficient to increase Ad gene expression for both viruses, emphasizing the role of Daxx in viral genome repression. In addition, wt but not M1 mutant protein VI could counteract Daxx mediated inhibition indicating that the PPxY motif of protein VI plays a significant role in initiating viral gene expression.
Next, we asked whether the Ad immediate early E1A and early E1B promoters are targeted by Daxx mediated repression and if this is the case whether it can be reversed by protein VI. To this end, we constructed luciferase expression vectors controlled by the Ad E1A and E1B promoters and measured luciferase expression in protein VI-wt or protein VI-M1 transfected H1299 cells (Figure 4A). Unlike VI-M1, VI-wt was able to stimulate expression from the E1A promoter ∼2.5-fold and ∼1.5-fold from the E1B promoter (Figure 4A). To show direct association of protein VI with E1 promoters, we performed chromatin immunoprecipitation assays (ChIP) at 48 h p.i from M1- or wt virus infected cells, using protein VI specific serum and Ad promoter-specific primers (Figure 4B). The results show that the VI-wt protein was much more strongly associated with the E1A and E1B promoter in infected cells than the VI-M1 protein, which is also reflected in their relative activation ability (Figure 4B, compare with 4A). To analyze whether protein VI associated activation of Ad early promoters is involved in Daxx de-repression, we cotransfected the E1B promoter driven luciferase expression vector in absence or presence of Daxx with protein VI-wt or VI-M1 expression vectors. Protein VI-wt, but not VI-M1, alleviated Daxx repression implying a role for the PPxY motif (Figure 4C). Although there was less binding to protein VI compared to the E1A promoter, we observed a strong effect on the activation of luciferase expression in that experiment. We also tested if protein VI (wt or M1) stimulates other Ad promoters using luciferase expression vectors for all viral promoters. The data showed that protein VI-wt was able to stimulate most of the Ad promoters in absence of other viral factors to various degrees (Figure S4). The strongest induction was observed for the immediate early E1A and E2A early promoter, which is in agreement with the weak E2A expression observed in HepaRG cells in M1-virus infected cells and the restoration of E2A expression following Daxx depletion (see Figure 3E). In contrast, E3 and E4 promoter activation was weak with no clear difference between wt and M1. In the context of an ongoing virus infection, the transcriptional activation of both promoter groups (E1/E2 vs. E3/E4) was shown to be regulated by E1A but via different mechanisms [40], [41]. Thus, our data showed that protein VI might also play a minor role in the transcriptional activation of the E1/E2 promoter group.
Altogether, the promoter analysis suggests that protein VI plays a so far not recognized role in the Ad gene expression program.
We next asked how the PPxY motif of protein VI contributes to Daxx de-repression. In previous work, we showed that this motif mediates protein VI interaction with cytoplasmic Nedd4 ubiquitin ligases [36]. Overexpression of protein VI and/or Nedd4 did not result in a change of steady-state Daxx levels (data not shown) suggesting that de-repression was not achieved through Daxx degradation as e.g. as shown for HCMV. However, when we tested if protein VI targets Nedd4 ligases to PML-NBs our analysis showed that protein VI-wt, but not VI-M1 targets Nedd4 ligases towards PML-NBs. This targeting required the PPxY motif and the amphipathic helix, but was independent of catalytical Nedd4 activity suggesting that Nedd4 ligases could be involved in other steps of counteracting Daxx repression by protein VI (Figure S5).
As a next step, we therefore analyzed whether the subcellular distribution of Daxx was altered in response to protein VI and Nedd4 expression. In non-transfected cells, endogenous Daxx (eDaxx) is nuclear in steady state with some Daxx localizing to dot-like intranuclear structures resembling PML-NBs (Figure 5a). When we transfected expression vectors for protein VI-wt or VI-M1 into U2OS cells, nuclear localization of eDaxx was lost and eDaxx colocalized with transfected protein VI in the cytoplasm (Figure 5b and e). In contrast, following transfection of expression vectors for protein VI-wt and Nedd4 ligases, eDaxx remained nuclear and instead protein VI-wt colocalized with Nedd4 ligases in the cytoplasm (Figure 5c). When we transfected expression vectors for Nedd4 ligases and protein VI-M1, protein VI retained the capacity of translocating eDaxx to the cytoplasm (Figure 5f). These data suggested that binding of Nedd4 to the PPxY motif of protein VI efficiently competed with protein VI-dependent cytoplasmic translocation and/or cytoplasmic retention of Daxx. This effect did not require Nedd4 ubiquitin ligase activity (Figure 5d). Thus, our results suggested that the PPxY motif present in wt protein VI could influence the dynamic nucleo-cytoplasmic distribution of Daxx.
To continue our analysis in a more physiological setting, we analyzed the subcellular localization of Daxx during Ad entry (Figure 6). In uninfected control cells, Daxx localized to the nucleoplasm and into PML-NBs. Within the first hour of infection, Daxx remained largely nuclear in wt- as well as M1-virus infected cells. Occasional cytoplasmic Daxx was never virus particle-associated. In contrast to non-infected cells, we observed a trend towards intranuclear displacement of Daxx from PML-NBs and PML clustering following infection (Figure 6A, red arrows), which could be clearly distinguished from Daxx spots in uninfected cells. This suggests that incoming viruses displace Daxx from PML-NBs by a mechanism independent of the PPxY motif of protein VI and prior to initial viral gene expression. Because we noticed occasionally large PML-NBs in infected cells, we next quantified the number of PML-NBs in wt- and M1-infected cells compared to non-infected cells. The results showed that on average, infected cells had less PML-NBs than non-infected cells, supporting our observation that PML-NBs were clustering (Figure 6B) and that the effects where PPxY motif independent. To show that the Daxx displacement from PML-NBs in the very early infection phase was caused by protein VI, we analyzed Daxx dissociation from PML-NB also in VI-wt and VI-M1 transfected cells (Figure S6). Compared to non-transfected cells, expression of protein VI-wt or VI-M1 led to translocation and cytoplasmic colocalization of Daxx (as seen in Figure 5). In addition, in several cells, Daxx was partially or completely displaced from PML-NBs and PML formed large nuclear clusters similar to those observed in infected cells (Figure S6, red arrows). We also transfected cells with expression vectors for HCMV pp71 tegument protein, known to interact with Daxx [42]. Unlike for protein VI, in pp71 transfected cells, Daxx remained nuclear and localized to some degree with PML into pp71 induced, ring-like structures also partially displacing Daxx from PML-NBs (Figure S6).
To directly follow Daxx displacement from PML-NBs and from the nucleus, we used microinjection of recombinant protein VI (Figure 7 and Videos S1, S2, S3). We transfected U2OS cells with Daxx-mCherry and PML-GFP expression constructs, and injected the cytoplasm with either control buffer, recombinant VI-wt or with recombinant VI-M1 (Figure 7B) and followed the distribution of Daxx-mCherry using live-cell imaging (Figure 7A). Daxx-mCherry was exclusively localized to the nucleoplasm and PML-NBs, while PML-GFP showed an intranuclear dot-like distribution with some cytoplasmic aggregates at higher levels of expression. Cytoplasmic injection of protein VI-wt or VI-M1 led to displacement of Daxx from PML-NBs and cytoplasmic accumulation of Daxx within minutes of injection (Figure 7A, first and second row compared to buffer controls in the last row). We quantified the cytoplasmic accumulation of Daxx by measuring nuclear Daxx fluorescence loss following microinjection. This quantification revealed that Daxx nuclear export occurred more rapidly post injection of protein VI-wt than VI-M1, suggesting that the PPxY motif accelerated the process of Daxx displacement (Figure 7C). Notably, Daxx displacement was paralleled by a strong increase in intranuclear mobility of PML-GFP and by fusion events between individual bodies (Videos S1 and S2), thus providing evidence that the large clustered PML-NBs, observed in fixed cells, result from the mobilization of Daxx out of the bodies.
We also microinjected recombinant protein VI (VI-delta54), lacking the amphipathic helix required for PML-NB targeting of protein VI, to see whether PML-NBs association is required for Daxx displacement. In contrast to protein VI-wt and VI-M1, injection of VI-delta54 only transiently displaced Daxx from PML-NBs and did not result in Daxx cytoplasmic translocation (Figure 7A third row and Video S3). The Daxx residence time in PML-NBs is ∼2 seconds [43]. Therefore our observation could be explained by competitive binding of VI-delta54 to Daxx, which could transiently prevent Daxx from association with PML-NBs. In summary, these data strongly suggested that protein VI from incoming adenoviral capsids can displace Daxx from PML-NBs, which in turn affects the PML-NB architecture leading to the accumulation of PML in large intranuclear clusters. Our analysis further indicate that association of protein VI with PML-NBs through the amphipathic helix is not strictly required for Daxx displacement from PML-NBs and that the PML-NB rearrangements take place prior to or are concomitant with the initiation of adenoviral transcription.
Our data showed that protein VI activates the Ad E1 promoters by reversing Daxx repression, presumably until newly synthesized E1A can secure the Ad gene expression program. In this case, virion proteins derived from other DNA viruses known to abrogate Daxx repression should be able to substitute this function. To test this possibility, we tested whether the expression from the E1A promoter can be activated by the HCMV pp71 tegument protein or by the HPV L2 minor capsid protein, which both target Daxx [26], [44]. Similar to protein VI-wt, pp71 and L2 were able to stimulate the Ad E1A promoter (Figure 8A). Furthermore, we observed that like protein VI-wt, pp71 and L2 could also drive efficient E1A and E1B expression from a subviral construct, preserving the virus context encoding the E1A and E1B transcription units (Figure 8B, lane 3, 6 and 7). These results show that non-adenoviral virion proteins are also capable of inducing immediate early adenoviral gene expression in the absence of any further Ad protein. This induction of gene expression was through mediating transcriptional activation, as shown by elevated E1A and E1B mRNA levels (Figure 8C). Similarly, this result confirmed that elevated E1A mRNA and protein expression levels driven by protein VI require the PPxY motif, thus directly linking entry and early viral gene expression (Figure 8B, lanes 1–4). To extend the analysis for other regions of protein VI, we used the expression construct encoding protein VI-delta54, lacking the amphipathic helix, which is required to target protein VI to PML-NBs (Figure S3d). The results showed that like protein VI-M1, the construct expressing VI-delta54 only marginally stimulated the E1A promoter (compare wt-, M1 and delta54 in Figure 8A and C). In contrast, the expression of protein VI-delta54 resulted in somewhat elevated protein expression levels compared to VI-M1 suggesting that it might promote E1A expression on a post-transcriptional level. This could result from the diffuse localization of VI-delta54 in the nucleoplasm of transfected cells (compare with Figure S3). In summary, this analysis showed that efficient transcriptional activation of the E1A promoter requires the amphipathic helix in addition to the PPxY motif.
If the HCMV tegument protein pp71, that is known to remove Daxx repression from the immediate early HCMV promoter [45], activates the Ad E1A promoter, it was conceivable to speculate that protein VI would also be able to stimulate the immediate early HCMV promoter. To test this hypothesis, we constructed viral vectors encoding wt- or M1-mutated protein VI where the E1 region was replaced by a HCMV promoter controlled GFP (wt) or mCherry (M1) expression unit. We transduced U2OS cells with M1-vectors and increasing amounts of wt virus and quantified gene expression using fluorescent activated cell sorting. The results showed partial restoration of the (HCMV promoter controlled) marker gene expression from VI-M1 vector transduced cells only in cells that were co-transduced with the M1-vector and the wt-vector (Figure S7). This analysis suggested that protein VI stimulated the HCMV promoter in trans, like pp71 could stimulate the Ad E1A promoter in trans (Figure S7). Taken together the effects that protein VI has on the E1A promoter are comparable, and moreover compatible and interchangeable, with the HCMV or papillomavirus virion derived immediate early enhancing activities.
Because protein VI, pp71 and L2 can stimulate Ad E1A expression independently, we next asked if they could compensate for the lack of functional PPxY motif in the replication competent HH-Ad5-VI-M1 virus. We transfected cells with expression vectors for protein VI-wt, VI-M1 and VI-delta54 (Figure 9A) and HCMV tegument protein pp71 and HPV small capsid protein L2 (Figure 9B) followed by infection with HH-Ad5-VI-wt or HH-Ad5-VI-M1 virus. The analysis showed that protein VI-wt was able to fully compensate for the M1 mutation in the virus and restored progeny virus production to wt levels, while protein VI-M1 was not able to rescue virus production and VI-delta54 resulted only in partial rescue (Figure 9A). Amazingly, HCMV pp71 and HPV L2 were also fully capable of complementing the M1 mutant virus and restored progeny virus production to wt levels (Figure 9B). Lastly, we wanted to know if the adenoviral protein VI capsid protein was also able to stimulate an immediate early promoter in the context of a non-related virus infection. We transfected U2OS cells with protein VI-wt and VI-M1 or a control vector and infected the transfected cells with a murine cytomegalovirus (MCMV) expressing luciferase under the control of the HCMV immediate early promoter (MCMV-Luc). Luciferase expression was measured 2 h after a synchronized infection to quantify the activation of the immediate early promoter. The results showed that only protein VI-wt was able to stimulate immediate early promoter in the context of MCMV infection (Figure 9C).
Taken together these results showed that protein VI promotes immediate early gene expression from the adenoviral E1A promoter, but it was also able to act on the immediate early gene expression of a non-related virus.
In summary, our analysis provides an intriguing mechanistic basis for cross genome activation of at least three unrelated DNA viruses. Our data suggest that initiation of viral gene expression can be achieved in cases where the respective virion proteins of one virus are capable of removing Daxx dependent transcriptional repression from the genome of the other virus.
Here, we show that the capsid protein VI is necessary for efficient initiation of Ad gene expression by activating the E1A promoter and promoting initial expression of the E1A transactivator, a function that had not been previously identified. E1A is a crucial global transcriptional activator promoting early adenoviral gene expression [37]. We show that E1A transcription and E1A protein expression at the onset of viral gene expression are reduced when cells are infected with an Ad mutant in which the PPxY motif in the capsid protein VI is inactivated. E1A mRNA production in this mutant increases with time and reaches wildtype levels, suggesting that newly expressed E1A compensates for the mutation in protein VI and drives adenoviral gene expression as soon as critical concentrations have been reached [37]. In addition, protein VI also stimulates other E1A dependent Ad promoters in the absence of any viral protein suggesting that it may act as a capsid derived E1A surrogate prior to the onset of E1A expression. Thus, protein VI is an important regulator of viral gene expression and links virus entry to the onset of gene expression. This is at least in part mediated by counteracting transcriptional repression imposed by the cellular Daxx protein and can be substituted by functionally homologous capsid proteins from unrelated DNA viruses.
In the nucleus, Daxx associates with chromatin and PML-NBs. PML-NB association with Daxx is thought to alleviate gene repression and activate apoptosis, while chromatin bound Daxx is thought to act in a transcriptionally repressive manner [7], [46], [47]. A dynamic equilibrium of Daxx between PML-NBs and chromatin association may thus govern the response status of the host cell upon infection. Moreover, an antiviral interferon response increases expression of PML and sensitizes cells for apoptosis. Artificial knock down of PML increases replication of Ad and other viruses, an observation that supports antiviral functions of PML [reviewed in 4], [21]. However, PML knock down also decreases Daxx steady state levels by an unknown mechanism, showing that antiviral activity might be mediated by Daxx rather than PML [21]. This would be in line with our observation that Daxx knock down has much stronger pro-replicative effects on Ads.
Here we demonstrate that Daxx directly represses Ad E1 promoters. So far, it has been shown that Daxx inactivates the major immediate early promoter of HCMV [45], is recruited to HSV genomes via SUMO dependent pathways [48] and is likely to associate with incoming avian sarcoma virus (ASV) and human immunodeficiency virus (HIV) genomes [49], [50]. Therefore, Daxx could act as a cytoplasmic and/or nuclear DNA sensor and may be part of a cellular innate defence mechanism against DNA virus infection (or other pathogens) by simply assembling repressive complexes on incoming DNA [51]. This is supported by two recent studies showing that Daxx selectively represses procaryotic DNA expression [52] and that frequent epigenetic silencing of integrated retroviral genomes could be reversed by Daxx depletion, showing epigenetic control of pathogen DNA by Daxx associated mechanisms [53]. Daxx mutants that fail to associate with the HSV genome also fail to induce repression on the HSV genome, underlining the important role of Daxx as part of the cellular innate antiviral defence mechanism [48].
If Daxx serves in antiviral intrinsic immunity to repress viral genomes, virion proteins are viral countermeasures. Several structural proteins from viral particles have been reported to interact with Daxx, including tegument protein pp71 [HCMV]; [ 42,54], minor capsid protein L2 [HPV; 26], DENVC [Dengue virus; 55], p6 [HIV GAG; 56], nucleocapsid protein PUUV-N [Hantavirus; 57], Integrase [ASV], [ HIV; 49,53] and protein VI (Ad, this study).
The best studied is the tegument protein pp71 of HCMV, which enhances infectivity and replication through activation of the immediate early promoter. This requires colocalization of the viral genome with PML-NBs and Daxx degradation via pp71 [23], [42], [44], [58], [59]. In addition, pp71 was also shown to activate gene expression from HSV-1, a different herpesvirus, showing that its function is not restricted to HCMV [60]. Unlike for HCMV, degradation of Daxx [through E1B-55K; 21] during Ad infection requires early gene expression. Here we observe quantitative removal of Daxx from PML-NBs upon infection without degradation before gene expression is established. We propose that this is caused by protein VI derived from the entering capsid, which partially associates with PML-NBs during entry. Similar to what we observe early in infection, transfected protein VI also displaces Daxx from PML-NBs and translocates it into the cytoplasm. Similarly, microinjected protein VI leads to rapid exclusion of Daxx from PML-NBs and cytoplasmic accumulation suggesting active removal following protein VI nuclear import. Deletion of the N-terminal amphipathic helix from protein VI, which serves as PML-NB targeting domain, still mediated the transient dissociation of Daxx from PML-NBs suggesting that competitive binding and a short residence time of Daxx in PML-NBs can also cause Daxx removal from PML-NBs [43]. Daxx depletion from PML-NBs also provokes intranuclear mobility and clustering of PML, reminiscent of infected cells and showing that Daxx contributes to the integrity of PML-NBs, which confirms previous observations [10].
Ad-wt, but not a virus with the mutated PPxY-motif in protein VI, counteracts Daxx repression for efficient viral gene expression. Protein VI wt also induces a more rapid Daxx displacement from PML-NBs and subsequent nuclear export than its mutated counterpart. In contrast, binding of Nedd4-family ubiquitin ligases to the PPxY of protein VI abolished cytoplasmic translocation of Daxx at steady state, suggesting that Nedd4 binding to protein VI competes with the interaction between Daxx and protein VI. Increasing the efficiency of Daxx mobilization in the nucleus, and simultaneously preventing Daxx nuclear export or limiting the time Daxx resides in the cytoplasm through competitive binding to Nedd4, could lead to efficient derepression and prevent Daxx from activating apoptosis (via JNK pathways), which could explain why Nedd4 binding is beneficial for the virus [10], [61].
Displacing Daxx from PML-NBs immediately after virus entry prevents antiviral apoptotic processes, possibly increasing Daxx mediated repression by epigenetic silencing [reviewed in 5]. We observe Daxx removal from PML-NBs for wt- as well as M1 mutated protein VI. In contrast, only wt-VI shows a strong stimulation and direct association with viral E1 promoters as determined by ChIP. In addition, proper transcriptional activation of the E1A promoter required the presence of the amphipathic helix. Thus, reversal of Daxx repression by protein VI from viral promoters might provide an additional explanation for Nedd4 function and the role of the PPxY motif. Targeting Nedd4 to viral promoters via the PPxY could result in ubiquitylation of histone or the histone-like DNA bound viral protein VII or other Daxx interactors, to open the chromatin structure for transcription.
In this scenario, protein VI would prevent formation or disassemble already bound repressive complexes from viral promoters via the PPxY motif and Nedd4. This would explain why the M1 mutant still displaces Daxx from PML-NBs, but retains only a minor capacity of stimulating viral gene expression presumably through interfering with the assembly of new Daxx repressive complexes. This model would also support the observation that, like protein VI-M1, protein VI without amphipathic helix (but intact PPxY and still capable of Daxx binding) hardly stimulates the E1A promoter. This mutant is diffusely distributed in the nucleus showing that the helix contributes to proper intranuclear targeting of protein VI. Mislocalization therefore could reduce the capacity to remove or prevent assembly of Daxx repressive complexes on the E1A promoter. How this mutant still retains some capacity of stimulating E1A protein expression (and as a consequence partially rescues the M1-virus) without activating the E1A promoter is currently unclear.
Removal of Daxx by components of incoming virions to initiate gene expression is a common viral strategy. Our experiments are the first to show that the consequences are not virus-family specific, but provoke global changes in transcriptional activity that allow transcriptional activation of one viral genome (here the Ad or MCMV genome) by the virion protein of unrelated viruses (here pp71 from HCMV and L2 from HPV; or the CMV promoter by protein VI). All three virion proteins (VI, pp71 and L2) target Daxx repressive complexes. The details of these interactions are not fully understood but they share similarities as highlighted in the model in Figure 10. We suggest that activation of viral gene expression for the three viral systems (Ad, HCMV and HPV) involves prevention and removal of Daxx repressive complexes. This is achieved by preventing Daxx-PML interaction or association of Daxx repressive complexes with the viral genome and (in some cases) involves the degradation of components of the complex (Figure 10). Neither pp71 nor L2 contain a PPxY motif suggesting different modes of action on Daxx or components of the Daxx repressive complex. Protein VI is also not restricted to Ads in its de-repressive activity and is able to stimulate the immediate early HCMV promoter. Several other viral capsid proteins have been reported to encode PPxY motifs [reviewed in 62]. The research focus for those motifs has been on their role in virus budding despite the presence of these proteins in the capsids of many viruses during virus entry. If virion derived PPxY (and related motifs) are part of a more general activation mechanism for several viruses then this could also mean that co-infections with different viruses, frequently observed in vivo, could promote each other. Similarly it is an interesting question, whether superinfections of a latently infected cell by another de-repressive virus would support reactivation of the latent genome. Epidemiological data from a recent study show that Ad/HCMV co-infections in vivo happen as often as mono-infections and the authors suggest that this could reflect co-viral reactivation [63]. Our data would provide a mechanistic basis for this observation, which is potentially applicable to several types of viral co-infections.
Lastly, we believe that gene regulatory functions of viral structural proteins should be considered when addressing safety issues for the application of viral vectors (e.g. adenoviral vectors) in therapeutic settings where (re)activation of unrelated (latent) viruses is unwanted.
U2OS, H1299 and HepaRG cells were grown in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum (FCS), 100 U of penicillin, 100 µg of streptomycin per ml in a 5% CO2 atmosphere at 37°C. For HepaRG and HAD (Daxx knock down) cells media was supplemented with 5 µg/ml of bovine insulin and 0.5 µM of hydrocortisone [20], [21].
Tagged protein VI, PML, Daxx and Nedd4 expression vectors have been described previously [36], [64]. E1A was expressed from constructs encompassing the left part of the viral genome including left inverted terminal repeat (ITR) and the E1 genes (pPG-S3). N-terminal flag-tagged human PML-isoforms I-VI were expressed from pLKO.1-puro vector (kindly provided by R. Everett). Codon optimized HPV (type 16) L2 expression vector was kindly provided by M. Mueller, DKFZ Heidelberg. Expression vector pCGN71 [65] encodes an XbaI-BamHI PCR fragment 0corresponding to the HCMV strain AD169 UL82. Dual luciferase assays were performed according to manufacturers instructions and have been described previously [66]. Promoter constructs are based on the pGL3-basic vector (Invitrogen, cloning details will be provided upon request).
E1-deficient viral vectors BxAd5-VI-wt-GFP and BxAd5-VI-M1-mCherry are based on human Ad serotype 5 and have been cloned using homologous and site-specific recombination using bacterial artificial chromosomes (BACs) as described in detail recently [36]. Replication competent wt virus HH-Ad5-VI-wt is identical to the previously described H5pg4100 [67]. The virus mutant HH-Ad5-VI-M1 carries an altered PPxY motif in the protein VI open reading frame [PPSY = >PGAA; Fig. S1; [36]]. Viruses were constructed, propagated and titrated on HEK293 cells as detailed in Figure S1.
For immunofluorescence analysis cells were washed in PBS and fixed for 20 min using 4% paraformaldehyde. Detection of endogenous antigens using primary and secondary Ab was done in IF-buffer (PBS with 10% FCS and 0.2% Saponin) followed by washing and embedding in Prolong Gold (Invitrogen). A list of primary and secondary Ab used in this study is given in Protocol S1 in Text S1. Images are presented as maximum image projections if not indicated otherwise. For protein analysis total-cell lysates were prepared and analyzed by western blot using standard protocols. The list of the antibodies used in this study and details for immunoprecipitation (IP) procedures are given in Protocol S1 and Protocol S2 in Text S1.
H1299 cells were infected with HH-Ad5-VI-wt or HH-Ad5-VI-M1 at 50 fluorescence forming units/cell (FFU/cell) and harvested 24 h p.i. ChIP analysis was performed as described previously with some modifications [68], [69]. For ChIP, proteins from 2×106 cells were cross-linked to DNA with 1% formaldehyde in PBS for 10 min at room temperature. The reaction was quenched and cells were washed with PBS and harvested by scraping off the dish. Nuclei were isolated by incubation of cross-linked cells with 500 µl buffer I (50 mM Hepes-KOH, 140 mM NaCl, 1 mM EDTA, 10% glycerol, 0.5% NP-40, 0.25% Triton X-100) for 10 min on ice and pelleted by centrifugation. The nuclei were subsequently washed with 500 µl buffer II (10 mM Tris-HCl, 200 mM NaCl, 1 mM EDTA, 0.5 mM EGTA), pelleted again and resuspended in 500 µl buffer III (1% SDS, 10 mM EDTA, 50 mM Tris-HCl). Chromatin was fragmented by sonication using a Bioruptor (Diagenode) to an average length of 100–300 bp. After addition of 10% Triton X-100, cell debris were pelleted by centrifugation (20,000× g, 4°C) and supernatants were collected. Chromatin was diluted with dilution buffer (0.01% SDS, 1.1% Triton X-100, 1.2 mM EDTA, 16.7 mM Tris-HCl, 167 mM NaCl). To reduce non-specific background, chromatin was pre-incubated with salmon-sperm DNA protein-A agarose beads (Upstate). Antibodies were added and incubated for 16 h at 4°C. Fifty µl agarose beads were added to precipitate the chromatin-immunocomplexes for 4 h at 4°C. Beads were washed once with low-salt buffer (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl, 150 mM NaCl), once with high-salt buffer (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl, 500 mM NaCl), once with LiCl-wash buffer (0.25 M LiCl, 1% Nonidet P-40, 1% Na-deoxycholate, 1 mM EDTA, 10 mM Tris-HCl) and twice with TE buffer. Chromatin was eluted from the beads in elution-buffer (50 mM Tris-HCl pH 8.0, 10 mM EDTA, 1% SDS) for 10 min at 95°C. Proteinase K was added for protein degradation and samples were incubated for 1 h at 55°C. For preparation of input controls, samples were treated identical to IP samples except that non-specific Ab were used. qPCR analysis was performed using a Rotor Gene 6000 (Corbett Life Sciences, Australia) in 0.5 ml reaction tubes containing 1/100 dilution of the precipitated chromatin, 10 pmol/µl of each synthetic oligonucleotide primer (E1A fwd 5′TCCGCGTTCCGGGTCAAAGT3′; E1A rev5′GTCGGAGCGGCTCGGAG3′; E1B fwd 5′GGTGAGATAATGTTTAACTTGC3′ ¸ E1B rev 5′TAACCAAGATTAGCC CACGG3′), 5 µl/sample SYBR Green PCR Master Mix (Applied Biosystems). The PCR conditions used: 7 min at 95°C, 45 cycles of 12 s at 95°C, 40 s at 60°C and 15 s at 72°C. The average Ct-value was determined from triplicate reactions and normalized against non-specififc IgG controls with standard curves for each primer pair. The identities of the products obtained were confirmed by melting curve analysis. For qPCR analysis, U2OS cells were infected with 1, 10 and 200 physical particles/cell and genome copy numbers were determined in nuclear and cytoplasmic fractions using hexon specific primers [70].
4sU (Sigma) was added to the cell culture media for 1 h, made up to a final concentration of 200 µM, during indicated time points throughout infection. Cells were harvested using Trizol reagent (Invitrogen) and total RNA isolated by phenol-chloroform extraction. Biotinylation and purification of 4sU-tagged RNA (newly transcribed RNA), was performed as described previously [71]. Five hundred ng of each newly transcribed RNA per reaction was reverse transcribed in 25 µl reactions using Superscript III (Invitrogen) and oligo-dT primers (Invitrogen) following the manufacturer's instructions. PCR was performed on a Light Cycler (Roche Molecular Biochemicals). Each reaction, every sample in duplicates, was carried out using 5 µl of cDNA (1∶10 dilution) and 15 µl reaction mixtures of Quantitect SYBR Green PCR master mix and 0.5 µM of the primers. PCRs were subjected to 10 min of 95°C hot-start, and SYBR Green incorporation was monitored for 45 cycles of 95°C denaturation for 10 s, 58°C annealing for 3 s, and 72°C elongation for 10 s. The data were analyzed using the ΔΔCt method using GAPDH as an endogenous reference, and the mock-infected sample as a calibrator. Values were normalized to 100% for wt-infected cells. The E1A 13S mRNA specific and the GAPDH specific primers were described in [72]. Primers used are listed below: E1A13S-fwd (5′-GGC TCA GGT TCA GAC ACA GGA CTG TAG), E1A13S-rev (5′-TCC GGA GCC GCC TCA CCT TTC), GAPDH-fwd (5′-TGG TAT CGT GGA AGG ACT CA), GAPDH-rev (5′-CCA GTA GAG GCA GGG ATG AT).
Details for microinjection are given in the Figure 8 and video legends (Video S1). Briefly, U2OS cells were cotransfected with PML-GFP and Daxx-mCherry expression plasmids and cultivated on a heated stage (37°C) in CO2 stabilized medium attached to a SP5 confocal microscope (Leica) equipped with a microinjection device (Eppendorf). Microinjected cells were imaged within a single confocal plane at the nuclear midsection at 20 s intervals for 10 frames prior to injection and 40 frames post injection. Injected proteins were purified as His-tagged proteins using standard procedures and dialyzed into transport buffer as detailed previously [30], [36].
Data are presented as mean, error bars as standard deviation (STD). Statistical analysis was done using paired students t-test except for Figure 6B where a two-tailed two sample t-test was used. The p-values are indicated.
Human Daxx CAG33366.1, Protein VI AAA96411.1, Human Adenovirus Type 5 HY339865, PML-I AAG50180, PML-II AF230410, PML-III S50913, PML-IV AAG50185, PML-V AAG50181, PML-VI AAG50184, HCMV pp71 ACZ79993.1, humanized HPV L2 (HPV16).
|
10.1371/journal.pcbi.1005359 | Memory replay in balanced recurrent networks | Complex patterns of neural activity appear during up-states in the neocortex and sharp waves in the hippocampus, including sequences that resemble those during prior behavioral experience. The mechanisms underlying this replay are not well understood. How can small synaptic footprints engraved by experience control large-scale network activity during memory retrieval and consolidation? We hypothesize that sparse and weak synaptic connectivity between Hebbian assemblies are boosted by pre-existing recurrent connectivity within them. To investigate this idea, we connect sequences of assemblies in randomly connected spiking neuronal networks with a balance of excitation and inhibition. Simulations and analytical calculations show that recurrent connections within assemblies allow for a fast amplification of signals that indeed reduces the required number of inter-assembly connections. Replay can be evoked by small sensory-like cues or emerge spontaneously by activity fluctuations. Global—potentially neuromodulatory—alterations of neuronal excitability can switch between network states that favor retrieval and consolidation.
| Synaptic plasticity is the basis for learning and memory, and many experiments indicate that memories are imprinted in synaptic connections. However, basic mechanisms of how such memories are retrieved and consolidated remain unclear. In particular, how can one-shot learning of a sequence of events achieve a sufficiently strong synaptic footprint to retrieve or replay this sequence? Using both numerical simulations of spiking neural networks and an analytic approach, we provide a biologically plausible model for understanding how minute synaptic changes in a recurrent network can nevertheless be retrieved by small cues or even manifest themselves as activity patterns that emerge spontaneously. We show how the retrieval of exceedingly small changes in the connections across assemblies is robustly facilitated by recurrent connectivity within assemblies. This interaction between recurrent amplification within an assembly and the feed-forward propagation of activity across the network establishes a basis for the retrieval of memories.
| The idea of sequential activation of mental concepts and neural populations has deep roots in the history of the cognitive sciences [1–3] as well as its share of criticism [4]. In one of the most influential works in neuroscience, Donald Hebb extended this concept by suggesting that neurons that fire simultaneously should be connected to each other, thus forming a cell assembly that represents an abstract mental concept [5]. He also suggested that such assemblies could be connected amongst each other, forming a network of associations in which one mental concept can ignite associated concepts by activating the corresponding assemblies. Hebb referred to the resulting sequential activation as well as the underlying circuitry as “phase sequence”. We will refer to such connectivity patterns as “assembly sequences”.
The notion of Hebbian assemblies has triggered a huge number of experimental studies (reviewed in [6]), but relatively few experiments have been dedicated to the idea of assembly sequences [7, 8]. Many theoretical studies focused on feedforward networks, also known as synfire chains [9–12]. Synfire chains are characterized by a convergent-divergent feedforward connectivity between groups of neurons, where pulse packets of synchronous firing can propagate through the network. Few works were also dedicated on synfire chains embedded in recurrent networks [13–15], however, without explicitly considering recurrent connectivity within groups.
In this study, we combine the concept of feedforward synfire chains with the notion of recurrently connected Hebbian assemblies to form an assembly sequence. Using numerical simulations of spiking neural networks, we form assemblies consisting of recurrently connected excitatory and inhibitory neurons. The networks are tuned to operate in a balanced regime where large fluctuations of the mean excitatory and inhibitory input currents cancel each other. In this case, distinct assemblies that are sparsely connected in a feedforward fashion can reliably propagate transient activity. This replay can be triggered by external cues for sparse connectivities, but also can be evoked by background activity fluctuations for larger connectivities. Modulating the population excitability can shift the network state between cued-replay and spontaneous-replay regimes. Such spontaneous events may be the basis of the reverberating activity observed in the neocortex [16–18] or in the hippocampus [19–21]. Finally, we show that assembly sequences can also be replayed in a reversed direction (i.e., reverse replay) as observed during replay of behavior sequences [22, 23].
To test Hebb’s hypothesis on activity propagation within a recurrent network, we use a network model of excitatory and inhibitory conductance-based integrate-and-fire neurons. The network has a sparse random background connectivity prand = 0.01 [24]. We form a neural assembly (Fig 1A) by picking M excitatory (M = 500 if not stated otherwise) and M/4 inhibitory neurons and connecting them randomly with probability prc, resulting in a mutually coupled excitatory and inhibitory population. The new connections are created independently and in addition to the background connections. To embed an assembly sequence in the network, we first form 10 non-overlapping assemblies. The assemblies are then connected in a feedforward manner where an excitatory neuron from one group projects to an excitatory neuron in the subsequent group with probability pff (Fig 1B). Thus, by varying the feedforward and the recurrent connectivities, we can set the network structure anywhere in the spectrum between the limiting cases of synfire chains (pff > 0, prc = 0) and uncoupled Hebbian assemblies (pff = 0, prc > 0), as depicted in Fig 1C.
To ensure that the spontaneous activity of the network is close to an in-vivo condition, we use Hebbian plasticity of inhibitory connections [25], which has been shown to generate a balance of excitatory and inhibitory currents in individual neurons (Fig 2A). As a consequence, spikes are caused by current fluctuations (Fig 2B), and the network settles into a state of asynchronous irregular (AI) firing (Fig 2C).
To simulate a one-shot sequence learning paradigm, we initially embed assemblies that have recurrent connectivity prc only and are not connected via feedforward connections (Fig 2, left-hand side). A stimulation of the first assembly does not evoke a replay. Then, in a sham learning event, new feedforward connections are created between subsequent assemblies followed by a short phase (∼ 5 seconds) with inhibitory plasticity turned on in order to properly balance the network. If we then stimulate the first group in the embedded assembly sequence (Fig 2C, right-hand side), the network responds with a wave of activity that traverses the whole sequence, as hypothesised by Hebb [5]. We refer to such a propagation of activity wave as replay. As excitatory and inhibitory neurons are part of the assemblies, they both have elevated firing rates during group activation. Despite the high population transient firing rates (∼ 100 spikes/sec for excitatory, and ∼ 60 spikes/sec for inhibitory neurons when using a Gaussian smoothing window with width σ = 2 ms) single neurons are firing at most one spike during assembly activation. Because excitatory neurons in an assembly transiently increase their population firing rate from 5 to 100 spikes/sec, a replay can be inferred from the large change in activity, which resembles replay in hippocampal CA networks [19]. On the other hand, interneurons have higher background firing rates of ∼ 20 spikes/sec and smaller maximum firing rates of ∼ 60 spikes/sec during replay. As a result, interneurons have a much lower ratio of peak to background activity than excitatory neurons in our model, in line with the reported lower selectivity of interneurons [26].
We chose the particular wiring scheme of discrete assemblies partly due to the resemblance of the discrete windows of activity defined by the fast oscillations during hippocampal replay: ripples during sharp-wave ripples (SWRs) and gamma cycles during theta sequences. Additionally, our approach facilitates the model description and gives a leverage for an analytical treatment. Accordingly, in Fig 2A–2C, we modeled discrete assemblies of size M = 500, which have a distinct recurrent connectivity prc = 0.1 within each assembly, and a feedforward connectivity pff = 0.04 between two assemblies in the sequence. However, in biological networks, assemblies could potentially overlap, making a clear-cut distinction between feedforward and recurrent connectivities difficult. To study assembly sequences in a more continuously wired sequence, we use an extreme case where no assemblies are defined at all. All neurons are arranged in a linear sequence, and every neuron is connected to its M = 500 neighbouring excitatory cells (M/2 preceding and M/2 succeeding) with probability prc. Recurrent connections to and from inhibitory neurons are embedded analogously in a continuous manner. To imitate the connectivity pattern from the discrete model, every excitatory neuron is connected to the M following neurons without overlapping with the recurrent connections (i.e., the range from 1 2 M to 3 2 M) with probability pff. After stimulating the first M neurons with a transient input, the whole sequence is replayed (Fig 2D). Compared to the discrete assembly sequence (Fig 2C) where the same connection probabilities were used, the replay is continuous and qualitatively similar. In what follows, however, we return to the discrete assemblies because this description facilitates a connection of simulations with an analytical treatment.
Whether an assembly sequence is replayed is largely determined by the connectivities within and between assemblies. Therefore, we first study how the quality of replay depends on the recurrent (prc) and the feedforward (pff) connectivities. The network dynamics can be roughly assigned to regimes where the connectivity is too weak, strong enough, or too strong for a successful replay. We use a quality measure of replay, which determines whether activity propagates through the sequence without evoking a “pathological” burst of activity (Fig 3). In such “pathological” cases the spatiotemporal structure of replay is often preserved while the background activity deviates from the AI state, or the whole network is involved in the events. To disregard such events, the quality measure punishes replays that (1) evoke bursting of neurons within assemblies during activation or (2) activate the whole network (for details see Materials and Methods).
Naturally, for a random network (pff = 0, prc = 0, Fig 3a) the replay fails because the random connections are not sufficient to drive the succeeding groups. In the case of uncoupled Hebbian assemblies (e.g., pff = 0, prc = 0.30), groups of neurons get activated spontaneously (Fig 3c), which is reminiscent to the previously reported cluster activation [27] but on a faster time scale. Already for sparse connectivity (e.g., pff = prc = 0.06) the assembly-sequence replay is successful (Fig 3b). In the case of denser recurrence (prc ≈ 0.10), a pulse packet propagates for even lower feedforward connectivity (pff ≈ 0.03). The feedforward connectivity that is required for a successful propagation decreases with increasing recurrent connectivity because assemblies of excitatory and inhibitory neurons can increase small fluctuations of the input through “balanced amplification” [28, 29] as summarized in Materials and methods, section “Balancing the Network”.
For high feedforward (pff ≲ 0.10) but low recurrent (prc ≲ 0.10) connectivity, the replay has low quality. In this case, excitatory neurons receive small recurrent inhibitory input compared to the large feedforward excitation, because the recurrent connection probability is lower than the feedforward one. Due to the lack of sufficiently strong inhibitory feedback within the assembly (compared to the strong feedforward excitation), the propagating activity either leads to run-away excitation (Fig 3e), also called synfire explosion [30, 31], or to epileptiform bursting (Fig 3d). When both recurrent and feedforward connectivities are high, the inhibition is able to keep the propagating activity transient (Fig 3f). However, because of the strong input each neuron is firing multiple times within a small time window. The fact that neurons in each group (except the first) are firing multiple times during a replay alters the spatio-temporal structure of the sequence. While activity propagates from one group to another, neurons do not necessary spike in order due to the many emitted spikes. Another reason to assign low quality to such replays is the fact that the network dynamics is deviating from the AI background state because neurons that are part of the sequence tend to fire almost exclusively during replays but not outside replays.
To get an analytical understanding of the network, we use a linear approximation of the network dynamics to derive conditions under which replay is successful. The key determinant for replay is an amplification factor κ ( p ff , p rc ) = r i + 1 r i, which measures how large is the rate ri+1 in group i+1 in relation to the rate in the previous group i.
In the case where the amplification factor is smaller than one (ri+1 < ri), the activity propagating through the assembly sequence will decrease at each step and eventually vanish, while for amplification larger than one (ri+1 > ri) one would expect propagating activity that increases at each step in the sequence. An amplification factor κ(pff, prc) = 1 represents the critical value of connectivity for which the replay is marginally stable, and the magnitude of activations is similar across groups. In the Materials and Methods we show that a linear model can approximate the amplification factor by
κ = c M p ff g E ( 1 + c M p rc g E ) (1)
where c = 0.25 nS-1 is a constant that fits the model to the data (see Materials and Methods). We can interpret κ as an “effective feedforward connectivity” because the recurrent connectivity (prc) effectively scales up the feedforward connectivity pff. We can match the analytical results for critical connectivities to the numerical simulation, and show a qualitative fit between the approaches (black line in Fig 3).
We note that the number of excitatory synapses that is needed for an association, M2(prc + pff), weakly depends on the position on the line κ = 1. By solving argmin p rc , p ff ∈ κ = 1 M 2 ( p rc + p ff ) we find that the minimum number of new connections required for a replay is obtained for prc = 0 because lines for which prc + pff = const have slope of −1 in Fig 3, and the slope of the line defined by κ = 1 is more negative. For example, when prc = 0.0, we need 40 new synapses; for prc = 0.05, we need 50 new synapses; and for prc = 0.2, 111 synapses are required for a new association. However, as feedforward connections might be created/facilitated on demand in one-shot learning, it is advantageous to keep their number low at the cost of higher recurrent connectivity, which has more time to develop prior to the learning. We extend this arguments further in the Discussion.
In summary, the recurrent connections within an assembly play a crucial role in integrating and amplifying the input to the assembly. This facilitation of replay is predominantly due to the excitatory-to-excitatory (E-E) recurrent connections, and not due to the excitatory-to-inhibitory (E-I) connections, a connectivity also known as “shadow pools” [31]. We tested that embedding shadow pools and omitting the E-E connectivity within assemblies has no beneficial effect on the quality of replay.
Neural systems have to deal with obscure or incomplete sensory cues. A widely adopted solution is pattern completion, that is, reconstruction of patterns from partial input. We examine how the network activity evolves in time for a partial or asynchronous activation of the first assembly.
To determine the capability of our network to complete patterns, we quantify the replay when only a fraction of the first group is stimulated by external input. If 60% of the neurons in the first group (strong cue) are synchronously activated (Fig 4A, left panel), the quality of replay is virtually the same as in the case of full stimulation (100% activated) in Fig 3. However, when only 20% of the neurons (weak cue) are simultaneously activated (Fig 4A, middle panel), we see a deterioration of replay mostly for low recurrent connectivities. The effect of the recurrent connections is illustrated in the right-most panel in Fig 4A where quality of replay is shown as a function of prc while the feedforward connectivity was kept constant (pff = 0.05).
Small input cues lead to a weak activation of the corresponding assembly. In the case of stronger connectivity (e.g., prc) this weak activity can build up and result in a replay as shown in the example from Fig 4B. The top and bottom rows of raster plots correspond to two assembly sequences with different recurrent connectivities, as highlighted by the rectangles in Fig 4A, while left and right columns show the activity during strong and weak cues, respectively. In the case of pff = 0.05 and prc = 0.10 (Fig 4B, top-right), the weak cue triggers a wide pulse packet with large temporal jitter in the first groups, which gradually shapes into a synchronous pulse packet as it propagates through the network. On the other hand, for a smaller recurrent connectivity (prc = 0.06), the 20% partial activation triggers a rather weak response that does not result in replay (Fig 4B, bottom-right).
The quality of replay depends not only on the number of neurons that are activated but also on the temporal dispersion of the pulse packet. Here, we adopt a quantification method that represents the activity evolution in a state-space portrait [10]. Fig 4C shows the time course of the fraction α of cells that participate in the pulse packet and the temporal dispersion σ of the packet as the pulse propagates through the network. The state-space representation of two assembly sequences with equal feedforward (pff = 0.05) but different recurrent connectivity are shown in Fig 4C (top: prc = 0.10, bottom: prc = 0.06). For each assembly sequence we repeatedly stimulated the first group with varying cue size α and time dispersion σ, depicted by the black dots. Depending on the strength and dispersion of the initial stimulation, the dynamics of a network can enter one of two attractor points. For high α and low σ the pulse packet propagates, entering the so-called synfire attractor (white background). On the other hand, for low α and high σ the pulse packet dies out resulting in low asynchronous firing (gray background). The black-arrow traces in Fig 4C are example trajectories that describe the propagating pulse packets from Fig 4B in the (α − σ) space.
To summarize, increasing both the recurrent and feedforward connectivity facilitates the replay triggered by weak and dispersed inputs. Recurrent connectivity is particularly important for pattern completion.
An interesting feature of assembly sequences is the potential emergence of spontaneous activations, that is, a replay when no specific input is given to the network. Random fluctuations in the network can be amplified by the feedforward structure and give rise to a spontaneous wave of propagation.
We find that spontaneous and evoked replay share various features such as sequential group activation on the background of AI network activity (Fig 5A, rasters a and b). As in the case of evoked replay, for exceedingly large connectivities the network dynamics can be dominated by epileptiform bursting activity (Fig 5A, rasters c and d).
To assess spontaneous replay, we quantify the number of replay events per time taking into account their quality, i.e., huge bursts of propagating activity are disregarded as replay. The rate of spontaneous activation increases as a function of both the feedforward (pff) and the recurrent (prc) connectivity (Fig 5A). For large connectivities (pff, prc > 0.20) the quality of the spontaneous events is again poor and mostly dominated by strong bursts (Fig 5A, raster c). The dynamics of networks with large feedforward and low recurrent connections is dominated by long-lasting bursts of activity consisting of multiple sequence replays within each burst (Fig 5A, raster d). The maximum rate of activations does not exceed 4 events per second because the inhibitory synaptic plasticity adjusts the inhibition such that the excitatory firing rate is close to 5 spikes/sec.
The starting position of spontaneous replays largely depends on the network connectivity. Sequences with low prc are seldom initiated in the first group(s), while for high prc spontaneous replays occur predominantly at the beginning of the embedded sequence. Spontaneous replays for sequences with low prc arise from noise fluctuations that are amplified mainly by the underlying feedforward connections. Fluctuations propagate through a few groups until they result in a full-blown replay. On the other hand, to explain the preference of starting position at the beginning of the sequence for high prc, we refer to the case of disconnected Hebbian assemblies (Fig 3A, panel c) that get activated by the noise fluctuations. In case of weak feedforward connectivity (e.g., pff < 0.02), these fluctuations do not always activate the following assemblies due to insufficient feedforward drive. On the other hand, for pff > 0.03 even a weak activation of an assembly will lead to a replay of the rest of the sequence. If replays were to start at random locations in the sequence, neurons in the later section of the sequence would participate in more replays than those earlier in the sequence, increasing the firing rate in these neurons. The inhibitory plasticity, which homeostatically regulates the rate, will hence increase the amount of inhibition in these later assemblies, with the effect of reducing the background activity. Because this in turn suppresses the fluctuations that trigger replays, spontaneous replays are less likely to be initiated in later assemblies.
To better characterize spontaneous dynamics, we refer to more extensive measures of the network dynamics. First, to account for deviations from the AI network state, we measure the synchrony of firing among neurons within the assemblies. To this end, we calculate the average pairwise correlation coefficient of spike trains of neurons within the same group. A low synchrony (value ∼0) means that neurons are uncorrelated, while a high synchrony (value ∼1) reveals that neurons fire preferentially together and seldom (or not at all) outside of an assembly activation. Because the synchrony builds up while activity propagates from one group to the next, a synchronization is most pronounced in the latter groups of the sequence. Therefore, we use correlations within the last group of the sequence as a measure of network synchrony (Fig 5B). The average synchrony is low (∼0) for low connectivities (pff, prc < 0.10) and increases as a function of both pff and prc. In the case of high prc, neurons participating in one assembly excite each other, and hence tend to fire together. On the other hand, for high pff, neurons within an assembly receive very similar input from the preceding group, so they fire together. This attachment of single neurons to group activity has two major consequences: first, it alters the AI state of the network, and second, it alters the stochastic behavior of the neurons, leading to more deterministic firing and bursting.
The network exhibits frequent epileptiform bursting in the case of high feedforward and low recurrent connectivities (raster plot examples in Fig 3, panel d, and Fig 5A, panel d). To assess this tendency of neurons to fire in bursts, we calculate the coefficient of variation (CV) for individual neurons’ spike trains. The average CV of neurons in the last group of the sequence exhibits Poisson-like irregular firing (CV value ∼ 1) for a large range of parameters (Fig 5C). However, for high pff (≥ 0.10) and low prc (≤ 0.10), the CV value exceeds 1, in line with irregular and bursting firing. In this parameter region, small fluctuations of activity in the first groups of the sequence are strongly amplified by the underlying feedforward connectivity, leading to ever increasing activity in the following groups (Fig 5A, panel d). Because of the variable shapes and sizes of these bursts, they are not always classified as spontaneous activations in Fig 5A. Highly bursty firing (CV > 3) and high synchrony (∼ 1) suggest that the network cannot be properly balanced.
To test whether the inhibitory plasticity can balance the network activity when assembly sequences are embedded, we measure the average firing rate in the last group of the sequence (Fig 5D). The firing rate deviates from the target rate of 5 spikes/sec mostly for high feedforward connectivity (pff ≳ 0.15). This inability of inhibition to keep the firing rate at the target value can be explained by the frequent replays that shape a stronger inhibitory input during the balancing of the network. Once the inhibition gets too strong, neurons can fire only when they receive excessive amount of excitation. Thus, in the case of high clustering, e.g., strong assembly connectivity, the inhibitory plasticity prevents the neurons from reaching high firing rates, but is unable to sustain an AI state of the network.
Further, we investigate how spontaneous and cued replay are related. The black line in Fig 5A refers to the analytical approximation for connectivities that enable evoked replay. Compared to the connectivity region of successfully evoked replays in Fig 3, the region for spontaneous replays in Fig 5 is slightly shifted to the top and to the right. Therefore, in only a narrow area of the parameter space, sequences can be replayed by external input but do not get spontaneously activated. This finding suggests that to embed a sequence with high signal-to-noise ratio of propagation, the connectivities should be chosen appropriately, in line with previous reports [32]. In what follows we show that the size of this region can be controlled by external input to the network.
We demonstrate how a small amount of global input current to all excitatory or all inhibitory neurons can modulate the network and shift it between AI and spontaneous-replay regimes (Fig 5E and 5F). In the first example, the connectivities are relatively low (pff = prc = 0.06) such that replay can be evoked (Fig 3) but no spontaneous activations are present (Fig 5A and 5E, left). After injecting a small additional current of only 1 pA into the whole excitatory population, the network becomes more excitable, i.e., the firing rate rises from 5 to 12 spikes/sec and spontaneous replays do arise (Fig 5E, right).
On the other hand, in a network with high connectivities (pff = prc = 0.12), replay can be reliably evoked (Figs 3 and 4A) and also occurs spontaneously (Fig 5A). An additional input current of 3 pA to the inhibitory population decreases the firing rate of the excitatory population from 5 to 0.33 spikes/sec and shifts the network from a regime showing frequent spontaneous replays to a no-replay, AI regime (Fig 5F, left and right, respectively). Nevertheless, replays can still be evoked as in Fig 3. Hence, the spontaneous-replay regime and the average firing rate in the AI state can be controlled by global or unspecific external current.
In summary, the balanced AI network state and successfully evoked replay of assembly sequences can coexist for a range of connectivities. For higher connectivities, the underlying network structure amplifies random fluctuations, leading to spontaneous propagations of activity between assemblies. A dynamical control of the rate of spontaneous events is possible through external input, which modulates the network activity and excitability. In the brain, such a switching between regimes could be achieved via neuromodulators, in particular via the cholinergic or adrenergic systems [33, 34].
So far, we have shown basic properties of sequences at fixed assembly size M = 500. To determine the role of this group size in replay, we vary M and the connectivity while keeping the size of the network fixed. As we have already explored how recurrent and feedforward connections determine replay individually, we now consider the case where they are equal, i.e., pff = prc = p.
Assembly sequences can be successfully replayed after stimulation for various assembly sizes (Fig 6A). Smaller assemblies require denser connectivity (e.g., p = 0.25 for M = 100), while larger assemblies allow sparser connectivity (e.g., p = 0.05 for M = 500). Moreover, assemblies as small as 20 neurons are sufficient to organize a sequence given the condition of all-to-all connectivity within and between assemblies. The analytically derived critical value of effective connectivity κ = 1 is in agreement with the numerical simulations (black line in Fig 6A).
To further characterize the network dynamics for varying group size, we measure the rate of spontaneous activations of assembly sequences in undisturbed networks driven solely by constant input. As indicated in Fig 6B, spontaneous replays occur for a limited set of parameters resembling a banana-shaped region in the (M, p) plane. The parameter region for spontaneous replays partly overlaps with that of evoked replay. Again, there is a narrow range of parameters to the right of the black line in Fig 6B for which sequences can be evoked by external input while not being replayed spontaneously. As shown above, the size of this region can be controlled by external input to the whole network (Fig 5E and 5F).
To further assess the spontaneous dynamics, we measure the firing synchrony of neurons within the last group. The synchrony grows as function of both connectivity and group size (Fig 6C). The fact that the synchrony approaches the value one for higher connectivity and group size indicates that the network dynamics gets dominated by spontaneous reactivations. The simulation results reveal that neurons always fire rather irregular with coefficient of variation (CV) between 0.7 and 1.4 (Fig 6D). Because the recurrent and the feedforward connectivities are equal (pff = prc = p), the inhibition is always strong enough and does not allow epileptiform bursting activity. This behavior is reflected in a rather low maximal value of the (CV<1.4) compared to the results from Fig 5, where the CV could exceed values of 4 for low prc. The measured firing rates in the last assembly are at the target firing rate of ρ0 = 5 spikes/sec for parameter values around and below the critical value κ = 1 (Fig 6E). However, for increasing connectivity p and increasing group size M, the firing rate deviates from the target, indicating that the inhibitory plasticity cannot keep the network fully balanced.
To conclude, the assembly size M plays an important role in the network activity. The critical values of connectivity and group size for successful propagation are inversely proportional. Thus, the analytics predicts that larger assemblies of several thousands neurons require only a fraction of a percent connectivity in order to propagate synchronous activity. However, for this to happen, the group size M must be much smaller than the network size NE. Here NE was fixed to 20,000 neurons for easier comparison of scenarios, but results are also valid for larger networks (see Materials and Methods). The good agreement between the mean-field theory and the numerical results suggests that the crucial parameter for assembly-sequence replay is the total input one neuron is receiving, e.g., the number of input synapses.
Up to this point, all excitatory synaptic connections in our model had constant and equal strengths. By encoding an assembly sequence we implicitly altered the structural connectivity by creating new synaptic connections. This case of structural plasticity can also occur when silent synapses are turned into functionally active connections upon learning [35, 36]. However, learning new associations might also be possible through a change of synaptic strength of individual connections [37, 38]. If a sequence is to be learned through synaptic plasticity, then instead of increasing the connectivity between groups of neurons, the synaptic conductances could be increased as well. To test whether these two types of plasticity are equivalent in our approach, we embed assembly sequences with various feedforward connectivities pff and various feedforward conductances g ff E, while keeping the recurrent connectivity (prc = 0.06) and recurrent conductances (gE = 0.1 nS) constant.
Numerical results show that feedforward connectivity and feedforward conductance have identical roles in the replay of a sequence. That is, the sparser the connections, the stronger synapses are required for the propagation of activity. The analytical estimate (Fig 7A, black line corresponds to κ ∼ p ff g ff E = const .) predicts that the product of pff and g ff E is the essential parameter for replay.
That this analytical prediction is fulfilled in the numerical simulations becomes clearer when we show the replay quality as a function of the feedforward connectivity and the total feedforward input p ff g ff E / g E a neuron is receiving (Fig 7B). It is irrelevant whether the number of connections are changed or their strength, what matters is their product. This rule breaks only for sparse connectivities (pff < 0.01), i.e. when the mean number of feedforward connections between two groups is low (< 5). Therefore, the number of relevant connections cannot be reduced to very low numbers.
Consistent with earlier findings, the quality of replay is high above a certain strength of the total feedforward conductance (≳ 0.05 in Fig 5B) and for pff ≥ 0.01. However, for sufficiently large feedforward input (p ff g ff E / g E > 0 . 12), the replay of sequences is severely impaired as the network is in a state of highly synchronous bursting activity (Fig 7B), which is similar to the results shown in Figs 5 and 6.
We also examined sequences that are formed by increasing existing background connections between the assemblies by a factor pff/prand, rather than by adding additional connections. Replays are possible also in this condition and they are indistinguishable from networks with increased feedforward connectivities.
The rule that the total input p ff g ff E determines the network behavior also holds for spontaneous activity. Spontaneous replay rate, CV, synchrony, and firing rate all vary as a function of the total input (Fig 7C), and only weakly as a function of the connectivity or the conductance alone. Similar to the previous results in Figs 5 and 6, for 0 . 05 ≤ p ff g ff E / g E < 0 . 10 it is possible to evoke a replay while preserving the AI state of the network. Increasing the total input beyond this value drives the network into a state of spontaneous replay with increased synchrony.
The assembly-sequence model discussed until now contains asymmetric connections, i.e., neurons from one group project extensively within the same and the subsequent group but not to the previous group. We showed that such feedforward assembly sequences are capable of propagating activity, which we call replay. Thus, the proposed model may give an insight on the replay of behavioral sequences that have been observed in the hippocampus [19]. However, further experiments revealed that sequences are also replayed in the inverse temporal order than during behavior, so-called reverse replay [22, 23]. The direction of this replay also depended on the context, i.e., when the animal was at the beginning of the path, forward replays prevailed; while after traversing the path, more reverse replays were detected (but see [39]). This suggests that the replay activity might be cued by the sensory input experienced at the current location of the animal.
As the feedforward structure adopted in the network model is largely asymmetric, the assembly sequence is incapable of reverse replay in its current form. To be able to activate a sequence in both directions, we modify the network and add symmetric connectivity between assemblies [40, 41]. The symmetric STDP window that has been reported recently in the hippocampal CA3 area in vivo [42] would allow for strong bidirectional connections. In such a model, an assembly of neurons does not project only to the subsequent assembly but also to the preceding, and both projections are random with probability pff (Fig 8A). While this connectivity pattern decreases the group clustering and makes the sequence more continuous, it does not lead to full merge of the assemblies because the inhibition remains local for each group.
Interpreting this network as a model for hippocampal activity during spatial navigation of a virtual rat on a linear track (Fig 8B, top), we test the idea that external input can switch the network between a spontaneous-replay state during rest and a non-replay, spatial-representation state during locomotion. During immobility at the beginning of the track, a context-dependent input cue is mimicked by a constant current Ie = 2 pA injected into the excitatory neurons of the first assembly (Fig 8B, red bar from 0 to 500 ms). The elevated firing rate of the first assembly results in a spontaneous forward replay, similar to the experimental findings during resting states at the beginning of a linear track [22, 23].
After the initial 500 ms resting period, an external global current of −10 pA is injected into the whole excitatory population to decrease network excitability and to mimic a state in which the rat explores the environment. In addition, to model place-specific sensory input that is locked to theta oscillations, we apply a strong and brief conductance input (as in Fig 2) every 100 ms to the assembly that represents the current location. In this situation, the assemblies fire at their corresponding locations only. There is, however, a weak activation of the neighboring assemblies that does not result in a replay. An extension of the model including lateral inhibition and short-term plasticity would possibly enable theta sequences that span in one direction only [43]. Such an extension is, however, beyond the scope of the current manuscript.
At the end of the track, we retract the global external current to return to the virtual resting state for the last 500 ms of the simulation, and the network switches back to higher mean firing rates. A context-dependent sensory cue to the last group (Ie = 2 pA current injected continuously) then leads to a spontaneous reverse replay, similar to experimental findings at the end of a linear track [22, 23].
In the absence of a context-dependent current injection during virtual resting state, spontaneous replays start at around the middle of the sequence (as in Fig 5) and propagate in forward or reverse direction. As noise fluctuations are gradually amplified while propagating between assemblies, it is rare to find a spontaneous event that is simultaneously replayed in both directions. In our simulations (Fig 8), we assumed that the starting position of replay is cued by the sensory input from the current location. However, it has been shown that replays during theta sequences are rather segmented and represent the environment in discrete “chunks” [44]. These segments are not uniformly distributed but tend to cover the space between physical landmarks, noteworthy positions in the environment. The finding of Gupta et al. [44] suggests that there might be other mechanisms controlling the starting position of replay other than the sensory input. Currently, it is an open question whether SWR replays represent the environment also in a segmented manner from a landmark to a landmark.
In summary, we show that given symmetric connectivity between assemblies, transient activity can propagate in both directions. Large negative external currents injected into all excitatory neurons can decrease network excitability and thus block the replay of sequences. On the other hand, spontaneous replay can be cued by a small increase in the firing rate of a particular assembly. Interestingly, once a replay is initiated, it does not change direction, in spite of the symmetric connectivity. An active assembly receives feedback inhibition from its inhibitory subpopulation, which prevents immediate further activations and hence a reversal of the direction of propagation.
We revived Hebb’s idea on assembly sequences (or “phase sequences”) where activity in a recurrent neural network propagates through assemblies [5], a dynamics that could underlie the recall and consolidation of memories. An important question in this context is how learning of a series of events can achieve a strong enough synaptic footprint to replay this sequence later. Using both numerical simulations of recurrent spiking neural networks and an analytical approach, we provided a biologically plausible model for understanding how minute synaptic changes can nevertheless be uncovered by small cues or even manifest themselves as activity patterns that emerge spontaneously. We showed how the impact of small changes in the connections between assemblies is boosted by recurrent connectivity within assemblies. This interaction between recurrent amplification within an assembly and the feedforward propagation of activity establishes a possible basis for the retrieval of memories. Our theory thus provides a unifying framework that combines the fields of Hebbian assemblies and assembly sequences [5], synfire chains [9, 10], and fast amplification in balanced recurrent networks that are in an asynchronous-irregular state [25, 28].
Main conclusions from our work are that the effective coupling between assemblies is a function of both feedforward and recurrent connectivities, and that the network can express three main types of behavior: 1. When the coupling is weak enough, assembly sequences are virtually indistinguishable from the background random connections, and no replays take place. 2. For sufficiently strong coupling, a transient input to some assembly propagates through the sequence, resulting in a replay. 3. For even stronger coupling, noise fluctuations get amplified by the underlying structure, resulting in spontaneous replays. Each of these three regimes has a certain advantage in performing a particular task. Weak coupling is appropriate for imprinting new sequences if the network dynamics is driven by external inputs rather than controlled by the intrinsically generated activity. Intermediate coupling is suitable for recollection of saved memories; sequences remain concealed and are replayed only by specific input cues; otherwise, the network is in the asynchronous-irregular, spontaneous state. For strong coupling, spontaneous replays might be useful for offline recollection of stored sequences when there are no external input cues. Importantly, the network behaviour and the rate of spontaneous events depends not only on the coupling but can be controlled by modulating the network excitability through external input. Neuromodulator systems, for example the cholinergic and the adrenergic systems [33, 34] might therefore mediate the retrieval process.
Assembly sequences are tightly related to synfire chains, which were proposed [9] as a model for the propagation of synchronous activity between groups of neurons. Diesmann et al. [10] showed for the first time that synfire chains in a noisy network of spiking neurons can indeed support a temporal code. It has been shown, however, that the embedding of synfire chains in recurrent networks is fragile [13, 30], because on the one hand, synfire chains require a minimal connectivity to allow propagation, while on the other hand, a dense connectivity between groups of neurons can generate unstable network dynamics. Therefore, Aviel et al. [31] introduced “shadow pools” of inhibitory neurons that stabilize the network dynamics for high connectivity. The network fragility can also be mitigated by reducing the required feedforward connectivity: inputs from the previous assembly are boosted by recurrent connections within the assembly. This approach was followed by Kumar et al. [14], who examined synfire chains embedded in random networks with local connectivity, thus, implicitly adopting some recurrent connectivity within assemblies as proposed by the assembly-sequence hypothesis; nevertheless, their assemblies were fully connected in a feedforward manner. Recently, it was shown that replay of synfire chains can be facilitated by adding feedback connections to preceding groups [45]. However, this Hebbian amplification significantly increased the duration of the spike volleys and thus decreased the speed of replay. Our model circumvents this slowing effect by combining the recurrent excitation with local feedback inhibition, effectively replacing Hebbian amplification by a transient “balanced amplification” [28].
Other analytical studies have used the Fokker-Planck approach to describe the propagation of pulse packets in synfire chains [46, 47]. In particular, Monasson and Rosay [48] have used diffusion analysis to explore the interplay between different environments encoded in the network and their effects on the activity propagation during replay. To store sequences, further classes of models were proposed, e.g., “winner-takes-all” [49–51] and “communication through resonance” [52]. However, the activity propagation in these models has an order of magnitude slower time scales than the synfire chain or the assembly sequence, and thus, are not suitable for rapid transient replays.
The spontaneous replay in our network bears some resemblance with the population bursts that occur in a model with supralinear amplification of precisely synchronised inputs [53, 54]. Adding such nonlinearities to the conductances in our model might decrease even further the connectivity required for the assembly-sequence replay. Another model class, which relies on lognormal conductance distributions, has been proposed as a burst generator for sharp-wave ripples (SWRs) [55]. The model accounts for spontaneously generated stereotypical activity that propagates through neurons that are connected with strong synapses.
Other computational models have focused more on different aspects of the SWR events. Taxidis et al. [56], for example, have proposed a hippocampal model where a CA3 network rhythmically generates bursts of activity, which propagate to a predominantly inhibitory CA1 network that generates fast ripple oscillations. The ripple generation by inhibitory networks is studied in a greater detail in Malerba et al. [57]. Azizi et al. [58] have explored the properties of a network that stores the topology of several environments and have shown that spike-frequency adaptation is an important mechanism for the movement of the activity bump within and between environments. In another modeling study, Romani and Tsodyks [43] proposed that short-term synaptic depression is a potential mechanism for explaining the hippocampal activity both during mobility (theta-driven activity) and during immobility (fast replays).
Another class of models that aims to explain the origin of SWR events relies on the electrical coupling between axons of pyramidal cells in the CA3/CA1 regions [59–61]. In a numerical model [62] it has been shown that the axonal plexus could explain the occurrence of SWs, the fast ripple oscillation, and moreover, account for the forward and reverse replay of sequences. Nevertheless, anatomical data to show the existence of such connections is still scarce [63].
What is the most efficient set of connectivities in terms of numbers of synapses used? To create an assembly of M neurons and to connect it to another assembly of the same size, we need M2(prc + pff) excitatory-to-excitatory synapses. The constraint κ = 1 (Eq 7) then leads to a minimum total number of synapses at prc = 0. This result is somewhat surprising because it suggests that our proposed recurrent amplification provides a disadvantage.
However, another constraint might be even more important: to imprint an association in one-shot learning, as for example required for episodic memories, it might be an advantage to change as few synapses as possible so that one can retrieve the memory later via a replay. Therefore, pff should be low, in particular lower than the recurrent connectivity that is bound by the morphological connectivity that includes also weak or silent synapses. Minimizing pff under the constraint κ = 1 implies, however, maximizing prc. Such large connectivities might require longer time to develop. A large prc is compatible with one-shot learning only if assemblies (that are defined by increased prc among a group of neurons) can be set up prior to the (feedforward) association of assemblies. Thus, episodic memories could benefit from strong preexisting assemblies. For setting up such assemblies, long time periods might be available to create new synapses and to morphologically grow synapses. Thus, we predict that for any episodic memory to be stored in one-shot learning in hippocampal networks such as CA3, a sufficiently strong representation of the events to be associated does exist prior to successful one-shot learning. In this case, pff (i.e., connectivity in addition to prand) can be almost arbitrarily low. A natural lower limit is that the number of synapses per neuron Mpff is much larger than 1, say 10 as a rough estimate (in Fig 3 we have Mpff ∼ 30 for a rather low value of prc = pff, and 10 for prc = 0.30; even 5 or more very strong synapses are sufficient in Fig 7). This can be interpreted in two ways: (1) Every neuron should activate several neurons in the subsequent assembly, and (2) every neuron in an assembly to be activated should receive several synapses from neurons in the previous assembly.
For example in the modeled network, for pff = 0.02 and Mpff > 10 we obtain M > 500, which is in agreement with an estimated optimal size of assemblies in the hippocampus [64]. The total number of feedforward synapses required for imprinting an association is then M2 pff > 5,000, which is a relatively small number compared to the total number of background synapses ( N E ) 2 p rand = 4 · 10 6 for NE = 20,000 and prand = 0.01. Scaling up the network accordingly (see Materials and Methods) to the size of a rodent CA3 network, i.e., NE = 240,000 (a typical number for the rat hippocampus, e.g., [65, 66]), the number of new associative synapses is M2 pff > 17,000, while the total connections are more than 0.5 ⋅ 109.
To conclude, abundant recurrent connections within assemblies can decrease the feedforward connectivity required for a replay to almost arbitrary low values. Moreover, the ratio of memory synapses to background synapses decreases as the network is scaled to bigger size.
For sequence replay, increasing the number of connections between groups has the same effect as scaling up the individual connection strengths. We conclude that structural and synaptic plasticity could play an equivalent role in the formation of assembly sequences. In the current study we have not considered plasticity mechanisms that could be mediating the formation of assembly sequences. Previous attempts of implementing a spike-timing-dependent plasticity (STDP) rule with an asymmetric temporal window [67–69] in recurrent networks led to structural instabilities [70–72]. However, it has been shown that under certain conditions the asymmetric STDP rule could encode sequences of connections [54], and moreover, maintain strong bidirectional synapses [73]. More sophisticated learning rules better matched the experimentally observed plasticity protocols [74–76], and these rules combined with various homeostatic mechanisms could form Hebbian assemblies that remained stable over long time periods [40, 41, 77]. Moreover, it has been shown that the triplet-based STDP rules [74, 75] lead to strong bidirectional connections [40, 41], a network motif that has been reported in multiple brain regions [24, 78–81]. Recent experimental work on the plasticity of the CA3-to-CA3 pyramidal cell synapses has revealed a symmetric STDP temporal curve [42]. Such a plasticity rule can be responsible for the encoding of stable assembly representations in the hippocampus.
Several plasticity rules have been successfully applied in learning sequences [7, 54, 73, 82–85]. However, these studies focused purely on sequence replay and did not take into account its interaction with a balanced, asynchronous irregular background state.
The present model may explain the fast replay of sequences associated with the sharp-wave ripple (SWR) events, which originate in the CA3 region of the hippocampus predominantly during rest and sleep [86]. SWRs are characterized by a massive neuronal depolarization reflected in the local field potential [87]. Moreover, during SWRs, pyramidal cells in the CA areas fire in sequences that reflect their firing during prior awake experience [19]. Cells can fire in the same or in the reverse sequential order, which we refer to as forward and reverse replay, respectively [22, 23]. Our model, however, can not account for the slower replays that occur at near behaviour time scales during REM sleep [88].
According to the two-stage model of memory trace formation [86], the hippocampus is encoding new episodic memories during active wakefulness (stage one). Later, these memories are gradually consolidated into neocortex through SWR-associated replays (stage two). It has been proposed that acetylcholine (ACh) modulates the flow of information between the hippocampus and the neocortex and thereby mediates switches between these memory-formation stages [89]. During active wakefulness, the concentration of ACh in hippocampus is high, leading to partial suppression of excitatory glutamatergic transmission [33] and promoting synaptic plasticity [90]. In this state, a single experience seems to be sufficient to encode representations of the immediate future in an environment [91]. On the other hand, the level of ACh decreases significantly during slow-wave sleep [92], releasing the synaptic suppression and resulting in strong excitatory feedback synapses, which suggests that this boost of recurrent and feedback connections leads to the occurrence of SWRs. In line with this hypothesis, the present model shows that increasing the synaptic strengths shifts the assembly-sequence dynamics from a no-replay regime to a spontaneous-replay regime. Also, we demonstrated that this regime supports both forward and reverse replay if assemblies are projecting symmetrically to each other and if recurrent connectivity exceeds severalfold the feedforward coupling.
Dragoi and Tonegawa [20, 93] showed that sequences can be replayed during SWRs also prior to the first exposure of the environment in which these sequences are represented. This finding challenges the standard framework according to which sequences are imprinted during exploration of the environment, i.e., the two-stage memory model [86]. An alternative model was presented by Sen Cheng [94] proposing that the recurrent CA3 synaptic weights are relatively constant during learning, and no plasticity in CA3 is required during the formation of new memories. According to the CRISP model [94], the storage of sequences is an intrinsic property of the CA3 network, and these sequences are formed offline prior to utilization due to the maturation of newly generated granule cells in the dentate gyrus. The model presented in this manuscript concerns the storage of sequences in a recurrent network and is not in contradiction with the idea of preexisting sequences.
Our model deploys a single uniform inhibitory population which is, likely, an oversimplification of cortical and subcortical networks that are rich in expressing various interneuron types [95, 96]. However, the roles of the different inhibitory neurons during various brain states, and in particular, during SWRs are not well known. Strong candidates for interneurons that might be balancing the run-away excitation during SWR replay are the basket cells due to their fast dynamics and strong synapses. Moreover, they are one of the most active inhibitory neurons during SWRs. OLM cells with their slower input on the distal dendrites are good candidates for priming which assemblies/sequence might be replayed prior to the event.
In summary, a prediction of our assembly-sequence model is that prior to being able to store and recall a memory trace that connects events, strong enough representations of events in recurrently connected assemblies are necessary because recalling a minute memory trace requires amplification within assemblies. Another prediction of this model is based on the fact that the network is in an asynchronous-irregular state during the time intervals between replays. Hence, by increasing the activity of the excitatory neurons or by disinhibiting the network, e.g., by decreasing the activity of the interneuron population specialized in keeping the balance, one could increase the rate of spontaneous replays. Such disinhibition might explain the counter-intuitive observation that SWRs can be evoked by the activation of interneurons [97, 98]. Our model thus links a diverse set of experimental results on the cellular, behavioral, and systems level of neuroscience on memory retrieval and consolidation [99].
The network simulations as well as the data analyses were performed in Python (www.python.org). The neural network was implemented in Brian [100]. For managing the simulation environment and data processing, we used standard Python libraries such as NumPy, SciPy, Matplotlib, and SymPy.
Neurons are described by a conductance-based leaky integrate-and-fire model, where the subthreshold membrane potential Vi(t) of cell i obeys
C d V i d t = G leak ( V rest - V i ) + G i E ( V E - V i ) + G i I ( V I - V i ) + I ext . (2)
The cells’ resting potential is Vrest = −60 mV, its capacitance is C = 200 pF, and the leak conductance is Gleak = 10 nS, resulting in a membrane time constant of 20 ms in the absence of synaptic stimulation. The variables G i E and G i I are the total synaptic conductances describing the time-dependent synaptic inputs to neuron i. The excitatory and inhibitory reversal potentials are VE = 0 mV and VI = −80 mV, respectively. Iext = Iconst + Ix is an externally applied current. To evoke activity in the network, a constant external current Iconst = 200 pA is injected into each neuron, which evokes a regular, intrinsically oscillating activity in the neuron, if considered in isolation. However, embedding such neurons in random recurrent networks can lead to irregular activity, as outlined below in the following two subsections. Only if explicitly stated (e.g., Figs 5 and 8), small additional current inputs Ix are applied to excitatory or inhibitory neurons, which we denote as Ie and Ii, respectively. As the membrane potential Vi reaches the threshold Vth = −50 mV, neuron i emits an action potential, and the membrane potential Vi is reset to the resting potential Vrest for a refractory period τrp = 2 ms.
The dynamics of the conductances G i E and G i I of a postsynaptic cell i are determined by the spiking of the excitatory and inhibitory presynaptic neurons. Each time a presynaptic cell j fires, the synaptic input conductance of the postsynaptic cell i is increased by g i j E for excitatory synapses and by g i j I for inhibitory synapses. The input conductances decay exponentially with time constants τE = 5 ms and τI = 10 ms. The dynamics of the total excitatory conductance is described by
d G i E ( t ) d t = - G i E ( t ) τ E + ∑ j , f g i j E δ ( t - t j ( f ) ) . (3)
Here the sum runs over the presynaptic projections j and over the sequence of spikes f from each projection. The time of the fth spike from neuron j is denoted by t j ( f ), and δ is the Dirac delta function. The inhibitory conductance G i I is described analogously.
Amplitudes of recurrent excitatory conductances and excitatory conductances on inhibitory neurons are denoted with g i j E and g i j I E, respectively. If not stated otherwise, all excitatory conductance amplitudes are fixed and equal (g i j E = g i j I E = g E = 0 . 1 nS), which results in EPSPs with an amplitude of ≈ 0.1 mV at resting potential. The recurrent inhibitory synapses are also constant (g i j I = 0 . 4 nS) while the inhibitory-to-excitatory conductances g i j E I are variable (see below). Irrespectively of the synaptic type, the delay between a presynaptic spike and a postsynaptic response onset is always 2 ms.
The modeled network consists of NE = 20,000 excitatory and NI = 5,000 inhibitory neurons. Our results do not critically depend on the network size (see Section ‘Scaling the network size’ below). Initially, all neurons are randomly connected with a sparse probability prand = 0.01.
A cell assembly is defined as a group of recurrently connected excitatory and inhibitory neurons (Fig 1A). The assembly is formed by picking M excitatory and M/4 inhibitory neurons from the network; every pair of pre- and post-synaptic neurons within the assembly is randomly connected with probability prc. The new connections are created independently and in addition to the already existing ones. Thus, if by chance two neurons have a connection due to the background connectivity and are connected due to the participation in an assembly, then the synaptic weight between them is simply doubled. Unless stated otherwise, assemblies are hence formed by additional connections rather than stronger synapses.
In the random network, we embed 10 non-overlapping assemblies with size M = 500 if not stated otherwise. The groups of excitatory neurons are connected in a feedforward fashion, and a neuron from one group projects to a neuron of the subsequent group with probability pff (Fig 1B). Such a feedforward connectivity is reminiscent of a synfire chain. However, classical synfire chains do not have recurrent connections (prc = 0, pff > 0), while here, neurons within a group are recurrently connected even beyond the random background connectivity (prc > 0, pff > 0). We will refer to such a sequence as an “assembly sequence”. By varying the connectivity parameters prc and pff, the network structure can be manipulated to obtain different network types (Fig 1C). In the limiting case where feedforward connections are absent (prc > 0, pff = 0) the network contains only largely disconnected Hebbian assemblies. In contrast, in the absence of recurrent connections (prc = 0, pff > 0), the model is reduced to a synfire chain embedded in a recurrent network. Structures with both recurrent and feedforward connections correspond to Hebbian assembly sequences.
To keep the network structure as simple as possible and to be able to focus on mechanisms underlying replay, we use non-overlapping assemblies and we do not embed more than 10 groups. Nevertheless, additional simulations with overlapping assemblies and longer sequences indicate that our approach is in line with previous results on memory capacity [15, 64, 101]. Advancing the theory of memory capacity is, however, beyond the scope of this manuscript.
A naive implementation of the heterogeneous network as described above leads, in general, to dynamics characterized by large population bursts of activity. To overcome this epileptiform activity and ensure that neurons fire asynchronously and irregularly (AI network state), the network should operate in a balanced regime. In the balanced state, large excitatory currents are compensated by large inhibitory currents, as shown in vivo [102, 103] and in vitro [104]. In this regime, fluctuations of the input lead to highly irregular firing [105, 106], a pattern observed in the cortex [9, 107] as well as in the hippocampus during non-REM sleep [108, 109].
Several mechanisms were proposed to balance numerically simulated neural networks. One method involves structurally modifying the network connectivity to ensure that neurons receive balanced excitatory and inhibitory inputs [110, 111]. It was shown that a short-term plasticity rule [112] in a fully connected network can also adjust the irregularity of neuronal firing [113].
Here, we balance the network using the inhibitory-plasticity rule [25]. All inhibitory-to-excitatory synapses are subject to a spike-timing-dependent plasticity (STDP) rule where near-coincident pre- and postsynaptic firing potentates the inhibitory synapse while presynaptic spikes alone cause depression. A similar STDP rule with a symmetric temporal window was recently reported in the layer 5 of the auditory cortex [114].
To implement the plasticity rule in a synapse, we first assign a synaptic trace variable xi to every neuron i such that xi is incremented with each spike of the neuron and decays with a time constant τSTDP = 20 ms:
x i → x i + 1 , if neuron i fires, τ STDP d x i d t = − x i , otherwise .
The synaptic conductance g i j E I ( t ) from inhibitory neuron j to excitatory neuron i is initialized with value g 0 I = 0 . 4 nS and is updated at the times of pre/post-synaptic events:
g i j E I = g i j E I + η ( x i − α ) , for a presynaptic spike in neuron j , g i j E I = g i j E I + η x j , for a postsynaptic spike in neuron i
where 0 < η ≪ 1 is the learning-rate parameter, and the bias α = 2ρ0τSTDP is determined by the desired firing rate ρ0 of the excitatory postsynaptic neurons. In all simulations, ρ0 has been set to 5 spikes/sec, which is at the upper bound of the wide range of rates that were reported in the literature: e.g., 1–3 spikes/sec in [87]; 3–6 spikes/sec in [115]; 1–76 spikes/sec in [116]; 0.43–3.60 spikes/sec in [117]; 1–11 spikes/sec in [118].
Existence of background connections and an implementation of the described inhibitory STDP rule drives typically the network into a balanced AI state. The excitatory and the inhibitory input currents balance each other and keep the membrane potential just below threshold while random fluctuations drive the firing (Fig 2A and 2B). The specific conditions to be met for a successful balance are discussed in the Results section. Similar effects could be achieved also in the absence of random background connections when input with appropriate noise fluctuations is applied to the neurons. We find this scenario, however, less realistic as neurons would be largely disconnected.
In the AI network regime, any perturbation to the input of an assembly will lead to a transient perturbation in the firing rate of the neurons within it. In the case of strong recurrent connections within the assembly, a small excitatory perturbation will lead to a stronger firing of both the excitatory as well as the inhibitory neurons. This amplification of input fluctuations into larger activity fluctuations is, unlike the Hebbian amplification, fast and does not show slowing of the activation dynamics for large connectivities. This phenomenon of transient pattern completion is known as balanced amplification [28], where it is essential that each assembly has excitatory and inhibitory neurons and strong recurrent connectivity. Another advantage of the inhibitory subpopulations is the rapid negative feedback that can lead to enhanced memory capacity of the network [119].
Each network simulation consists of 3 main phases:
1. Balancing the network. Initially, the population activity is characterized by massive population bursts with varying sizes (avalanches). During a first phase, the network (random network with embedded phase sequence) is balanced for 50 seconds with decreasing learning rate (0.005 ≥ η ≥ 0.00001) for the plasticity on the inhibitory-to-excitatory synapses. During this learning, the inhibitory plasticity shapes the activity, finally leading to AI firing of the excitatory population. Individual excitatory neurons then fire roughly with the target firing rate of 5 spikes/sec, while inhibitory neurons have higher firing rates of around 20 spikes/sec, which is close to rates reported in the hippocampus [87, 117]. After 50 seconds simulation time, the network is typically balanced.
2. Reliability and quality of replay. In a second phase, the plasticity is switched off to be able to probe an unchanging network with external cue stimulations. All neurons from the first group/assembly are simultaneously stimulated by an external input so that all neurons fire once. The stimulation is mimicked by adding an excitatory conductance in Eq 3 (gmax = 3 nS) that is sufficient to evoke a spike in each neuron. For large enough connectivities (prc and pff), the generated pulse packet of activity propagates through the sequence of assemblies, resulting in a replay. For too small connectivities, the activity does not propagate. For excessively high connectivities, the transient response of one group results in a burst in the next group and even larger responses in the subsequent groups, finally leading to epileptiform population bursts of activity (Fig 3).
To quantify the propagation from group to group and to account for abnormal activity, we introduce a quality measure of replay. The activity of a group is measured by calculating the population firing rate of the underlying neurons smoothed with a Gaussian window of 2 ms width. We extract peaks of the smoothed firing rate that exceed a threshold of 30 spikes/sec. A group is considered to be activated at the time at which its population firing rate hits its maximum and is above the threshold rate. Activity propagation from one group to the next is considered to be successful if one group activates the next one within a delay between 2 and 20 ms. A typical delay is about 5 ms, but in the case of extremely small pff and large prc the time of propagation can take ∼ 15 ms. Additional rules are imposed to account for exceeding activity and punish replays that lead to run-away firing. First, if the activity of an assembly exceeds a threshold of 180 spikes/sec (value is chosen manually for best discrimination), the group is considered as bursting, and thus, the replay is considered as failed. Second, if the assembly activity displays 2 super-threshold peaks that succeed each other within 30 ms, the replay is unsuccessful. Third, a “dummy group” (of size M) from the background neurons is used as a proxy for detecting activations of the whole network. In case that the dummy group is activated during an otherwise successful replay, the replay is failed. Thus, for each stimulation the “quality of replay” has a value of 1 for successful and a value of 0 for unsuccessful replays. The quality of replay for each set of parameters (Fig 3) is an average from multiple (≳ 5) stimulations of 5 different realizations of each network.
Additionally, we test the ability of the assembly sequence to complete a pattern by stimulating only a fraction of the neurons in the first group (Fig 4). Analogously to the full stimulation, the quality of replay is measured.
3. Spontaneous activity. In the last phase of the simulations, no specific input is applied to the assemblies. As during the first phase of the simulation, the network is driven solely by the constant-current input Iconst = 200 pA applied to each neuron, and plasticity is switched off.
During this state, we quantify spontaneous replay (Fig 5). Whenever the last assembly is activated and if this activation has propagated through at least three previous assemblies, we consider this event as a spontaneous replay. Here, we apply the quality measure of replay, where bursty replays are disregarded. Additionally, we quantify the dynamic state of the network by the firing rate, the irregularity of firing, and the synchrony of a few selected groups from the sequence. The irregularity is measured as the average coefficient of variation of inter-spike intervals of the neurons within a group. As a measure of synchrony between 2 neurons, we use the cross-correlation coefficient of their spike trains binned in 5-ms windows. The group synchrony is the average synchrony between all pairs of neurons in a group.
How quickly do the neurons that receive a synchronous pulse packet react during a replay? Following the arguments of Diesmann et al. [10], the response time is not determined by the membrane time constant of the neuron, but rather by the time it takes the neurons to reach threshold in response to the pulse packet. An analytical calculation can hence be obtained by considering the membrane potential dynamics in Eq 2. Let us assume that a neuron is at some initial voltage V0. How fast does the neuron reach the threshold voltage when an external excitatory conductance Ginj is applied to the membrane? We can express the membrane potential V(t) explicitly:
V = ( V 0 - V * ) exp - t τ * + V *
where the “driving” voltage is
V * = G leak V rest + G E V E + G I V I + I ext + G inj V E G leak + G E + G I + G inj
and the time constant is
τ * = G leak G leak + G E + G I + G inj τ m .
Here, τm = C/Gleak = 20 ms is the leak time constant from Eq 2. The time that is needed for a neuron with initial membrane potential V0 to reach the voltage threshold Vth is:
t AP = τ * log V 0 - V * V th - V * .
Substituting with parameter values corresponding to the simulations (GE = 0.6 nS, GI = 5 nS, Gleak = 10 nS, Ginj = 3 nS, V0 = −51 mV), we obtain tAP = 1.4 ms. Here, for Ginj we use a typical value of the peak excitatory conductance during a replay.
We also measured the activation time of the assemblies during pulse propagation in the simulated balanced network. A stimulation with step conductance Ginj applied to a group of random neurons leads to a fast increase in firing rates (20%-to-80% rise time is 1 ms).
In summary, in agreement with the literature [105, 106, 120], the response time of the modeled network is indeed fast, i.e., faster than the membrane time constant τm = 20 ms and the inter-spike interval (ISI ∼ 12 ms when Ginj is injected).
An analytical description of conditions for successful replay is not easy to obtain. The most appropriate ansatz would be a generalization of the pulse-packet description of Goedeke & Diesmann [121], which is unfortunately not trivial and beyond the scope of this paper. Instead, we choose a phenomenological approach and portray the network dynamics during replay by a linear dynamical system, which could be thought of as a linearization of a more accurate model. This ansatz allows to estimate a lower bound for the connectivities required for a successful replay.
The dynamics of an assembly i (Fig 1A and 1B) in the AI state is approximated by two differential equations:
τ d r i E d t = - r i E + w rc r i E - k w rc r i I + ξ i E ( t ) τ d r i I d t = - r i I + w rc r i E - k w rc r i I (4)
where r i E and r i I are the deviations of the population firing rates of the excitatory (E) and inhibitory (I) populations from the spontaneous firing rates r 0 E and r 0 I, respectively. The parameter wrc and the term −kwrc represent the respective strengths of the excitatory and the inhibitory recurrent projections. The constant k describes the relative strength of the recurrent inhibition vs. excitation; for a balanced network, we assume that inhibition balances or dominates excitation, e.g., k ≥ 1. The weight wrc is proportional to the average number Mprc of recurrent synapses a neuron receives, and proportional to the synaptic strength gE. The function ξ i E describes the external input to the assembly from the rest of the network. In this mean-field analysis, we neglect the influence of the noise on the network dynamics. Activities r i E and r i I are assumed to approach the steady state 0 with a time constant τ. Based on the discussion in the previous subsection, we assume this time constant to be much faster than the membrane time constant.
The excitatory assemblies are sequentially connected, and we denote the strength of the feedforward projections as wff. The feedforward drive can be represented as an external input to an assembly:
ξ i E = w ff r i - 1 E , for i > 1 .
Taking into account the feedforward input to population i from the preceding excitatory population i − 1, Eq 4 can be rewritten as
τ d r i d t = - 1 + w rc - k w rc w rc - 1 - k w rc r i + w ff r i - 1 E 0 , for i > 1 (5)
where r i = ( r i E r i I ) is the 2-dimensional vector of firing rates in group i.
Assuming that the time duration of a pulse packet in group i − 1 is much longer than the population time constant τ in group i, we consider the solution of the stationary state (τ d r i d t = 0) as an adequate approximation. By setting the left-hand side of Eq 5 to zero, we can express the firing rate r i E as a function of r i - 1 E:
r i E = 1 + k w rc 1 + ( k - 1 ) w rc w ff r i - 1 E = κ r i - 1 E , (6)
where κ is the “effective feedforward connectivity”.
Interestingly, the recurrent connections effectively scale up the efficiency of the feedforward connections and facilitate the propagation of activity. Assuming that (k − 1)wrc ≪ 1, that is, either small recurrent connectivity wrc or an approximately balanced state k ≈ 1, we can linearize in wrc:
κ ≈ w ff ( 1 + w rc ) . (7)
For small κ, i.e. κ ≪ 1, even large changes of the firing rate in group i − 1 do not alter the rate in group i. For κ < 1, the pulse packet will steadily decrease while propagating from one group to another as r i E < r i - 1 E. On the other hand, if κ = 1, the propagation of a pulse packet is expected to be marginally stable. In the case of κ > 1, any fluctuation of firing rate in one assembly will lead to a larger fluctuation in the following assembly.
To connect the analytical calculations to the numerical simulations, we again note that a total connection strength is proportional to the number of inputs a neuron is receiving (e.g., the product of group size M and connection probability) and proportional to the synaptic strength:
w rc = c M p rc g E and w ff = c M p ff g ff E , (8)
where M is the group size, and prc and pff are the recurrent and feedforward connectivities, respectively. gE is the conductance of an excitatory recurrent synapse within a group, and g ff E is the conductance of feedforward synapses between groups. Unless stated otherwise, we assume g ff E = g E. The parameter c is related to the slope of the neurons’ input-output transfer function, but given the phenomenological nature of the theoretical treatment, an accurate ab initio calculation of c is non-trivial. Instead, we use it as a fitting parameter. Using the critical value κ(prc = 0.08, pff = 0.04) = 1 extracted from the simulation results (Fig 3), we find c = 0.25 nS−1. This value of c is used in all further analytical estimations for the effective connectivity κ.
In summary, the lower bound for the connectivities for a successful replay can be described as
p rc = 1 c M g E 1 c M p ff g ff E - 1 ,
which is represented as a black line in Figs 3 and 5. For Figs 6 and 7, the black line is calculated analogously using the same constant c = 0.25 nS−1.
In the previous section, the constant c was manually fitted to a value of 0.25 nS−1 to match analytical and numerical results. Here we express c analytically by utilizing a non-linear neuronal model and by using the parameter values from the simulations.
The resting firing rate ρ of a neuronal population that is in an asynchronous irregular (AI) regime can be expressed as a function of the mean μ and the standard deviation σ of the membrane potential distribution [47, 122–124]:
μ = ∑ k J k ρ k σ = ∑ k J k 2 ρ k , (9)
where the sums over k run over the different synaptic contributions, ρk is the corresponding presynaptic firing rate, and Jk and J k 2 are the integrals over time of the PSP and the square of the PSP from input k, respectively. Here PSPs are estimated for the conductance-based integrate-and-fire neuron from Eq 2 for voltage values near the firing threshold Vth,
J k = ∫ t PSP ( t ) d t = τ syn ( V syn - V th ) g k syn G leak J k 2 = ∫ t PSP 2 ( t ) d t = τ syn g k syn ( V syn - V th ) 2 2 ( τ + τ syn ) G leak 2 ,
where τ is the membrane time constant, τsyn is the synaptic time constant, Vsyn is the synaptic reversal potential, and g k syn is the synaptic conductance of connection k. Connections can be either excitatory or inhibitory.
Here we consider a network with random connections only, and look at a subpopulation of size M, where M ≪ NE. For a more convenient analytical treatment, the recurrent connections within the group are neglected. This assumption does not affect the estimation of the transfer function slope, as c is independent on the type of inputs. The firing rate-fluctuations of the neuronal group are calculated as in Eq 6:
r = c M g E r ext . (10)
The membrane potential of an excitatory neuron from this subpopulation has several contributions: NE prand excitatory inputs with firing rate ρ0 and efficacy JE; inhibitory inputs due to the background connectivity: N I p rand J E I ρ 0 I; injected constant current: Iext/Gleak; and input from an external group: M ext J ext E ρ ext. In summary, we find:
μ = N E p rand J E ρ 0 + N I p rand J E I ρ 0 I + M ext J ext E ρ ext + I ext G leak .
The standard deviation of the membrane potential is then, accordingly:
σ 2 = N E p rand J E 2 ρ 0 + N I p rand J E I 2 ρ 0 I + M ext J ext E 2 ρ ext .
In the case of uncorrelated inputs, the following approximation can be used for the firing rate estimation [47, 122–124]:
ρ = τ rp + τ π ∫ V rest - μ σ V th - μ σ e u 2 1 + erf ( u ) d u - 1 , (11)
where τrp is the refractory period, and Vth and Vrest are membrane threshold and reset potential, respectively (see also section “Neural Model”).
To find the constant c used in the linear model, we estimate the firing rate ρ from Eq 11 and substitute in Eq 10, assuming a linear relation between firing-rate fluctuations:
ρ ( ρ ext ) - ρ 0 = c M ext g E ( ρ ext - 0 ) , (12)
and find:
c = ρ ( ρ ext ) - ρ 0 c M ext g E ρ ext . (13)
Before calculating the constant c according to the method presented above, a preliminary step needs to be taken. As we set the firing rate of the excitatory population in the network to a fixed value ρ0 = 5 spikes/sec, there are two variables remaining unknown: the firing rate of the inhibitory population ρ 0 I and the inhibitory-to-excitatory synaptic conductance g rand E I that changes due to synaptic plasticity. Therefore, we first solve a system of 2 equations for the firing rates of the excitatory and the inhibitory populations expressed as in Eq 11. Once the unknowns ρ 0 I and g rand E I are calculated, we can estimate ρ(ρext) and c according to the method presented above. We note that the analytically calculated values of g rand E I and ρ 0 I match the measured values in the simulations.
The value we get after applying the above mentioned method for estimation of c is 0.13 nS−1. The fit corresponding to the estimate of c is shown in Fig 3 with a white dashed line. It is worth noting that a slightly more involved calculation relying on the estimate c = 1 M g ∂ ρ ∂ ρ e x t gives a similar result, concretely c = 0.11 nS−1.
Although the analytically calculated value c is a factor of 2 smaller than the manual fit c = 0.25 nS−1, it is in the same order of magnitude and not too far from describing the results for critical connectivity from the simulations.
The method applied above finds the slope of the transfer function for stationary firing rates. However, the spiking network replay is a fast and brief event, where a transient input in one assembly evokes a transient change in the output firing rate. The value discrepancy suggests that the transfer function of transients is even steeper than at the resting AI state.
So far we have been dealing with networks of fixed size NE = 20,000 neurons. How does the network size affect the embedding of assembly sequences? Is it possible to change the network size but keep the assembly size fixed?
Scaling the network size while keeping the connectivity prand constant leads to a change in the number of inputs that a neuron receives, and therefore, affects the membrane potential distributions. To compare replays in networks with different sizes NE but identical M, we need to assure that the signal-to-noise ratio is kept constant, and the easiest way is to keep both the signal and the noise constant, which requires to change connectivities prc and pff and conductances.
While scaling the network from the default network size NE = 20,000 to a size N ˜ E = γ N E, we see that the noise σ scales as ∼ g γ N E (Eq 9). To keep the input current fluctuations constant as we change N ˜ E, all synaptic conductances are rescaled with a factor of 1 / γ: g ˜ = g / γ [105]. However, such a synaptic scaling leads to a change in the coupling between assemblies of fixed size M, which is proportional to the conductance. Therefore, the connectivities prc and pff are scaled with γ to compensate the conductance decrease, leading to a constant coupling (c M p ˜ rc g ˜ E = c M p rc g E and c M p ˜ ff g ˜ E = c M p ff g E), and hence, a constant signal-to-noise ratio.
What is the impact of such a scaling on the network capacity to store sequences? The number of connections needed to store a sequence is changed by a factor γ as we change prc and pff. However, the number of background connections to each neuron is scaled with γ, resulting in sparser memory representations in larger networks. More precisely, for a neuron participating in the sequence, the ratio of excitatory memory connections to the total number of excitatory connections is
u = ( p rc + p ff ) γ M ( p rc + p ff ) γ M + p rand γ N E .
Therefore, the proportion of connections needed for an association is scaled as 1 / γ for N ˜ E ≫ M. To give a few numbers, u is equal to 0.23 for N ˜ E = 20 , 000, and u = 0.09 for N ˜ E = 180 , 000. Other parameter values are: M = 500, prc = pff = 0.06, prand = 0.01.
The chosen scaling rule is applicable for networks of simpler units such as binary neurons or current-based integrate-and-fire neurons [106, 123]. This scaling is not valid in a strict mathematical framework for very large networks (N ˜ E → ∞) consisting of conductance-based integrate-and-fire neurons (see [110] for a detailed discussion). Simulations results, however, reveal that replays are possible in network sizes up to 2 ⋅ 105 neurons.
|
10.1371/journal.pntd.0002332 | Polyethyleneimine Mediated DNA Transfection in Schistosome Parasites and Regulation of the WNT Signaling Pathway by a Dominant-Negative SmMef2 | Schistosomiasis is a serious global problem and the second most devastating parasitic disease following malaria. Parasitic worms of the genus Schistosoma are the causative agents of schistosomiasis and infect more than 240 million people worldwide. The paucity of molecular tools to manipulate schistosome gene expression has made an understanding of genetic pathways in these parasites difficult, increasing the challenge of identifying new potential drugs for treatment. Here, we describe the use of a formulation of polyethyleneimine (PEI) as an alternative to electroporation for the efficacious transfection of genetic material into schistosome parasites. We show efficient expression of genes from a heterologous CMV promoter and from the schistosome Sm23 promoter. Using the schistosome myocyte enhancer factor 2 (SmMef2), a transcriptional activator critical for myogenesis and other developmental pathways, we describe the development of a dominant-negative form of the schistosome Mef2. Using this mutant, we provide evidence that SmMef2 may regulate genes in the WNT pathway. We also show that SmMef2 regulates its own expression levels. These data demonstrate the use of PEI to facilitate effective transfection of nucleic acids into schistosomes, aiding in the study of schistosome gene expression and regulation, and development of genetic tools for the characterization of molecular pathways in these parasites.
| Schistosomiasis is a global disease infecting more than 240 million people worldwide and is ranked second only to malaria in global health importance. The causative agents of human schistosomiasis are parasitic worms that ingest red blood cells and can live for decades producing hundreds of eggs daily. There is one primary drug for treatment of schistosomiasis, but its use for over 30 years has raised concern over the development of drug resistance and thus created a need for new drugs. A challenge to the rational development of effective antischistosomals has been the difficulty in manipulating schistosome gene expression, and thus a limitation in our understanding of schistosome gene function. Here, we present a new and straightforward method for inserting genes into schistosomes and expressing them. In addition, to our knowledge we provide the first example of dominant negative gene expression to modify transcriptional regulation using a molecular genetics approach to study this globally important parasite.
| The use of transgenesis and other technological advances has had a powerful impact in the molecular characterization and functional analysis of gene function in model organisms [1], [2]. However, like many parasitic worms, the natural characteristics of the schistosome (its complex life cycle involving multiple hosts, the absence of an immortalized cell line, and the inability to maintain the entire life cycle in vitro) have made in-depth genetic modifications challenging [3]. Schistosomes are the causative agents of human schistosomiasis, a parasitic disease that is endemic in 78 countries worldwide and that infects almost 240 million people [4]. In terms of morbidity and mortality, schistosomiasis is considered to be the most important helminth infection [5]. Although our knowledge of schistosome biology has increased over the last few years, the lack of simple and effective methodologies to manipulate schistosomes has slowed our understanding of schistosome molecular biology significantly behind other systems.
With the sequencing of the schistosome genome and recent updates to schistosome annotation [6], [7], research has focused on the functional analysis of schistosome genes. This includes approaches to insert DNA/RNA into schistosomes and to induce gene expression. Strategies used thus far for transfection of DNA/RNA molecules include the use of particle bombardment [8]–[10], soaking [11]–13, electroporation [14]–[18], chemical or lipofectamine based approaches [16], [19], and viruses [20], [21] (for review see [22]).
The insertion of genetic material into schistosomes by soaking in high concentrations of DNA/RNA has been successful for delivering siRNA and dsRNA. [11], [12], [19], [22], [23]. This approach is straightforward; however, the transfection efficiency is highly restricted by the size of nucleic acid fragment delivered, and worm death resulting from the use of highly concentrated nucleic acids [16]. In addition, this approach is not suitable for long-term modification of the parasite genome, which requires the use of compatible sized vectors that carry information for transcription, self-amplification, and the insertion of transposable elements.
The use of biolistic particle delivery for Schistosoma mansoni (S. mansoni) transfection has been successful for several developmental stages of the parasite. These include the adult, sporocyst, and miracidia [10], [24]–[26]. However, the square wave electroporation approach to introduce naked plasmid-based and non-plasmid-based exogenous genes into schistosomes has been more successful [8], [15]–[17], [27]. Square wave electroporation is more effective for transfection of schistosome eggs than the use of pseudotyped murine leukemia virus [21]. Consequently, electroporation has become the method of choice for schistosome transgenesis, specifically for the delivery of siRNA, dsRNA and vector based shRNA for gene silencing studies by RNAi [14]–[19], [23], [27]–[30]. One report, however, has suggested that the biolistic particle delivery method is more effective than electroporation for the delivery of RNA into adult worms and miracidia [8]. Nonetheless, both particle bombardment and electroporation can be damaging or even lethal to cells and parasites due to the physical damage or intense electrical charges, respectively [31], [32].
Polyethyleneimine (PEI) is a synthetic polymer with a highly cationic charge that can facilitate gene transfection in cells, and was identified as an oligonucleotide transfection reagent in 1995 [33]. PEI tightly binds to DNA by electrostatic interaction, induces DNA condensation and packaging into nanosized particles, and protects DNA from degradation, increasing efficiency for entrance into cell nuclei [34]–[37]. PEI as a transfection reagent is available in either linearized or branched structures and in a range of molecular weights. Several reports suggest that linearized PEI is the most efficient and optimized reagent for transfection, compared to the branched, higher molecular weight form [38], [39]. Here we report the use of PEI for the transfection of schistosomes.
To our knowledge, this is the first report of the use of PEI for schistosome transfection. The idea was inspired by the use of PEI to transfect the S. mansoni intermediate host, Biomphalaria glabrata (B. glabrata) [38]. Here, we evaluate the use of PEI as a transfection reagent for schistosomes using a plasmid encoding the mCherry fluorescent protein, and a Neomycin selectable marker. We then assess two RNA polymerase II (pol II) promoters for their ability to drive transcription of the reporter gene.
Previously, we characterized the schistosome myocyte enhancer factor (SmMef2) [40], a DNA-binding transcriptional activator that is important for cellular development, morphogenesis and survival in mammals and Drosophila (for review on Mef2, see [41]). We identified potential SmMef2 DNA binding elements in the promoters of wingless-type MMTV integration site family members 1 and 2 (Wnt1 and Wnt2) homologs. WNT genes are conserved oncogenes that play a significant role in cell development, cell signaling and cell fate during early development [42]–[44]. Although the WNT pathway in schistosomes has not been extensively characterized, some WNT genes have been described in these worms [45]–[47]. We proposed that these two schistosome WNT genes homologs Smp_152900 (SmWnt1) and Smp_167140 (SmWnt2) could be potential targets of SmMef2.
Here, we developed a dominant negative form of SmMef2 (SmMef2,133) that lacks a transactivation domain and using this genetic mutant provide evidence that SmMef2 can regulate SmWnt1 and SmWnt2 gene transcription levels. Finally, we provide data supporting a role for SmMef2 in regulating its own transcription.
Cercariae of S. mansoni NMRI strain (NR-21962) or strain PR-1 (NR-21961) were shed from the infected B. glabrata snails obtained from the Biomedical Research Institute (Rockville, MD) and transformed into schistosomula as previously described [48], [49] Seven to ten thousand schistosomula were cultured in complete RPMI medium (RPMI, 5% Fetal Bovine Serum, 1× Pen/Strep) per well in 12-well cell culture plates (Greiner Bio-One, Orlando, FL) at 37°C and 5% CO2 for 4 hours before being utilized for transfection. For the longevity experiment, modified Basch Medium 169 (Basch Medium 1695, Fetal Bovine Serum, 1× Pen/Strep) was used for the first three days of culture. After three days, the media was changed and replaced with complete Basch Medium [48].
DNA primers were designed and ordered from Integrated DNA Technologies (IDT, Coralville, IA). Subcloning was performed using the In-Fusion HD Cloning kit (Clontech, Mountainview, CA). The full transcript of the mCherry gene from the transposon vector pKM225 (GenBank: HQ386859.1), the first 399 bp (133 amino acids) of SmMef2, and the wild-type SmMef2 [40] (NCBI accession number: JN900476) were amplified by PCR using Phusion High-Fidelity DNA Polymerase (NEB, Ipswich, MA) with three sets of primers: oEJ1020 forward (5′-TCA CGC GTG GTA CCT CTA GAA TGG TGA GCA AGG GCG AGG AG) and oEJ1021 reverse (5′-GCC CGG GTC GAC TCT AGA TTA CTT GTA CAG CTC GTC CAT GCC), oEJ1026 forward (5′-CAC TAT AGG CTA GCC TCG AGA TGG GTC GCA AAA AAA TAC TCA TC) and oEJ1027 reverse (5′-GCG TGA ATT CTC GAG CTA CGG TGT TTT AGT TCC TGT TCG TAT), and oLS197b forward (5′-ATA GGC TAG CCT CGA GAT GGG TCG CAA AAA AAT ACT CAT CA-3′) and oLS198 reverse (5′-TAA AGG GAA GCG GCC GCT CAA AGG TGG CGC ACA CGT TTA AGA-3′), respectively. mCherry and truncated SmMef2 (SmMef2,133) amplicons were subcloned into the pCI-neo plasmid (Promega, Madison, WI) at the XbaI and XhoI sites, respectively. The wild-type SmMef2 (SmMef2) amplicon was subcloned into the pCI-neo plasmid at the XhoI and NotI sites. Constructs were transformed into chemically competent One Shot TOP10 cells (Invitrogen, Carlsbad, CA). The mCherry reporter plasmid (pEJ1175), SmMef2,133 expression plasmid (pEJ1181) and SmMef2 expression plasmid (pLS068) (Figure 1) were purified using the Nucleospin Plasmid miniprep kit (Clontech, Mountainview, CA) and verified by restriction digestion analysis.
Plasmid pEJ1116 contains 2000 base pairs of the Sm23 upstream activation sequence (UAS) regulating the expression of mCherry. To make this construct, the mCherry transcript was amplified from plasmid pEJ604 using primers oJM16 forward (5′-CGT TTG AAA GTA TGG GAT CCA TGG TGA GCA AGG GCG AGG AG) and oJM17 reverse (5′-CTG TTT TCT TTG CAG TGT CTG CAG TTA CTT GTA CAG CTC GTC CAT GCC), then subcloned between the BamHI and PstI sites in the pGBKT7 vector (Clontech, Mountainview, CA). The 2000 base pair region containing the upstream activation sequence of Sm23 was amplified from schistosome genomic DNA using oligos oJM12 forward (5′-ATG GAG GCC GAA TTC CCG GGA CCC GAA CAC TAT AGT GTG ATG CAG) and oJM13 reverse (5′-CCG CTG CAG GTC GAG GAT CCC ATA CTT TCA AAC GGG ACA CAA TGC), then subcloned into the XmaI and BamHI sites of the same vector to make plasmid pEJ1116. To review, plasmid pEJ1116 contains the 2000 base pair UAS of the Sm23 promoter, followed by the mCherry reporter gene (Figure 1B).
InVitroPlex-Express-Parasite (Cat # IVTP-ExPA-002), a formulation of PEI optimized for nucleic acid delivery into parasites, was received as a gift from Dr. Puthupparampil Scaria (AparnaBio, Rockville, MD), and used for the transfection of schistosomes. PEI (7.2 µg) and DNA plasmid (4.8 µg), either pEJ1175, pEJ1181, or pEJ1116, were diluted in 1 mL of complete RPMI [49], separately. Then, the 1 mL PEI solution was added to the 1 mL DNA solution drop by drop to make a 2 mL PEI/DNA mixture with a PEI nitrogen and DNA phosphate (N/P) ratio of either 6∶1 or 11∶1, followed by 10–15 sec vigorous vortexing. The PEI/DNA RPMI solution was incubated at 37°C for 30 min to allow the PEI and DNA to form a nanoparticle complex. The complete RPMI from the 4 h schistosomula culture was carefully removed, leaving the schistosomula at the bottom of the culture well. Two mL of pre-warmed PEI/DNA solution was then added to the plate well and schistosomula were grown in the transfection mixture for another 40 h at 37°C and in 5% CO2. All above procedures were performed under sterile conditions. For each DNA transfection experiment, schistosomula were cultured in complete RPMI medium lacking PEI or DNA, or without both as negative controls.
At 40 hours post-transfection, the supernatant was removed from 44 h schistosomula by centrifuging the parasites at 1,500× g for 2 min. Recovered parasites were washed with 1.5 mL of 1× phosphate buffered saline (137 mM NaCl, 2.7 mM KCl, 4.3 mM Na2HPO4 and 1.47 mM KH2PO4 at a final pH of 7.4) twice to remove potentially contaminating residual DNA and PEI remaining in the tube, incubated for 15 minutes in 50 units of DNAse I to remove any remaining external DNA, and resuspended in 1× phosphate buffered saline.
Total DNA was purified using phenol-chloroform (Thermo Scientific, Waltham, MA). Five mg/mL glycogen (Invitrogen, Carlsbad, CA) and 3 M sodium acetate were added during the purification to increase the yields of DNA.
Total RNA was isolated following the standard manufacturer's protocol for the PureLink RNA Mini Kit using TriZol reagent (Invitrogen, Carlsbad, CA). DNase I digestion was performed to eliminate DNA contamination.
RNA and DNA were quantified on a Nanodrop 8000 spectrophotometer (Thermo Scientific, Waltham, MA) and the quality was verified by visualization on agarose gels.
The transfection of vector pEJ1175 into parasites was tested by standard PCR using Taq DNA Polymerase (NEB, Ipswich, MA). 150 ng of total DNA was used as a template. Forward oligo oEJ1022 (5′-TAA CAT GGC CAT CAT CAA GGA GTT C) and reverse oligo oEJ1019 (5′-ATA CTT TCT CGG CAG GAG CA) were added to amplify a 2377 base pair DNA fragment including a partial mCherry and a partial neomycin sequence within the plasmid (Figure 1).
Total RNA from each DNAse treated sample was used to make cDNA by RNA reverse transcription reaction using SuperScript III Reverse Transcriptase, RNase OUT and oligo (dT)12–18 (Invitrogen, Carlsbad, CA) in a total volume of 20 µL volume. A no reverse transcriptase control was used in all experiments. The reaction was performed at 50°C for 40 min and then treated with 10 U RNase H (New England Biolabs, Ipswitch, MA) at 37°C for another 20 min to digest mRNA thoroughly. Both reverse transcriptase and RNase H were inactivated by incubation at 70°C for 15 min. The quality of cDNA was tested by PCR amplification of a 374 bp Sm23 gene fragment using primers oJM18 forward (5′-CGT TTG AAA GTA TGG GAT CCA TGG CAA CGT TGG GTA CTG GTA TGC) and oJM20 reverse (5′-GCC CTT GCT CAC CAC GGA TCC TTT GTA AAC AAC TGC AAC TAT GGC) (Supplementary Figure S1, Figure S2).
To analyze the expression of the mCherry gene under control of a human cytomegalovirus (CMV) promoter (pEJ1175, Figure 1A) and Sm23 promoter (pEJ1116, Figure 1B), qRT-PCR was carried out using primers oEJ1022 forward and oEJ1023 reverse (5′-TAC ATG AAC TGA GGG GAC AGG ATG T), to clone a 192 bp mCherry gene fragment from 60 ng cDNA template.
Two sets of primers were designed for detection of SmMef2 transcripts by qRT-PCR. The first set of primers measure SmMef2 within the first 399-base pair (the truncated region). The second sets of primers are located at the 3-prime end of SmMef2 (outside of the first 399 base pairs), and were used to detect wild-type SmMef2 transcripts. The other primers sets were designed for detection of four SmMef2's potential downstream targets. All primers were verified by Primer3 online software (http://frodo.wi.mit.edu/, Supplementary Table S1).
Sixty nanograms of cDNA from parasites treated with both PEI and pEJ1181 was used as template in a 60 µL qRT-PCR reaction with Power SYBR Green Master Mix (Applied Biosystems, Foster City, CA). Each reaction was divided into 20 µL triplicates and PCR was carried out and analyzed by StepOnePlus Real-Time PCR System (Applied Biosystems, Foster City, CA). This was done at least in triplicates for each sample. Transcript levels in schistosomes transfected with pEJ1175 and pLS068 were quantified by qRT-PCR, to differentiate between the non-specific gene amplification (mCherry) and wild-type SmMef2 overexpression, as a result of overexpression SmMef2,133, respectively. Negative controls (No RT and DNA only treatment) were run in parallel. The qRT-PCR conditions were as follows: 95°C for 10 min, 45 cycles of 15 s at 95°C, 30 s at 60°C and 30 s at 72°C. The melt-curve analysis of each pair of primers showed that only one specific product was amplified by each reaction. The relative gene expression was calculated by the ΔΔCt method according to the formula: expression rate = 2−ΔΔCt and cyclophilin was used as an endogenous control gene [40]. Experimental data were verified by Student's t-test, and a p-value less than 0.05 was considered to be statistically significant [50]. Amplification efficiencies of target genes and the endogenous control gene, cyclophilin, are optimal and comparable.
Both SmMef2,133 and the wild type SmMef2 gene with a c-Myc tag at the 5-prime end were amplified from pEJ1114 [49] and subcloned into pCI-neo vector using the methods described above. Seven to ten thousand schistosomula transfected with one of the two c-Myc tagged plasmids and were harvested 44 h after cercarial transformation. Samples transfected with pEJ1175 were used as a negative control. Schistosomula were washed with 1× PBS twice and resuspended in the lysis buffer (20 mM Tris-HCl, 200 mM NaCl, 1× PMSF and 1× Halt Protease Inhibitor Cocktail; Thermo Scientific, Waltham, MA), followed by 6× sonication of 15 s pulses, 30% amplitude with 1 min interval between each pulse. Cell lysate was then added with 5× SDS loading buffer and boiled at 100°C for 10 min and incubation for 5 min on ice. Fifty microliters of the supernatant from each cell lysate was resolved on NuPAGE 4–12% Bis-Tris ready-made gels (Invitrogen, CA). The protein was transferred to a nitrocellulose membrane (Thermo Scientific, MA) and blocked in 5% milk. The specific expression of c-Myc tag protein was detected by the mouse monoclonal IgG1 c-Myc (Myc.A7) primary antibody and a goat anti-mouse IgG-HRP secondary antibody (Santa Cruz Biotechnology, CA). Similarly, mCherry protein expression was detected by the mouse monoclonal IgG2a primary antibody (Novus Biologicals, Littleton, CO) and the goat anti-mouse IgG-HRP secondary antibody (Santa Cruz Biotechnology, CA), and assayed by western blot analysis.
PEI has been used successfully for gene delivery in mammalian cells in vivo and in vitro [51]–[53], and recently in the snail, Biomphalaria glabrata [38]. The success of gene delivery using PEI in snails inspired us to ask whether PEI could be used as an alternative to electroporation to transport genetic material into schistosomes. To test this possibility, we incubated 4-hour schistosomula for 40 hours in a PEI/plasmid DNA mix in complete RPMI (Figure 2, Lanes 5 and 6, see Material and Methods). The plasmid DNA contained the mCherry gene regulated by a strong CMV promoter (Figure 1A). As negative controls for transfection, equal numbers of schistosomula were cultured in RPMI medium containing (1) only PEI, (2) only DNA plasmid, or (3) only schistosomes, no PEI and no DNA (Figure 2, Lanes 2–4). We examined the efficacy of the use of PEI for the introduction of plasmid DNA into schistosomula by using different N/P ratios, 6∶1 (Lane 5) and 11∶1 (Lane 6). Forty hours after transfection, we treated all schistosomula with DNase to remove any contaminating external DNA. Total schistosome DNA was extracted from each group and used for standard PCR analysis to test for the presence of the 2,377 base pair fragment stretching from the mCherry gene to the neomycin gene of plasmid pEJ1175 (Figure 1A). We found that the expected 2,377 base pair fragment was amplified only from samples containing both PEI and plasmid DNA (Figure 2, Lanes 5 and 6), whereas all negative control samples (Lanes 2–4) had no product. This result is consistent and reproducible (n>5), and demonstrates that PEI can be used to introduce plasmid DNA into schistosomes.
After treatment with PEI, we observed the parasites by light microscopy. Under some conditions, PEI can have potential toxic effects to human cells [54]. To test for potential lethality to schistosomes due to PEI exposure, we incubated approximately 8,000 schistosomula in the 2 mL PEI∶DNA mix in complete RPMI media for two days. We found no significant differences in viability between schistosomula incubated with PEI (with plasmid DNA) and schistosomula grown without PEI in the medium (Supplementary Table S2). To assay viability, we pipeted the schistosomula in media and removed 5–10% of the parasite culture after 1 hour, 1 day, and 2 days. The schistosomula were allowed to settle briefly and were counted. We counted the schistosomula that settled on the culture dish and that were motile as alive, but schistosomula that did not settle on the plate, were not observed to be motile, or that appeared to lysed, were counted as dead. We rationalized that if PEI is deleterious, then under stressful conditions where the schistosomula were crowded due to large numbers, toxicity might be exacerbated. Our data indicate that lethality due to exposure to PEI is not a major concern for culturing schistosomula at the concentrations utilized in these experiments.
The nitrogen/phosphate (N/P) molar ratio of PEI∶DNA complexes is an important factor for effective transfection of DNA into mammalian cells [55], [56]. An N/P ratio of 6∶1 is optimal for the transfection of most mammalian cells [57]. Using PCR to amplify a 2,377 base pair sequence from plasmid pEJ1175 (Figure 1), we assessed whether a change in the ratio of PEI to DNA would affect the efficacy of the transfection of the plasmid DNA in schistosomes. We assayed two N/P ratios, 6∶1 PEI∶DNA, and 11∶1 PEI∶DNA, and both ratios were found to be effective for transfection of DNA into schistosomes (Figure 2, Lanes 5 and 6, respectively).
Since PEI can be used to insert DNA into schistosomes, we assayed whether the mCherry reporter gene, under control of a CMV promoter, could be expressed from a plasmid in transfected schistosomes. Previously, schistosomes transfected using either particle bombardment or electroporation showed that the CMV promoter is capable of inducing heterologous gene expression in these parasites [17], [58]. Thus, the use of CMV as a testable promoter was considered valid. To test for expression from the CMV promoter after PEI mediated transfection, DNase treated total RNA was extracted from schistosomula after treatment with or without PEI, and subsequently followed by two-step reverse transcription PCR (RT-PCR) to amplify a 192 base pair fragment of the mCherry RNA transcript (Figure 1A). Since mCherry is not endogenous in schistosomes, only parasites that have been successfully transfected with the plasmid will be capable of expressing mCherry. Our RT-PCR analysis confirms that the CMV promoter is sufficient to induce transcription of the mCherry reporter gene in schistosomula (Figure 3A, Lane 2), but not in the negative control sample (Figure 3A, Lane 4). No product was observed in a control sample tested without reverse transcriptase (data not shown).
We next evaluated whether a larger DNA plasmid could be transfected into schistosomes, and assayed the expression of a reporter gene on the plasmid directed by the schistosome Sm23 promoter. Sm23 is an integral membrane protein in schistosomes that is constitutively expressed during the schistosome life cycle [59], [60]. We cloned 2000 base pairs of the Sm23 upstream activation sequence containing the Sm23 promoter into the vector pGBKT7 (Clontech). Directly under control of the Sm23 promoter, we subcloned the mCherry gene (Figure 1B) to produce the 10.4 kb plasmid, pEJ1116 (Figure 1B). We showed that PEI could be used for transfection of the smaller 6.2 kb mCherry vector pEJ1175 (Figure 1A). Here, we evaluated PEI for the transfection of a larger 10,4 kb DNA plasmid. We transfected the 10.4 kb plasmid pEJ1116 into 4-hour schistosomula. After transfection of the 10.4 kb plasmid pEJ1116 into 4 hr schistosomula, we assayed for the amplification of a 192 base pair mCherry product to test for the expression of the mCherry transcript, as described above. Expression of the mCherry transcript can only occur in schistosomula that are successfully transfected and then, only if the plasmid based promoter, Sm23, is functional. After RT-PCR analysis, we found that mCherry is expressed from the Sm23 promoter on the 10.4 kb plasmid, demonstrating that PEI is sufficient to aid in the transfection of large plasmids into schistosomes, and that the Sm23 UAS is sufficient for gene expression from a plasmid (Figure 3A, Lane 3).
We investigated whether schistosomes transfected with a plasmid transcribing mCherry, under control of the CMV promoter, were able to express the mCherry protein using Western blot analysis (Figure 3B). Using an antibody against mCherry, we observed a 28 kD mCherry protein in schistosomes expressing mCherry from the CMV promoter (Figure 3B, Lane 1), but this was not observed in untransformed schistosomes (Figure 3B, Lane 2).
We previously identified and characterized Mef2 in schistosomes (SmMef2) [40], a conserved transcriptional activator that is essential for myogenesis in Drosophila [41]. Mef2 also has diverse functions regulating cellular differentiation, morphogenesis and proliferation [61], [62]. Recent studies in mice provide evidence that Mef2 proteins can modulate signaling of the WNT pathway during skeletal muscle regeneration [63]. We previously reported that there were potential Mef2 DNA binding sites within 500 bp of the translation start sites of two schistosome genes encoding WNT homologs: Smp_152900, encoding for Wnt1, and Smp_167140, which we assert, based on conserved sequence analysis, encodes Wnt2 [64].
The ability to easily transfect and induce gene expression in schistosomes with low lethality using PEI, and the developmental question of whether Mef2 plays a role in regulating genes in the WNT pathway, provided an opportunity to test whether schistosome transfection with PEI could be used as a genetic tool to dissect basic gene functions in schistosomes. To address this, we propounded the idea that expression of a SmMef2 mutant that can (1) bind DNA, but (2) be unable to efficiently induce Mef2 transcriptional target genes, could potentially interfere with normal SmMef2 activator function in vivo by acting as a competitive inhibitor and act as a potential genetic dominant negative in schistosomes. SmMef2 has a N-terminal DNA binding and a C-terminal transactivation domain. We removed the C-terminal transactivation domain of SmMef2, producing a truncation mutant comprising the first 133 amino acids containing the MADS box and Mef2 DNA binding domains to make SmMef2,133. We cloned the truncated schistosome Mef2,133 gene so that its expression was controlled by the strong CMV promoter (Figure 1C) and transfected schistosomula with this construct as before. After 40 hours, we extracted RNA and used qRT-PCR to compare SmMef2,133 transcript levels to an untransfected control. The control was incubated with the Mef2,133 plasmid without PEI. We found that SmMef2 levels were increased twenty-fold higher than SmMef2 levels in the untransfected control (Figure 4 A), demonstrating significant upregulation of the SmMef2 transcript.
Since the promoters of SmWnt1 and SmWnt2 genes have Mef2 binding sequences, we tested whether SmMef2,133 overexpression has an effect on the transcript levels of SmWnt1 and SmWnt2 by qRT-PCR. When SmMef2,133 is overexpressed, we found that Wnt 1 transcript levels are downregulated some 2 fold (Figure 5A), and Wnt2 transcript levels were downregulated more than 5 fold compared to the untransfected control (Figure 5B). As a negative control, when mCherry was overexpressed we observed no significant changes in Wnt1 or Wnt2 transcript levels (Figure 5 A,B). When we tested a muscle LIM gene (Smp_143130) and a TGF beta family gene (Smp_063190) that have a potential Mef2 binding site, we found no significant difference in transcript levels (data not shown, Supplementary Table 1).
SmMef2 levels are highest in 4-hour schistosomula relative sporocysts, cercariae, and adult worms [40]. We cloned the SmMef2 gene under control of the CMV promoter, as was previously described for the truncation mutant, SmMef2,133, and overexpressed SmMef2.. To distinguish between expression of SmMef2 and the mutant SmMef2,133, we designed DNA oligonucleotides that recognize SmMef2 (Figure 1D, oligonucleotides f and g) but that do not recognize SmMef2,133 transcript. DNA oligonucleotides that recognize sequences in SmMef2,133 (Figure 1D, oligonucleotides d and e) also recognize SmMef2 sequences. We found that expression of SmMef2 was elevated 30-fold relative to the control (Figure 4B), whereas overexpression of the mCherry negative control had no effect on SmMef2 transcript levels. A second pair of oligonucleotides (Figure 1D, oligonucleotides d and e) that recognize sequences in SmMef2,133, and SmMef2, showed a 25 fold increase in Mef2 transcript levels (Figure 4D), indicating the confidence level of the qRT-PCR data.
We then assayed if overexpression of SmMef2 could have a positive effect on WNT levels. We found a very slight increase in Wnt1 transcript, but no significant change in Wnt2 transcript levels. The Mef2 protein is reported to positively regulate its own transcription [65]. We investigated whether SmMef2 was capable of regulating its transcription levels in schistosomes. To address this genetically, we overexpressed the truncated mutant, SmMef2,133, and measured SmMef2 levels by qRT-PCR using oligonucleotides pairs that distinguish SmMef2 transcript from SmMef2,133 transcript as described (Figure 1D). We found that overexpression of SmMef2,133 resulted in a 3-fold decrease of SmMef2 transcript (Figure 4C), strongly suggesting a role for Mef2 positively regulating its expression.
We assayed for distinct changes in viability between schistosomula expressing the dominant negative mutant, SmMef2. One thousand schistosomula were transfected with plasmid expressing SmMef2,133, SmMef2 or a nonspecific control, mCherry. These were grown for 7 days in Basch medium (see Material and Methods). After 7 days, all worms were quantified. We found 640, 550, and 620 schistosomula transformed with SmMef2,133, SmMef2, and the mCherry control respectively remained alive. Thus, we observed no significant differences in survival rate.
SmMef2 and SmMef2,133 transcript levels are upregulated when expressed from the CMV promoter. We addressed whether this expression led to production of protein. To assay protein expression from the reporter construct in schistosomes, we added a c-Myc tag sequence to the 5-prime ends of the truncated mutant SmMef2,133 and the wt SmMef2 genes. We then extracted protein from schistosomula transfected with the plasmids expressing myc tagged SmMef2 and myc-tagged SmMef2,133 at 40 h post-transfection, and we assayed protein expression by Western blot analysis. Both c-Myc tagged SmMef2,133 (18.3 kDa) and c-Myc tagged SmMef2 (77.1 kDa) were detected by Western analysis and visualized using a gel documentation system (Figure 6). These data confirm that exogenous gene transcripts are translated into protein.
Here, we have shown 1) that PEI facilitates the transfection of nucleic acids into schistosomes, and 2) that it facilitates the molecular genetic analysis of signaling and transcriptional pathways in schistosomes, addressed here by assessing SmMef2 function on SmWnt1 and SmWnt2 genes as proof of principle. 3) We provide an example of dominant-negative gene expression in schistosomes, and 4) provide evidence that SmMef2 is autoregulatory, and show data supporting its role in the regulation of the WNT pathway.
The idea to examine PEI for the transfection of DNA into schistosomes was inspired by the report that showed it was a useful agent for the successful transfection of B. glabrata snails [38]. PEI is an established transfection agent for individual mammalian cells [33], for tissue culture [66], and for tumor therapy [67]. The transfection of live snails led us to test whether it could also be efficacious as a transfection agent in schistosomes. We found that PEI (Aparna Biosciences, Rockville, MD) is extremely effective for the transfer of nucleic acids into schistosomula. DNA plasmids up to 10.4 kb in size were successfully introduced into schistosomula and were functional for transcription of a heterologous reporter. We recommend the use of PEI as an alternative to the aforementioned transfection approaches previously used in schistosomula. Although electroporation has been the most widely used method for transfection in schistosomes, electroporation can lead to significant mortality after passage of electrical amperage into worms. It also requires the purchase of an electroporator and cuvettes. The schistosomes must be transferred into cuvettes with nucleic acids in minimal salt solution to avoid arcing, prior to in vitro or in vivo culturing, increasing the possibility of contamination. Our data suggests that PEI at the levels used and described here does not increase lethality to transfected parasites. In addition, the use of PEI as a transfection agent is straightforward, requiring the addition of PEI and less than 10 µg of DNA in our studies. Mechanistically, DNA and PEI are incubated in the same culture medium containing the parasites, making the technique simple.
We tested two different promoters for expression in schistosomes after transfection using PEI- the CMV promoter, and the schistosome Sm23 promoter. Both promoters were capable of inducing gene expression from plasmids when tested 2 days after transfection was initiated. Initially, we used mCherry as a reporter gene for expression under the premise that we could visually screen for transfected schistosomula under a microscope and that we could be able to determine the exact efficiency of transfection by quantifying the percentage of fluorescent schistosomes. We found that the background autofluorescence of schistosomes masked consistent discrimination between transfected and untransfected schistosomes. Since the PEI does not specifically localize DNA during transfection to a discrete locations in the parasite (i.e. the gut, the nerves, schistosome surface) and there is as yet no organelle specific reporter described in schistosomes, it is possible that diffuse fluorescence of the mCherry reporter cannot be observed visually using our methods. Thus, we assayed transfection and reporter activity by directly quantifying schistosome RNA levels in the transfected parasites.
The ease of this approach to transfect schistosomes in combination with our interest in transcriptional regulation and our previous work on SmMef2 in schistosomes, stimulated us to inquire if we could develop a genetic model to investigate basic biological questions on SmMef2 gene expression in schistosomes. We predicted that expression of a truncated SmMef2 protein, that contains the DNA binding domain but no transactivation domain (SmMef2,133), could antagonize or compete with wild-type SmMef2 for binding to SmMef2 transcriptional targets, and potentially interfere with expression of SmMef2 target genes. We identified potential Mef2 binding elements in several schistosome promoters, including SmWnt1 and SmWnt2 [40] When we overexpressed the truncated SmMef2,133, we found that both SmWnt1 and SmWnt2 transcript levels were reduced by two-fold and five-fold, respectively. We similarly overexpressed a control mCherry gene to test whether overexpression of any gene could lead to general down regulation of schistosome gene expression, but observed no change in Wnt1 or Wnt2 transcript levels. This indicates that Wnt1 and Wnt2 transcription is regulated by SmMef2, either directly or indirectly. Although, the presence of Mef2 binding sites in the promoters of SmWnt1 and SmWnt2 might suggest that this interaction is direct.
We overexpressed SmMef2 to assay whether elevated levels of SmMef2 levels could lead to an increase in Wnt1 or Wnt2 transcript levels. We found an indication of change in Wnt1 transcript levels. Mef2 transcript levels are highest in schistosomula compared to sporocysts, cercariae, or adults. It could be that the normal high expression level of SmMef2 at this stage saturates Mef2 targets and increasing Mef2 levels higher has little effect. This rationale corresponds to work done on myoblast cells where a dominant negative version of Mef2 reduces MyoD induced myogenic colony formation, but overexpression of Mef2 had no effect on myogenic conversion [68]. Alternatively, it could simply be that Mef2 requires an interacting factor for the expression of some Mef2 targets or posttranslational modification of SmMef2, which has been established for several Mef2 target genes [69]–[73].
We also found that SmMef2 was capable of regulating its own transcript levels in whole schistosomes, which has been reported previously in mammalian cell culture [65]. When the dominant negative SmMef2,133 was overexpressed, SmMef2 transcript levels were reduced three fold, showing that SmMef2 regulates its own transcription. In Drosophila, Mef2 interacts with microRNAs (miRNA), specifically miR-1, that targets and reduces the mRNA stability and translation of class II histone deacetylases (HDACs), specifically HDAC4. HDAC4 is a transcriptional repressor of muscle specific genes [74]. In this model, downregulation of Mef2 would reduce expression of miR-1. With reduced miR-1, HDAC levels are not suppressed and in turn can repress of SmMef2 (Figure 7). Thus, overexpression of Mef2 could presumably lead to an exponential increase in its expression. On the contrary, Mef2 also induces targets that negatively regulate its levels. Mef2 activates the miR-92b, a recently identified microRNA that represses Mef2 transcript [75]. Similarly, Mef2 can activate HDAC9. HDAC9 in turn interacts with Mef2 proteins to repress Mef2 transcriptional ability [76]. Thus, Mef2 forms a feedback loop that maintains an equilibrium in Mef2 expression and Mef2 target induction. This may also explain why overexpression of SmMef2 did not produce a significant increase in Wnt1 and Wnt 2 expression levels.
Since Mef2 is important for myogenesis and neuron survival in other organisms, we predicted that overexpression of SmMef2,133 would produce a phenotypically distinct mutant due to lack of muscle development or a reduction in neuronal survival. However, after microscopic analysis, we unable to identify a visual physical difference in either motility or in shape between schistosomes overexpressing SmMef2,133, or SmMef2, or a mCherry control schistosomula, even after 7 days. Nor did we find a quantitative change in a predicted muscle Lim gene (Smp_143130) or a TGF beta family gene (Smp_152900) at two days, which have a potential SmMef2 binding site (data not shown). One reason for this could be that factors other than SmMef2 can participate in muscle or neuromuscular development [41], [77]. Alternatively, the transcription activation function of SmMef2 protein function may be inhibited by a HDAC9-like protein in schistosome preventing induction, or simply that the schistosomula should be cultured for a longer periods to observe any gross phenotypic changes contributed to overexpression of SmMef2,133. Nonetheless, these data, and recently published data in mice, corroborate that Mef2 plays a role in regulating the WNT pathway, a connection that has not been extensively explored.
In Drosophila and in mammals, Mef2 activates genes that participate in the Notch-Delta, Hedgehog, Fibroblast Growth Factor and Epidermal Growth Factor pathways [41]. This report adds the WNT pathway to that list (Figure 7). It will be of interest to further examine the role of SmMef2 in these pathways in schistosomes.
The use of PEI for transfection is a simple tool that can be used to dissect schistosome genetic pathways. We have not yet tested whether it facilitate the transfection of nucleic acids into other stages of schistosome development, nor have we made a direct comparison between PEI transfection and electroporation, which could be informative. The successful use of PEI for the transfection of mammalian cells in culture, for whole snails, and for schistosomula, provides promise that this approach may work in other schistosome stages, and that it could be successfully used for other flat or roundworm species that have been challenging to transfect. In addition, we are currently evaluating commercial and noncommercial promoters for their ability to drive gene expression in schistosomes, using PEI for nucleic acid delivery. Eventually, we expect that schistosome expression vectors could be selected based on promoter transcription rates, stage-specific or location dependent expression, or for use as cellular markers. In addition, since PEI is thought to function by protecting nucleic acid from digestion [35], RNA interference constructs may be potentially used for transcript knockdown by using this approach.
|
10.1371/journal.ppat.1007204 | Asymmetric antiviral effects of ebolavirus antibodies targeting glycoprotein stem and glycan cap | Recent studies suggest that some monoclonal antibodies (mAbs) specific for ebolavirus glycoprotein (GP) can protect experimental animals against infections. Most mAbs isolated from ebolavirus survivors appeared to target the glycan cap or the stalk region of the viral GP, which is the envelope protein and the only antigen inducing virus-neutralizing antibody response. Some of the mAbs were demonstrated to be protective in vivo. Here, a panel of mAbs from four individual survivors of ebolavirus infection that target the glycan cap or stem region were selected for investigation of the mechanisms of their antiviral effect. Comparative characterization of the inhibiting effects on multiple steps of viral replication was performed, including attachment, post-attachment, entry, binding at low pH, post-cleavage neutralization of virions, viral trafficking to endosomes, cell-to-cell transmission, viral egress, and inhibition when added early at various time points post-infection. In addition, Fc-domain related properties were characterized, including activation and degranulation of NK cells, antibody-dependent cellular phagocytosis and glycan content. The two groups of mAbs (glycan cap versus stem) demonstrated very different profiles of activities suggesting usage of mAbs with different epitope specificity could coordinate inhibition of multiple steps of filovirus infection through Fab- and Fc-mediated mechanisms, and provide a reliable therapeutic approach.
| Recent progress in isolation of mAbs from survivors of filovirus infections suggests that the human adaptive immune system is capable of producing strong antibody responses. However, the effects of mAbs with different epitope specificity on individual steps of filovirus infection are still unclear. We evaluated a panel of mAbs obtained from survivors of natural filovirus infections, specific for the glycan cap or stem region of GP, for their effects on the attachment of viral particles to the cell surface, intracellular traffic of viral particles, proteolytic processing of GP, its interaction with the NPC1 receptor, cell-to-cell virus transmission, virus egress from infected cells, activation of natural killer cells and antibody-dependent cellular phagocytosis through Fc-mediated mechanisms. We found that antiviral activity of glycan cap-specific antibodies results from inhibition of attachment, cell-to-cell transmission and inhibition of virion budding. In contrast, the antiviral mechanisms of stem-specific antibodies were found to be inhibition of virus release from endosomal network to the cytoplasm, and also activation of natural killer cells and phagocytosis mediated by monocytes and neutrophils. The data provide new insight into the development of immune protective mechanisms during natural human infection, and have important implications for the treatment of filovirus infections by passively-transferred antibodies and vaccine design.
| Filoviruses are enveloped, filamentous-like viruses with non-segmented RNA genome of negative polarity. The Ebolavirus genus of the Filoviridae family includes five species: Ebola (EBOV), Sudan (SUDV), Bundibugyo (BDBV), Taï Forest (TAFV) and Reston (RESTV) viruses. Most of these viruses are responsible for highly lethal disease outbreaks, for example the occurrence of 11,323 human fatalities during the 2013–2016 EBOV epidemic in West Africa [1, 2]. Despite intense international collaborative efforts, there is still no licensed therapeutic available against filovirus disease.
GP is the sole ebolavirus envelope protein responsible for cell entry and, hence, serves as the primary target for antibody-based therapies and as antigen for vaccine development [3]. The primary nucleotide sequence of the GP gene encodes soluble glycoprotein (sGP), which shares its 295 N-terminal amino acid residues with GP, whereas GP mRNA synthesis requires the insertion of an extra adenosine into the nascent mRNA via stuttering of the EBOV RNA-dependent RNA polymerase over the transcriptional editing site [4]. The mature GP at the surface of nascent virions represents a 450 kDa trimer assembled from GP1/GP2 heterodimers [3]. The GP1 subunit mediates cellular attachment of viral particles and includes base domain interacting with the GP2 subunit, and a chalice-like structure formed by the receptor-binding domain (RBD), glycan cap and heavily N- and O-glycosylated mucin-like domain (MLD). The RBD is sequestered in the chalice bowl, whereas the glycan cap and MLD are exposed and covered by a thick glycan layer that likely shields much of GP from effective humoral immune recognition [5, 6]. The GP2 subunit forms a GP stalk containing the hydrophobic internal fusion loop (IFL), two heptad repeats (HR1 and HR2), the membrane-proximal external region (MPER), the transmembrane anchor and the short cytoplasmic domain. This subunit is responsible for fusion of the viral and host cell membranes during the entry.
EBOV attachment to the cell surface occurs via two types of low affinity interactions. First, using a set of N- and O-linked glycans on the MLD and the glycan cap of GP1, virus can bind to multiple C-type lectins. Second, EBOV uses phosphatidylserine molecules incorporated into viral envelope to bind TIM/TAM receptors (reviewed in reference [7]). After adherence, virions internalize to the cell by macropinocytosis, and subsequently traffic through the labyrinth of endosomal compartments, where critical pH-dependent GP priming by cathepsin proteases takes place. The consecutive processing of GP by cathepsins L and B results in the excision of most of the GP1 subunit, which includes the glycan cap and MLD, and exposure of the RBD for interaction with the intracellular filovirus receptor, the cholesterol transporter protein NPC1 [8–11].
The first human EBOV neutralizing mAb, KZ52, was generated from RNA isolated from the bone marrow of a survivor of natural infection [12]. This mAb protected guinea pigs from lethal EBOV challenge [13], but failed to protect non-human primates (NHPs) [14]. The feasibility of post-exposure prophylaxis with antibodies in monkeys was demonstrated six years ago with total IgG purified from convalescent serum of macaques [15]. Several mAb cocktails that protect NHPs from EBOV infection have been developed subsequently: MB-003 (human or human/mouse chimeric mAbs c13C6, h13F6 and c6D8), ZMAb (murine mAbs m1H3, m2G4 and m4G7) and ZMapp (human/mouse chimeric mAbs c13C6, c2G4 and c4G7) [16, 17]. The latter cocktail, which showed a beneficial effect, however, failed to demonstrate the pre-specified statistical threshold for efficacy in a clinical trial performed during the West Africa epidemic [18]. Human mAbs from survivors of natural ebolavirus infection, rather than antibodies raised in experimentally vaccinated or infected animals, are preferable for the development of therapeutics against filovirus infections. Such antibodies have a full compatibility of Fc fragments with the receptors on human immune cells, which is expected to make them more effective due to Fc-mediated protective mechanisms. While several published studies demonstrate binding of filovirus mAbs from human survivors to GP at the atomic level [5, 19–24], none of them are characterized for the ability to affect multiple steps of viral replication. Here, we present a comprehensive comparative study of Fab- and Fc-mediated biological functions of a panel of ebolavirus mAbs from human survivors [20] targeting epitopes in the GP glycan cap and stalk region. The results indicate that both types of mAbs interfere with and target different steps of viral replication, including virus entry, egress, cell-to-cell transmission, secondary infection and facilitate destruction of infected cells through antibody-dependent cellular cytotoxicity (ADCC) and antibody-dependent cellular phagocytosis (ADCP) mechanisms. However, important differences between the two groups also were observed, suggesting complementary effects of various antibodies generated during natural filovirus infections.
In previous work, we isolated and characterized multiple mAbs from the blood of human survivors of natural BDBV infection [20]. To study mechanisms of inhibition of filovirus replication by antibodies, we selected a panel of mAbs from four donors, with differing virus neutralization properties and affinity to GPs of EBOV, BDBV and SUDV: BDBV52, BDBV270, BDBV41, BDBV289, BDBV259, BDBV317 and BDBV223 (Fig 1). Identification of epitopes demonstrated that most of the mAbs can be grouped into those recognizing two major antigenic sites: those specific for glycan cap and those specific for stem [20, 25, 26].
To test whether any of the mAbs inhibited attachment, we incubated BDBV virus-like particles (VLPs) and mAbs at 37°C for 1 hour, added the mixtures to Vero-E6 cell culture monolayers in chambered slides, and incubated on ice for 1 hour. Then, cells were fixed, and cell-bound VLPs were immunostained. Confocal microscopic analysis of cell monolayers demonstrated a strong binding inhibition only by mAb BDBV289 (Figs 2A and S1). Agreeing with the confocal microscopy results, flow cytometric analysis demonstrated 2-fold inhibition of viral binding by BDBV289, some enhancement of binding by BDBV52, BDBV270, BDBV259, and strong enhancement for BDBV223 (Figs 2B and S2). We did not observe any enhancement of viral binding in our confocal microscopic assay for the BDBV223-treated samples. The reasons for this are unclear and may be related to the use of attached cells for the confocal microscopy and suspension cells for flow cytometry. An irrelevant human mAb 2D22 of the IgG1 isotype, specific to dengue virus envelope protein in the dimeric structure [27], was used as a control. A significant difference was not observed between the 2D22 and the no-mAb groups (S3 Fig).
For the post-attachment inhibition assay, BDBV was adsorbed first on Vero-E6 cell culture monolayers for 20 min at 4ºC. Then, mAbs were added, incubated for 20 min at 4ºC, and viral plaques were developed at 37ºC. The BDBV223 mAb strongly reduced plaque numbers, suggesting that MPER-targeting mAbs can effectively block post-attachment steps of virus replication (Fig 2C). The inhibiting effect of the other MPER-specific mAb, BDBV317, was comparable to that of BDBV289 and BDBV259 in this assay. Only marginal post-attachment inhibition was demonstrated for BDBV41 and BDBV270 mAbs from the glycan cap-targeting group. To assess the total impact of mAbs on inhibition of virus entry (binding and post-attachment steps), we used a chimeric replication-competent EBOV in which GP was replaced with its counterpart from BDBV and that expresses eGFP from an added transcriptional cassette to visualize infected cells (EBOV/BDBV-GP) [28]. EBOV/BDBV-GP was incubated with mAbs for 1 hour and adsorbed on Vero-E6 cell culture monolayers for 40 min at 4ºC. Then, cells were incubated for 24 hours at 37ºC, and the percentages of infected eGFP+ cells were determined by flow cytometry (Figs 2D and S4A). As expected, MPER-specific mAbs completely abolished virus entry in cells. High levels of inhibition also were demonstrated by BDBV270, BDBV41, BDBV289 and BDBV259 mAbs. Unexpectedly, the non-neutralizing BDBV52 mAb slightly increased virus entry into the cells.
To investigate the effect of mAbs on intracellular steps of virus life cycle, trafficking of mAb-treated VLPs through the cell organelle network was analyzed by confocal microscopy. BDBV or EBOV VLPs were mixed with mAbs, placed on Vero-E6 cell culture monolayers, incubated for 30 or 60 min and immunostained for EBOV VLPs and for endosomal markers. Unexpectedly, BDBV259 and BDBV317, but not the other mAbs, caused accumulation of VLPs in late endosomes, as evidenced by GP/Rab7 co-localization (Figs 2E, S5, S6 and S7). However, the effect of BDBV317 was relatively short, as the co-localization disappeared after one hour of incubation (S5 and S6 Figs), probably suggesting instability of BDBV317/GP complexes in the acidic pH of endosomes. When treated with BDBV223, but not the other mAbs, VLPs were found to be co-localized with the lysosomal-associated membrane protein 1 (LAMP-1) marker of lysosomes as early as 30 min after infection (Figs 2E and S6), which was still observed at 60 min (S5 and S6 Figs).
We next tested binding of mAbs to BDBV GP at low or neutral pH by ELISA (Fig 2F). Binding of BDBV317 mAb was impaired at low pH compared to neutral pH, consistent with the short duration of BDBV317/Rab7 co-localization. In contrast, binding of BDBV259 was 1.5 times higher at low pH compared to neutral pH, whereas the difference for BDBV223 mAb was as much as 3.6 times higher. Hence, we propose that an acidic pH environment stabilizes BDBV223/GP complexes, allowing this antibody to retain viral particles inside lysosomal compartments and prevent nucleocapsid entry into the cytoplasm.
We hypothesized that the accumulation in acidic compartments observed for BDBV223, BDBV317 and BDBV259 mAbs was caused by inhibition of binding of GP to NPC1 in the late endosomes. To test this hypothesis, we developed Förster resonance energy transfer (FRET) analysis using NPC1 fused to red fluorescent protein (NPC1-RFP) and GP immunostained with AlexaFluor 647. Vero-E6 cells were transfected with NPC1-RFP-expressing plasmid and incubated overnight. A modified EBOV/BDBV-GP that does not express eGFP (EBOV/BDBV-GP_no eGFP) was pre-incubated with selected mAbs for 60 min at 37ºC. NPC1-RFP-transfected Vero-E6 cell culture monolayers then were inoculated with virus-mAb complexes at an MOI of 10 PFU/cell for 2 hours, fixed, and GP was immunostained. FRET analysis was performed by scanning confocal microscopy; the NPC1-GP interaction was quantified by changes in FRET efficiency when compared with virus in the absence of mAbs (S8 Fig). We analyzed BDBV223, BDBV259 and BDBV317 in comparison with the glycan cap-specific mAb BDBV289 as a negative control, since BDBV289 inhibits attachment and entry (Fig 2A and 2B) and is expected not to reach endosomes. The FRET efficiencies for virus samples treated with mAbs were equivalent to those without mAb, suggesting that the tested antibodies did not affect binding of the virus to NPC1. We also compared the numbers of FRET-positive events, and observed a dramatic increase with BDBV223 and a more modest increase with BDBV259. The increase in the number of virus-associated events was consistent with the increased trapping of the VLPs treated with these two mAbs in endosomes (Fig 2E). Notably, BDBV223 and BDBV259, but not the other mAbs tested, were found to bind to GP at low pH (Fig 2F).
As described above, GP is processed by cysteine proteases, cathepsins B and L, resulting in the removal of glycan cap and MLD from GP1 subunit followed by interaction of the exposed RBD with the C-loop of NPC1. Treatment of EBOV GP with the bacterial metalloproteinase thermolysin also results in deletion of the glycan cap and MLD, thus mimicking endosomal proteolysis of GP mediated by cathepsins [29, 30]. To test if selected mAbs interfere with late stages of virus cell entry by interacting with GP after its cleavage, we treated sucrose gradient-purified replication-competent vesicular stomatitis virus enveloped with BDBV GP (VSV/BDBV-GP) [31] with thermolysin and compared its neutralization with non-treated virus. As shown in Fig 3A, thermolysin treatment of virions resulted in complete proteolysis of GP1 subunit, and slight reduction of the virus titer (3.8-fold). Incubation of intact VSV/BDBV-GP with neutralizing glycan cap-specific antibodies led to a dramatic reduction of virus titers, whereas thermolysin-processed virus was resistant to BDBV270, BDBV41 and BDBV289, with no effect shown for the non-neutralizing BDBV52 mAb against either virus preparation. In contrast, GP cleavage with thermolysin did not reduce virus sensitivity to GP2-specific BDBV259, BDBV317 or BDBV223 mAbs, suggesting that these mAbs interact with the full-sized and processed fusion-active form of GP equally well, and, thus can inhibit multiple steps of virus entry. However, no mAb prevented the proteolysis of GP after treatment with cathepsin B and cathepsin L, as the 20 kDa GP1 fragment band resulting from the digestion of GP was present in all samples treated with cathepsin regardless of the mAb used (Fig 3B). The differences in the band intensity observed with different mAbs are probably caused by binding of mAbs to additional cathepsin cleavage sites, which are not involved in generation of the 20 kD fragment.
Secondary infection of cells by transfer of virions and intermediate products of viral replication (genome copies, viral proteins or the whole vRNP complexes) across the cytoplasmic bridges between infected and uninfected cells was shown to play an important role in the pathogenesis of HIV [32], influenza virus [33, 34] and EBOV [35]. Such cell-cell contacts can increase the effective viral MOI at the sites of transmission, making this route of infection spread 70-fold [35] to 2–3 orders of magnitude [32] more efficient compared to cell-free dissemination. Moreover, use of the alternative intercellular gateway for direct access to the cytoplasm of a new host cell allows virus to escape from antibodies targeting initial steps of cell entry and/or virus egress. To analyze the effects of mAbs on cell-to-cell transmission, we used a flow cytometry-based approach previously described for HIV studies [36]. THP-1 monocytes (donor cells) infected with EBOV/BDBV-GP (which expresses eGFP) were incubated with mAbs for 1 hour, placed at the top of Vero-E6 cell culture monolayers (acceptor cells) pre-stained with CellTrace Far Red, and incubated for 72 hours. Since all mAbs but BDBV52 are strong suppressors of viral entry (Fig 2D), their constant presence in the cell medium was expected to prevent spread of infection through the medium. Indeed, titration of supernatant aliquots harvested from co-cultures of THP-1 and Vero-E6 cells on days 3–5 after the inoculation of monocytes showed an absence of detectable live viral particles in samples containing BDBV270, BDBV289, BDBV223 or BDBV317 mAbs, but not in those with 2D22 or no mAb (S9 Fig). To measure cell-to-cell virus transmission, CellTrace FarRed+/eGFP+ cells were quantified by flow cytometry and expressed as percentages of the total CellTrace FarRed+ population (Figs 3C and S4B). Consistent with the previous experiments (Figs 2C–2F and 3A), the MPER-specific BDBV223 appeared to be the most potent mAb, as it completely suppressed the infection of acceptor cells at all concentrations tested. The other MPER-specific BDBV317, along with one glycan cap-specific BDBV270, demonstrated a clear dose-dependent inhibition of viral transmission, and the glycan cap-specific BDBV289 showed a somewhat lesser inhibition. The non-neutralizing glycan cap-specific BDBV52 mAb did not cause any detectable inhibitory effect. The overall ability of mAbs to inhibit virus transmission better corresponded to their ability to inhibit viral replication at higher (0.1 PFU/cell) than the lower (0.01 PFU/cell) MOI (S10 and S11 Figs), as was previously demonstrated for HIV [32]. The overall antibody potency in the cell-to-cell transmission assay was much lower compared to the mAb effects on primary virus entry (Fig 2D) suggesting that higher mAb doses may be required to overcome secondary virus infection through the intercellular connections.
Upon completion of replication cycle inside the cell, progeny virions release into the extracellular matrix and spread the infection to bystander cells. The budding of MARV can be prevented by antibodies even in the absence of virus neutralization detectable in plaque reduction assays, possibly by bivalent cross-linking of newly formed virions to each other and to the viral proteins exposed on the cell membrane [37]. We therefore analyzed the effect of mAbs on virus release from infected Vero-E6 cells. Since the presence of the neutralizing mAbs in the medium would interfere with analysis of released viral particles by plaque assay, we quantitated both live and neutralized virus released from infected cells and present in the medium by quantifying viral genomic RNA by droplet digital RT-PCR. The egress of virus was strongly and dose-dependently inhibited by all glycan cap-specific mAbs (Fig 3D). In contrast, stalk-specific mAbs BDBV259 and BDBV317 increased the viral load in the supernatants when provided in low doses. Only high doses of these mAbs (100 μg/ml) strongly reduced release of viral particles, which did not reach, however, the level of inhibition seen for the glycan cap-specific mAbs. Taken together, these data suggest that retention of produced virions on the cells is a common mechanism of antibodies targeting external, well-exposed domains of the viral envelope proteins involved in an interaction with a cell surface. Strikingly, the single tested non-neutralizing BDBV52 mAb abolished release of virus, even at concentrations as low as 1 μg/ml. The same inhibition level was demonstrated by the highly neutralizing BDBV223 mAb, highlighting the lack of correlation between suppression of virus egress and in vitro neutralization.
To ensure that viral RNA detected in the cell supernatants resulted from a bona fide viral egress process, but not from the exit of RNA from cells via the exosome pathway, we conducted an additional experiment with depletion of exosomes in supernatant samples (S12 Fig). Indeed, for each of the tested mAb, regardless of its concentration, incubation of supernatants with exosome-binding beads did not result in a significant change of the viral RNA level (p > 0.05, paired Student’s t-test).
Next, we tested if mAbs can suppress virus replication when added at different time points after infection (Figs 3E, S4A and S13). Administration of BDBV270, BDBV289, BDBV223 or BDBV317, which belong to different epitope recognition groups, 3 hours prior to or during cell inoculation completely blocked replication of virus. Infection inhibition by BDBV41 and BDBV259 was less prominent. In contrast, the two MPER-specific mAbs BDBV317 and BDBV223 caused the strongest reduction of infected cell numbers when added up to 3 hours after inoculation. When added at 24 hours post-inoculation, none of the mAbs tested could prevent virus replication by more than 20%. Addition of BDBV52 at any time point did not change percentages of eGFP+ cells (Fig 3E), despite the fact that this mAb efficiently inhibited virus egress from infected cells (Fig 3D). This finding can be explained by direct virus dissemination between cells skipping the step of virion release into the extracellular space, since BDBV52 did not block cell-to-cell transmission of the virus (Fig 3C). The results indicate that MPER-specific antibodies are important for control viral replication, as they effectively prevent virus replication when added late.
Glycosylation is one of the most common posttranslational modifications of viral surface proteins. Glycosylation of viral proteins may target key epitopes at the surface of virions, masking them from antibody recognition. A decade ago, an unusual post-translational modification, C-mannosylation, was found at the residue Trp288 of EBOV sGP, which was the first demonstration of this type of glycosylation in a viral protein [38]. Since then, no evidence of any biological significance of sGP C-mannosylation has been reported. BDBV GP possesses the same W288AFW291 motif in the glycan cap, which is considered to be the most immunogenic region of filoviral GP [3, 17]. We therefore hypothesized that C-mannosylation of BDBV GP can impact antibody virus neutralization (Fig 4). To test this hypothesis, we disabled the mannosylation site in EBOV/BDBV-GP by introduction of the mutation W291A (Fig 1). To avoid interference of sGP with neutralization of viruses and to assess the pure effect of spike GP C-mannosylation on their resistance to mAbs, we also disabled the expression of sGP by stabilization of GP gene editing site [28]. BDBV270 mAb demonstrated a striking difference between neutralization of W291- and A291-bearing (ΔC-mann) mutants. At a concentration 0.8 μg/ml, the levels of neutralization of ΔsGP and ΔsGP/W288 ΔC-mann viruses by this mAb were 7.2% and 72.5%, respectively. Neutralization of the two viruses by MPER-specific BDBV223 (Fig 4) or by other mAbs included in this study performed for comparison did not show any difference. These results demonstrated for the first time shielding of a viral epitope by C-mannosylation.
Next, we sought to compare protection by glycan cap- and MPER-specific antibodies. Since BDBV52, BDBV41 and BDBV259 mAbs are BDBV species-specific and do not bind EBOV GP [20], and no mouse-adapted BDBV exists so far, they were not included in animal studies. We have shown previously that BDBV289, BDBV223 and BDBV317 mAbs protect mice when given as a single 100 μg dose the day after challenge with 1,000 PFU of mouse-adapted EBOV delivered by the intraperitoneal route [20, 25]. Here, we extended the study by testing BDBV270 (S14 Fig). The in vivo activity of the antibody was similar to that of another glycan cap-specific mAb BDBV289 with 80% protection (4 out of 5 mice), with similar dynamics of changes in weight and disease score, although the difference in animal survival between BDBV270 and 2D22 mAb groups did not reach statistical significance (p = 0.0644, Mantel-Cox test). Thus, the two MPER-specific mAbs demonstrated complete protection and two glycan cap-specific mAbs demonstrated a high but not absolute protection in our present and previously reported studies [20, 25] with the selected dose and regiment of treatment.
Besides direct blocking of viral entry and/or exit through interaction with virions via Fab domains, mAbs also provide a second level of defense by cross-linking the viral proteins exposed on the surface of infected cells and Fc receptors on multiple immune cells to activate ADCC or ADCP mechanisms. Natural killer (NK) cells play a pivotal role in elimination of infected cells by ADCC. Activation of FcγRIIIa on NK cells causes the release of cytotoxic granules, which causes apoptotic death of target cells, as well as secretion of cytokines (IFNγ and TNFα) and chemokines (MIP-1α and MIP-1β), which correlate with their activation. Phagocytosis through engulfment of infected cells represents another important mechanism of rapid clearance of infection, which is mediated by FcγR-bearing immune cells including monocytes, macrophages, dendritic cells, neutrophils, and mast cells known as professional phagocytes [39].
Since the expression of CD107a correlates with cytokine production and cytotoxicity and it is used as a marker of NK cell degranulation [40], we used it as a marker of NK cell activation. In our experiments, the only two mAbs of IgG3 subclass, BDBV259 and BDBV223, both are stalk-specific, induced a high level of surface expression of CD107a and intracellular production of IFNγ and MIP-1β in NK cells directed against BDBV GP (Figs 5A–5C, S15A and S16A) that is consistent with the higher affinity of IgG3, compared to IgG1, for binding to FcγRs [41]. BDBV259 and BDBV223 also induced ADCP of GP-covered beads by THP-1 monocytes and neutrophils (Figs 5D, 5E, S15B, S16B and S16C). Interestingly, however, another stalk-specific mAb, BDBV317, belonging to the IgG1 subclass, showed only a slight increase in NK cell activation compared to glycan cap-specific mAbs, yet induced neutrophil phagocytosis similarly to BDBV259 and BDBV223. As interaction with FcRs can be modulated by both IgG subclass and Fc glycans structures, analysis of the glycans on the Fc domain was performed for each mAb (Fig 5F–5I). Interestingly, the stalk-specific IgG3 mAbs, BDBV259 and BDBV223, and the stalk-specific IgG1, BDBV317, were all characterized by higher sialylation of the Fc domain. As increased sialyation has been typically associated with anti-inflammatory activity [42, 43], the IgG3 subclass of BDBV259 and BDBV223 may underlie the enhanced functional activity associated with these mAbs. However, the IgG1 BDBV317 mAb was characterized by increased levels of galactose and bisecting N-acetylglucosamine (GlcNAc) glycan structures, and elevated levels of bisecting GlcNAc has been previously associated with greater phagocytic activity [44] and enhanced interaction with FcγRIIIa and ADCC activity [41]. The level of fucosylation, which negatively impacts binding of all IgG subclasses to FcγRIIIa and induction of ADCC [42], was equally high for all tested mAbs. Altogether, these data suggest that while IgG3 induced the highest level of Fc-mediated effects, the epitope location also contributed to some of the Fc-mediated effects, consistent with previously published studies with influenza virus [45–50].
Finally, we selected BDBV223 mAb, which has the broadest spectrum of inhibitory activities against different steps of viral infection in vitro, to address the physiological relevance of the observed Fc-mediated effects for MPER mAbs (Fig 5). We introduced the L234A/L235A (LALA) mutation, which impairs binding of antibodies to FcγRs [51–54], into the Fc region of the antibody, and compared the efficacy of mutated and non-mutated recombinant mAbs in a mouse model of EBOV infection (Fig 6). Human IgG1 and IgG3 have been shown previously to interact with mouse FcγRs [55]. Human IgG1 induces mouse innate immune effector functions at the levels equivalent to that induced by the most functional mouse subclass, IgG2a, while human IgG3 shows reduced activity with murine cells compared to human IgG1 [55]. Thus, it is possible that the human IgG3 mAbs cannot fully leverage the mouse innate immune system to maximize in vivo protective efficacy. We therefore generated the recombinant BDBV223 mAbs of IgG1 subclass, although the original BDBV223 subclass is IgG3 (Fig 1). Groups of BALB/c mice (5 animals per group) were inoculated with 1,000 PFU of mouse-adapted EBOV, strain Mayinga, and 24 hours later treated by the intraperitoneal route with 40 or 100 μg of wild-type rBDBV223-IgG1 or rBDBV223-IgG1-LALA. At both doses tested, wild-type antibody, but not the LALA mutant, provided complete protection of mice from the lethal EBOV infection. The differences between survival of animals in rBDBV223-IgG1 and rBDBV223-IgG1-LALA groups were statistically significant: 40 μg, p = 0.0158, 100 μg, p = 0.0494 (Mantel-Cox test). These data suggest that Fc-FcγR interactions can play a critical role in protection against EBOV infection mediated by MPER mAbs in vivo.
The unprecedented epidemic of EBOV in West Africa in 2013–2016 demonstrated the urgent need for treatments against this and related highly pathogenic filoviruses. Antibody-based therapy remains the only available effective strategy against the infection. Further progress in development of more broad and effective filovirus mAbs requires identification of the mechanism of the protective effect of these mAbs.
The glycan cap and MLD are excised by cathepsins during endosomal GP processing and, therefore they are dispensable for virus entry into the cytoplasm. It has been proposed that antibodies targeting these domains of GP are generally non-neutralizing, with some of them being able to confer protection likely through Fc-mediated mechanisms, such as ADCC or ADCP of infected cells [39]. In contrast, antibodies targeting the GP base could prevent membrane fusion [56] by blocking GP cleavage [57] or fusion-triggering conformational changes in proteolytic primed GP bound to NPC1 [58], and therefore are mostly neutralizing. However, we isolated glycan cap-specific mAbs from the blood of survivors of natural ebolavirus infection that protect mice and guinea pigs from lethal EBOV challenge [20]. Murine m8C4 mAb targeting the glycan cap was reported to neutralize EBOV and SUDV and confer partial protection of mice against these viruses; induction of ADCP by neutrophils, monocytes and dendritic cells was proposed as one of the mechanisms of protection [44]. The discovery of novel antibody epitopes in RBD [44, 59], glycan cap/RBD interface [60], IFL [19, 59, 61], and epitopes proximal to the viral membrane [19, 20] have substantially extended the concept of vulnerability sites on EBOV GP. Murine 6D6 cross-neutralizing mAb targeting the tip of the IFL prevented GP-mediated membrane fusion and protected mice against EBOV and SUDV [61]. Inhibition of cathepsin-cleaved EBOV GP binding to its endosomal receptor NPC1 was demonstrated to be the major mechanism of protection by human antibody mAb114 [57] and macaque-derived FVM04 mAb [21]. MAb114 recognizes an epitope spanning both the glycan cap and RBD, while FVM04 binds to the tip of the RBD crest. Interestingly, although antibody access to RBD is considered to be largely restricted by the surrounding glycan cap and MLD domains [5], the epitope of FVM04 is exposed in the full-sized GP. Thus, prevention of endosomal membrane fusion remains the only demonstrated mechanism of EBOV neutralization by RBD-, IFL- and GP base-specific antibodies, whereas antiviral mechanisms employed by antibodies targeting the glycan cap and novel epitopes proximal to the viral membrane are not clear.
Here, we investigated antiviral mechanisms for a diverse panel of human antibodies isolated from several human survivors of natural ebolavirus infections. Generation of escape mutant viruses resulted in mutations in the glycan cap of GP1 or in the IFL/stalk region of the GP2 subunit [26]. Glycan cap represents a well-exposed portion of the GP trimer in its native conformation, and therefore is a common target of the antibody response [5, 22], while the GP areas proximal to the viral membrane are less accessible, and have been only recently identified as a novel group of mAb epitopes [20, 62].
Since filoviruses attach to the cell surface through low-affinity interactions with multiple types of molecules, none of the filovirus-specific mAbs, including those described in the present study, were shown to completely inhibit cell attachment and infection. Moreover, all of the neutralizing mAbs studied here showed dose-dependent inhibition of viral replication when added after virus attachment to cells, suggesting they inhibit intracellular steps of entry. Interestingly, the non-neutralizing BDBV52 mAb caused an enhanced viral attachment and entry into cells, which perhaps can be mediated by the re-uptake of de novo synthesized viral particles retained at the cell surface by BDBV52 at the budding step.
We next analyzed mAb effects on VLP trafficking through the endosomal network. The tested mAbs did not prevent cathepsin cleavage of GP (Fig 3B) and had no effect on GP/NPC1 interaction (S8 Fig). Therefore, the co-localization of viral particles with LAMP-1 and Rab7 endosomal markers observed in the presence of stalk-binding mAbs is likely a consequence of events accompanying the merge of viral and endosomal membranes, such as conformational rearrangements of GP2 subunit after interaction of cleaved GP with NPC1.
Other than blocking of virus entry, mechanisms of restriction of infection can include inhibition of cell-to-cell transmission or budding of nascent virions from infected cells. Both steps of virus infection were found to be inhibited by glycan cap and MPER mAbs in our study. However, these mechanisms are not mutually exclusive, and, moreover, could be mediated at least in part by direct virus neutralization. The latter mechanism seems to pertain for the most potent neutralizer, BDBV223, which completely blocked virus transmission and egress, presumably by trapping it inside LAMP-1+ vesicles during cell entry. Unexpectedly, a comparable effect on egress inhibition was demonstrated for the non-neutralizing mAb BDBV52, with no impact on virus transmission to neighboring cells observed. Overall, mAbs demonstrated differing patterns of cell-to-cell transmission and virus egress inhibition, which could not be explained by simple differences in their neutralization activity, and is probably determined by a combination of factors, such as the location of epitope and affinity to GP at differing pH conditions.
The addition of N-linked glycans to envelope proteins is a commonly used strategy of immune evasion employed by HIV, influenza, Nipah and other viruses, which, at the same time, does not interfere with their attachment to the cell surface [63]. We found that C-mannosylation can also make virus less sensitive to a glycan cap-specific antibody. The C-mannosylation motif is located in the region shared by GP and sGP and is conserved in all known ebolavirus species: EBOV, SUDV, TAFV, BDBV and RESTV. Despite the fact that this modification was found in EBOV sGP protein [38], the results of comparative neutralization of viruses with intact or disrupted C-mannosylation site and the lack of sGP produced by the viruses used in the assay suggests that envelope GPs of BDBV, and likely of all other ebolaviruses, are also subjected to C-mannosylation. The neutralization kinetics showed that the mannose residue on W288 is likely to restrict epitope access for at least some of the glycan cap-specific mAbs.
Antibodies mediate antiviral effects both by binding epitopes on targeted pathogens by Fv region interactions and by activating Fc receptor-bearing effector cells, such as NK cells, neutrophils, macrophages and dendritic cells by Fc domain interactions. The spectrum of Fc-mediated effects induced by an antibody depends on its affinity for binding to particular FcγRs, which, in turn, depends on the IgG subclass and Fc region glycosylation. The conformational nature of the epitope recognized also impacts the efficiency of immune cell engagement. The disruption of Fc-FcγR linkage through either introduction of a D265A mutation in the Fc region or using knockout mice with disabled FcγRs leads to a complete loss of in vivo protection from influenza virus by broadly neutralizing HA stalk-targeting mAbs, but not by strain-specific mAbs binding to HA head domain [45, 64]. From this insight, it was interesting to observe a substantial activation of NK cells and induction of monocyte- and neutrophil-mediated phagocytosis by stalk-specific mAbs BDBV259 and BDBV223 in our study compared to the glycan cap-specific mAbs. While the observed increased activation may be due to their IgG3 subclass, which have higher affinity for FcγRIIIa and FcγRIIa compared to IgG1 antibodies [65], the stalk-specific BDBV317 IgG1 mAb also induced greater ADCP activity by neutrophils and stimulation of NK cells compared to the glycan cap-specific mAbs tested here, which also belong to the IgG1 isotype. Therefore, it is of interest to test if the direct contact between antibody-bound filovirus GP and the effector cell is required for optimal triggering of Fc mechanisms.
The biological effects of mAbs demonstrated in this study are summarized in Fig 7. In general, stalk-specific mAbs have greater Fab- and Fc-mediated effects, with the noticeable exception of the inhibition of viral egress, which was highly pronounced for all glycan cap-specific mAbs tested, and the greater level of protection in vivo. The current approach for treatment of filovirus infections with antibody cocktails demonstrated in animal models uses the principle of targeting of non-overlapping epitopes [20, 44, 59, 60, 66–68]; for example, our recent study demonstrated synergistic effects of the MPER-specific mAb BDBV223 and the glycan cap-specific mAb BDBV289 [20]. The data presented here suggest that there may be cooperative or synergistic effects of antibodies that block varying steps of viral replication, and cocktails based on combining such effects also should be tested. As the two contrast groups of mAbs tested in this study have different biological effects (Fig 7), the beneficial effects of cocktails of non-overlapping epitopes may be related not only to targeting different epitopes, but also to the ability of these antibodies to inhibit different steps of viral replication.
Wild-type BDBV, strain 200706291 Uganda, which was originally isolated from the serum of a patient during the first known outbreak [69] was passaged three times in Vero-E6 cells. The EBOV/BDBV-GP virus enveloped with glycoprotein of Bundibugyo strain, and EBOV/BDBV-GPΔsGP virus lacking sGP production were generated as described earlier [28]. To generate an EBOV/BDBV-GP derivative not expressing eGFP, the full-length clone was digested with BsiWI restriction endonuclease to remove eGFP gene, and then re-ligated. The resulting plasmid was transfected into 293T cell monolayers to rescue EBOV/BDBV-GP_no eGFP virus. To obtain EBOV/BDBV-GP ΔsGP derivative with disabled C-mannosylation site, we subjected pEBOwtΔBamHI-SbfI,AscI-PspOMI subclone with the ORF for the GP of BDBV with stabilized RNA editing site to PCR mutagenesis using the QuikChange site-directed mutagenesis kit (Stratagene). Amino acid substitution W291A in BDBV GP was introduced into the construct to disrupt C-mannosylation of W288 residue in W288AFW291 motif. For generation of full-length construct, ApaI-SacI restriction endonuclease fragment from the resulting subclone was used to replace those in pEBO-eGFP plasmid. The obtained construct was transfected into 293T cell monolayers to rescue chimeric virus with disrupted C-mannosylation site - EBOV/BDBV-GP ΔsGP/W288 ΔC-mann. Neutralization of viruses by mAbs was tested in high-throughput screening assay based on the detection of residual eGFP fluorescence [28].
To generate VLPs enveloped with BDBV GP, glycoprotein ORF in pWRG7077:64755-2010-233-1_GP_optGP was substituted with that of BDBV. First, BamHI restriction endonuclease sites were disabled in pEBOwtΔBamHI-SbfI,AscI-PspOMI subclone with the ORF for the GP of BDBV with stabilized RNA editing site (ΔsGP) by introduction of silent mutations using the QuikChange site-directed mutagenesis kit (Stratagene, La Jolla, CA). Then, BDBV GP ORF was amplified from the resulting construct with following primers: direct, AGTCACGTGCGGCCGCCACCATGGTTACATCAGGAATTCT; and reverse, AGTCACGTGGATCCTTATCATCAGAGTAGAAATTTGCAAA (the NotI or BamHI restriction endonuclease sites are underlined, and the start of the BDBV GP ORF direct sequence and the end of the BDBV GP ORF complementary sequence are italicized). The obtained PCR product was used to replace EBOV GP ORF in EBOV VLP GP plasmid by NotI and BamHI sites to get the final GP-bearing plasmid for BDBV VLP production. EBOV NP and codon optimized VP40 were cloned into the pCEZ vector [70]. pCEZ-NP was a kind gift from Drs. Kawaoka and Feldmann. The plasmids were transfected to 293T cells using TransIT-LT1 transfection reagent (Mirus). VLPs were harvested after 72 hours of the transfection, purified by sucrose gradient and quantified using ViroCyt Virus Counter (VC) 2100 (ViroCyt).
For confocal microscopy, Vero-E6 cell cultures (American Type Culture Collection) were grown in monolayers in chambered slides. BDBV VLPs were incubated in the presence of mAbs (200 μg/ml) for 1 hour at room temperature, added to Vero-E6 cell culture monolayers at a ratio of 500 VLP/cell, and cells were placed on ice for 1 hour. Then, cells were fixed in formalin (ThermoFisher Scientific) for 15 min, permeabilized with 0.5% Triton X-100 in phosphate buffered saline (PBS) for 15 min to increase sensitivity of the subsequent immunostaining of viral proteins. Cells then were blocked with 5% donkey serum diluted in PBS with 1% BSA and 0.1% Triton X-100 (PBS-T-BSA) for 1 hour. Next, VLPs were stained using rabbit immune serum raised against EBOV VLPs (IBT Bioservices) supplemented with rabbit polyclonal antibodies specific for BDBV GP (IBT Bioservices; all antibodies for virus staining were diluted at 1:100 in PBS-T-BSA). The slides then were incubated with donkey anti-rabbit antibodies conjugated with AlexaFluor 647 (ThermoFisher Scientific) for 1 hour at room temperature. Next, the slides were washed 3 times in PBS with 0.1% Triton X-100 (PBS-T), fixed in 10% formalin and incubated with 4',6-diamidino-2-phenylindole dihydrochloride (DAPI) (Invitrogen) at 1 μg/ml for 2 min. Then, slides were washed 5 times in PBS and mounted onto coverslips using PermaFluor mounting medium (ThermoFisher Scientific). The slides were analyzed by laser scanning confocal microscopy using an Olympus FV1000 confocal microscope housed in the Galveston National Laboratory. Lasers with 405 nm wavelength were used for DAPI excitation, and 635 nm for Alexa Fluor 647. All images were acquired using a 60x oil objective. For quantification, five representative randomly selected images were acquired and the AlexaFluor 647 fluorescence was analyzed using the FV1000 software image measurement tool. Statistical analysis was performed using ANOVA with Tukey post hoc test.
For flow cytometric analysis of virus binding, Vero-E6 cells were plated in U-bottom 96-well plates (ThermoFisher Scientific) at 106 cells per well and placed on ice. EBOV/BDBV-GP_no eGFP was incubated with mAbs (200 μg/ml) at 37ºC for 1 hour followed by 15 min on ice and used to inoculate cells at an MOI of 5 PFU/cell. Cells were incubated for 2 hours on ice and washed with 2% fetal bovine serum (FBS) in PBS. Thereafter, cells were immunostained with rabbit immune serum against EBOV VLPs (IBT Bioservices) supplemented with anti-BDBV GP rabbit polyclonal antibody (IBT Bioservices); both the immune sera and antibody were added at 1:100 dilution in PBS with 2% FBS and incubated for 30 min at room temperature. After staining, cells were washed three times with 2% FBS in PBS, fixed in 10% formalin for 15 min, stained with donkey anti-rabbit antibodies labeled with Alexa Fluor 647 (ThermoFisher Scientific) and washed again 3 times with 2% FBS in PBS. Flow cytometry was performed using an LSRII Fortessa cytometer (BD Biosciences). For each sample, 10,000 events were acquired.
BDBV was adsorbed on Vero-E6 monolayer cell cultures in 24-well plates at an MOI of 0.1 PFU/cell for 20 min at 4ºC. Cells were washed 3 times with cold PBS, incubated with four-fold serial dilutions of mAbs for 20 min at 4ºC, washed again and covered with a 0.45% methylcellulose overlay in minimal essential medium (MEM) with 2% fetal bovine serum. Cells were incubated for 6 days at 37ºC, and plaques were visualized by immunostaining with BDBV52 mAb [20] followed by secondary goat anti-human IgG conjugated with horseradish peroxidase and 4CN two-component peroxidase substrate system (KPL). Post-attachment inhibition was calculated as a percent reduction of numbers of viral plaques developed after incubation with antibody compared to no mAb control, as previously described [71, 72]. For the no-mAb control samples, the average number of plaques per well was 263.
Three million PFU of eGFP-expressing EBOV/BDBV-GP were incubated with various mAbs at the final concentration 100 μg/ml for 1 hour at 37ºC and then adsorbed on Vero-E6 cell culture monolayers for 40 min at 4ºC. Cells were washed 3 times with MEM containing 10% FBS and incubated in fresh medium for 24 hours. Then, cells were treated with trypsin, harvested, washed twice with PBS and fixed with 4% paraformaldehyde for 24 hours for virus inactivation. Cells were analyzed by flow cytometry using an Accuri C6 cytometer (BD Biosciences) to determine the percentages of infected eGFP+ cells and their mean fluorescence intensity (MFI). On average, 7,728 events were acquired per sample.
BDBV VLPs were generated as described above. EBOV VLPs were purchased from IBT Bioservices. BDBV or EBOV VLPs were incubated with 200 μg/ml of mAbs for 60 min at 37ºC. Monolayers of Vero-E6 cells were inoculated with VLP/mAb complexes, incubated for 30 or 60 min and fixed with 4% paraformaldehyde for 15 min. Monolayers were washed and permeabilized with 0.5% Triton-X100 solution in PBS for 15 min. Monolayers were blocked with 5% donkey serum diluted in PBS-T-BSA for 30 min. Cell monolayers were stained with mouse mAb specific for lysosomal marker LAMP-1 (Santa Cruz) at a 1:50 dilution and goat polyclonal antibodies specific for late endosome marker Rab7 (Santa Cruz) at a 1:50 dilution. VLPs were stained with rabbit immune serum against EBOV VLPs or the same rabbit immune serum supplemented with rabbit anti-BDBV GP polyclonal antibody (IBT Bioservices) at a 1:100 dilution for each antibody. Slides were incubated for 1 hour at 37ºC, washed 3 times as above, and incubated with a mixture of three secondary antibodies, each at 1:200 dilution in PBS-T-BSA: donkey anti-mouse conjugated with Alexa Fluor 488, donkey anti-goat conjugated with Alexa Fluor 594 and donkey anti-rabbit conjugated with AlexaFluor 647 (ThermoFisher Scientific). Next, cells were washed 3 times in PBS-T, and nuclei were stained with DAPI, as described above. Slides were analyzed by laser scanning confocal microscopy using an Olympus FV1000 confocal microscope with 405 nm wavelength laser for DAPI excitation, 488 nm for Alexa Fluor 488, 543 nm for Alexa Fluor 594, and 635 nm for Alexa Fluor 647.
VSV/BDBV-GP was propagated in Vero-E6 cells; at 48 hours after inoculation, the virus suspension was harvested and clarified from cell debris by low-speed centrifugation. To purify the virus, supernatants were placed atop a 25% sucrose cushion and pelleted in an ultracentrifuge for 2 hours at 175,000 x g, 4ºC. Pellets were resuspended in 1x STE buffer (10 mM Tris, 1 mM EDTA, 0.1 M NaCl) and further purified by ultracentrifugation in 20–60% sucrose gradient (1.5 hours at 288,000 x g, 4ºC). The virus-containing band was harvested, and VSV/BDBV-GP virions were washed from sucrose by final ultracentrifugation in 1x STE buffer (1 hour, 4ºC, 175,000 x g). The obtained viral particles were resuspended in 1x STE buffer. Flat-bottom high-binding 96-well microplates (Greiner Bio-One) were coated overnight with purified VSV/BDBV-GP particles diluted in PBS. Bound antigen was blocked with 1% bovine serum albumin (Sigma-Aldrich) in PBST buffer (0.1% Tween-20 in PBS), and treated for 20 min with 20 mM sodium citrate, pH 5.0 (Sigma-Aldrich), or PBS for 20 min. MAbs were added at 1 μg/ml in 0.1% Tween-20 containing 20 mM sodium citrate, pH 5.0, or PBST buffer, respectively, and incubated for 1 hour at 37ºC. Plates were washed three times in PBST buffer, secondary goat anti-human IgG conjugated with horseradish peroxidase (KPL) were added at a 1:2,000 dilution in PBST buffer, and plates were incubated for 1 hour at 37ºC. Next, plates were washed three times in PBST buffer, 1-component SureBlue Reserve TMB Microwell Peroxidase Substrate (KPL) was added, and plates were incubated for 20 min at room temperature and scanned in a Synergy microplate reader (BioTek) at the emission wavelength 630 nm.
VSV/BDBV-GP purified as described above was resuspended in thermolysin digestion buffer (50 mM Tris, pH 8.0, 0.5 mM CaCl2) and divided into two aliquots; one aliquot was treated with 0.5 mg/ml of thermolysin (Promega) and another one with an equal volume of thermolysin digestion buffer (mock-treated virus) for 40 min at 37ºC. The reactions were stopped by addition of EDTA up to the final concentration 10 mM. Virus samples were re-pelleted through a 25% sucrose cushion as described above, and washed by ultracentrifugation in 10 mM Tris, 0.1 M NaCl for 1 hour at 175,000 x g, 4ºC. The resulting preparations were resuspended in 10 mM Tris, 0.1 M NaCl, incubated with 100 μg/ml mAbs for 1 hour at 37ºC, or mock-incubated, and titrated on triplicate Vero-E6 cell culture monolayers using plaque reduction assay.
Aliquots of thermolysin-treated or mock-treated purified virions were heated for 10 min at 95ºC and separated in Nu-PAGE 4 to 12% Bis-Tris gel with Novex Sharp Pre-Stained Protein Standard used as a molecular weight marker. Proteins were transferred to a nitrocellulose membrane using the iBlot Gel transfer system (Life Technologies). The membrane was incubated with primary rabbit polyclonal antibodies against BDBV GP (1:500; IBT Bioservices) and secondary goat anti-rabbit IgG antibodies conjugated with horseradish peroxidase (1:500; KPL). Protein bands were visualized using the chromogenic 4CN two-component peroxidase substrate system (KPL).
EBOV VLPs alone or in the presence of 200 μg/ml of mAbs were incubated in sodium acetate buffer, pH 5.0, with 0.1 μg/μl of cathepsin B and cathepsin L at 37ºC overnight. Thereafter, VLPs were denatured in Laemmli buffer (Novex) in reducing conditions, and GP cleavage was confirmed by immunoblotting with a pan-filovirus GP-specific monoclonal antibody (IBT Bioservices). Densitometry was performed using ImageJ gel analyzer plug-in. For normalization, we used VP40 as a housekeeping protein and the 20 kDa band of GP as the target protein.
THP-1 monocytic cells (American Type Culture Collection) were inoculated with EBOV/BDBV-GP virus expressing eGFP at MOI of 2 PFU/cell, incubated for 48 hours, washed two times to remove unbound virus, and incubated with 100 μg/ml of mAbs or no mAb. Following a one hour-long incubation, cells were placed atop of monolayers of Vero-E6 cells pre-stained with CellTrace Far Red (ThermoFisher Scientific) according to the manufacturer’s recommendations, incubated for 72 hours and fixed with 4% paraformaldehyde. Cells were analyzed by flow cytometry to determine the percentages of cells double-positive for CellTrace Far Red and eGFP of total cells positive for CellTrace Far Red. The percentage of double-positive cells indicated the percentage of cells that became infected due to cell-to-cell transmission of virus. For each sample, 30,000 events were counted. In a separate experiment, supernatant aliquots were harvested from co-cultures of THP-1 and Vero-E6 cells on days 3–5 after the inoculation of monocytes and then titrated on Vero-E6 cell monolayers.
Vero-E6 cell culture monolayers were inoculated with EBOV/BDBV-GP expressing eGFP at an MOI of 0.1 PFU/cell, incubated for 1 hr, washed 3 times to remove non-attached viral particles, and covered with medium containing 1, 10 or 100 μg/ml of mAbs or no mAb. Cells were incubated for 48 hours, supernatants were collected, and RNA was isolated. Viral genomes were quantitated by one-step reverse transcription droplet digital RT-PCR (Bio-Rad) according the manufacturer’s instructions. Sequences of primers are available upon request. In a separate experiment, cell supernatants were incubated with exosome removal beads (Exosome-Human CD63 Isolation/Detection Reagent, ThermoFisher Scientific) for 30 min at ambient temperature, or mock-incubated, centrifuged for 5 min at low speed for sedimentation of beads, transferred to the clean tubes and subjected to RNA isolation and droplet digital RT-PCR analysis.
Vero-E6 cell culture monolayers in 24-well plates were inoculated with EBOV/BDBV-GP expressing eGFP at an MOI of 0.1 PFU/cell, with mAbs added at final concentration 100 μg/ml 3 hours prior to, at the moment of infection, or 3 or 24 hours after virus inoculation. Forty-eight hours after inoculation, cells were washed twice with PBS, treated with trypsin, harvested, fixed with 4% paraformaldehyde, and infected (eGFP+) cells were quantified by flow cytometry. For each sample, 10,000 events were counted.
Seven-week-old BALB/c mice (Charles River Laboratories) were placed in the ABSL-4 facility of the Galveston National Laboratory. Groups of mice at five animals per group were injected intraperitoneally with 1,000 PFU of the mouse-adapted EBOV. Twenty-four hours later, animals were injected with mAbs at indicated amounts by the intraperitoneal route. Animals treated with the 2D22 mAb specific for dengue virus served as controls. The recombinant versions of BDBV223 mAb with or without LALA mutation in the Fc fragment (rBDBV223-IgG1-LALA and rBDBV223-IgG1, respectively) were generated as described elsewhere [25, 26, 73]. The animal observation procedure was performed as previously described [20]. The extent of illness was scored using the following parameters: dyspnea (possible scores 0–5), recumbence (0–5), unresponsiveness (0–5), and bleeding/hemorrhage (0–5). Moribund mice were euthanized as per the protocol approved by the UTMB Institutional Animal Care and Use Committee. The humane endpoint for weight loss was 20%. The overall observation period lasted for 28 days.
The NPC1-encoding plasmid was purchased from OriGene. NPC1 was amplified by PCR and cloned into the p3xFLAG-CMV9 plasmid (Sigma-Aldrich). To add the red fluorescent protein (RFP) at the N-terminus, RFP gene cDNA was PCR-amplified from pcDNA3-mRFP (Addgene) and added upstream of the NPC1 coding sequence using NotI and BamHI restriction endonuclease sites. Vero-E6 cell culture monolayers were electroporated with a P3X_NPC1-RFP using Neon transfection system (ThermoFisher Scientific) with 2 pulses of 20 msec at 1,150 V, placed in chambered slides (Nalge Nunc International) and incubated overnight at 37ºC. EBOV/BDBV-GP_no eGFP was incubated with 200 μg/ml of mAbs for 1 hour at 37ºC and used for inoculation of transfected cells at an MOI of 10 PFU/cell for 30 min. Thereafter, cells were fixed with 4% paraformaldehyde for 15 min. Cell monolayers were washed 3 times in PBS-T, and viruses were incubated with rabbit immune serum against EBOV VLPs (IBT Bioservices) supplemented with rabbit anti-BDBV GP polyclonal antibody (IBT Bioservices) at a 1:100 dilution for both antibodies for 1 hour. Next, cells were washed 3 times with PBS-T and incubated with donkey anti-rabbit antibody conjugated with Alexa Fluor 647 (ThermoFisher Scientific) diluted 1:200 in PBS-T-BSA for 30 min. Next, the slides were washed 3 times in PBS-T, fixed in 10% formalin for 72 hours and removed from the BSL-4. The slides were washed 3 times in PBS and mounted onto coverslips using PermaFluor mounting medium (ThermoFisher Scientific). FRET analysis was performed by scanning confocal microscopy using an Olympus FV1000 confocal microscope with the 543 nm laser for excitation and a far-red emission filter for detection. FRET efficiency (E) was calculated using Olympus FV1000 software. The effect of mAbs on NPC1-GP interaction was measured by changes of FRET efficiency when compared with the effect of virus inoculated in the absence of mAbs.
Human NK cells were enriched from peripheral blood by negative selection using RosetteSep negative selection kit (Stem Cell Technologies) followed by Ficoll separation. NK cells were rested overnight in the presence of 1 ng/ml recombinant IL-15 (PeproTech). 3 μg/ml of BDBV GP (IBT Bioservices) was coated on a Maxisorp ELISA plate (Nunc) at 4°C overnight, and plates were blocked with 5% BSA prior to addition of antibodies (5 μg/ml) in PBS for 2 hours at 37°C. The control EBOV-specific mAb c13C6 was purchased from IBT Bioservices. Unbound antibodies were removed by washing wells 3X with PBS prior to addition of NK cells. The NK cells were added at 5 x 104 cells/well in the presence of brefeldin A (Sigma Aldrich), GolgiStop (BD Biosciences), and anti-CD107a PE-Cy5 antibody (BD Biosciences clone H4A3) and incubated for 5 hours at 37°C. NK cells were stained with flow cytometry antibodies for the following surface markers: CD3 AlexaFluor700 (BD Biosciences clone UCHT1), CD56 Pe-Cy7 (BD Biosciences clone B159), and CD16 APC-Cy7 (BD Biosciences clone 3G8), followed by intracellular staining for IFNγ (FITC, BD Biosciences clone B27) and MIP-1β (PE, BD Biosciences clone D21-1351) to detect the production of cytokines and chemokines. Cells were analyzed by flow cytometry on a BD LSRII flow cytometer and data was analyzed using FlowJo software.
Recombinant BDBV GP was biotinylated and conjugated to streptavidin-coated Alexa488 beads (Life Technologies). BDBV-coated beads were incubated with antibodies at 5 μg/ml in culture medium for 2 hours at 37°C. Human THP-1 cells (ATCC) were added at a concentration of 2.5 x 104 cells/well and incubated for 18 hours at 37°C in 96-well plates. Cells were fixed with 4% paraformaldehyde and analyzed by flow cytometry on a BD LSRII using Diva software and FlowJo analysis software. The phagocytic score was determined using the following calculation: (% of AlexaFluor488+ cells)*(AlexaFluor488 geometric MFI of AlexaFluor488+ cells)/10,000.
Recombinant BDBV GP was biotinylated and conjugated to streptavidin-coated Alexa488 beads (Life Technologies). BDBV-coated beads were incubated with antibodies at 5 μg/ml in culture medium for 2 hours at 37°C. Human white blood cells were isolated from peripheral blood by lysis of red blood cells using ammonium chloride potassium lysis buffer. Cells were washed with PBS, and 5.0 x 104 cells/well were added to bead-antibody immune complexes, and then incubated for 1 hour at 37°C. Cells were stained with the following antibodies to identify neutrophils: CD66b Pacific Blue (BioLegend clone G10F5), CD14 APC-Cy7 (BD Biosciences clone MφP9) and CD3 AlexaFluor700 (BD Biosciences clone UCHT1). Cells were fixed with 4% paraformaldehyde and were analyzed on a BD LSRII flow cytometer. A phagocytic score was determined as described above.
20 μg of antibodies were digested with 120 U of IDEZ (NEB) for 1 hour at 37°C to separate the F(ab′)2 and Fc regions. The Fc region was purified by incubating digested antibodies with magnetic protein G beads (NEB) for an additional hour at room temperature. Beads were washed with 2X with distilled water. Beads were then incubated with PNGaseF (ThermoFisher Scientific) to remove the N-linked glycan at 50°C for 1 hour. Released glycans were purified and labeled using the GlycanAssure APTS labeling kit (ThermoFisher Scientific) according to manufacturer’s instructions. Labeled glycans were analyzed on a 3500xL Genetic Analyzer (Applied Biosystems) using a POP7 polymer. Glycan peaks and relative abundance of glycan content was analyzed using the GlycanAssure Data Analysis Software v1.0 (Applied Biosystems).
Factorial ANOVA and two-sided t-test were used for statistical analysis of in vitro data. Animal survival data were analyzed by log-rank (Mantel-Cox) test.
The animal protocol for testing of mAbs in mice was approved by the UTMB Institutional Animal Care and Use Committee (protocol №1307033) in compliance with the Animal Welfare Act and other applicable federal statutes and regulations relating to animals and experiments involving animals. Challenge studies were conducted under maximum containment in an animal biosafety level 4 (ABSL-4) facility of the Galveston National Laboratory.
|
10.1371/journal.pntd.0003693 | A DNA Vaccine against Yellow Fever Virus: Development and Evaluation | Attenuated yellow fever (YF) virus 17D/17DD vaccines are the only available protection from YF infection, which remains a significant source of morbidity and mortality in the tropical areas of the world. The attenuated YF virus vaccine, which is used worldwide, generates both long-lasting neutralizing antibodies and strong T-cell responses. However, on rare occasions, this vaccine has toxic side effects that can be fatal. This study presents the design of two non-viral DNA-based antigen formulations and the characterization of their expression and immunological properties. The two antigen formulations consist of DNA encoding the full-length envelope protein (p/YFE) or the full-length envelope protein fused to the lysosomal-associated membrane protein signal, LAMP-1 (pL/YFE), aimed at diverting antigen processing/presentation through the major histocompatibility complex II precursor compartments. The immune responses triggered by these formulations were evaluated in H2b and H2d backgrounds, corresponding to the C57Bl/6 and BALB/c mice strains, respectively. Both DNA constructs were able to induce very strong T-cell responses of similar magnitude against almost all epitopes that are also generated by the YF 17DD vaccine. The pL/YFE formulation performed best overall. In addition to the T-cell response, it was also able to stimulate high titers of anti-YF neutralizing antibodies comparable to the levels elicited by the 17DD vaccine. More importantly, the pL/YFE vaccine conferred 100% protection against the YF virus in intracerebrally challenged mice. These results indicate that pL/YFE DNA is an excellent vaccine candidate and should be considered for further developmental studies.
| DNA and other nucleic acid vaccine technologies are advancing quickly, and new potent delivery methods are demonstrating great potential in human clinical trials. In this manuscript, we report a highly protective DNA vaccine against the yellow fever virus. This vaccine was engineered with a molecular adjuvant technology to enhance the exposure of the vaccine antigens to the immune system, resulting in augmented CD4+ helper responses. We postulate that the robust CD4+ responses help the B cells and the CD8+ cells mature more efficiently and produce better antibodies and cytotoxic cells, respectively. Our results show that vaccination with this yellow fever DNA formulation elicited protective levels of neutralizing antibodies and very strong cellular responses at similar levels to the responses elicited by the live attenuated 17DD vaccine. In addition, these results also suggest a very important role for cellular responses in mediating protection against yellow fever virus. The results reported here are very promising and further studies may lead to a new yellow fever vaccine for human use.
| The yellow fever (YF) virus is considered the prototype member of the family Flaviviridae, which includes several other viruses of medical importance, such as the dengue, Japanese encephalitis, tick-borne encephalitis and West Nile viruses [1]. According to the World Health Organization (WHO), more than 200,000 cases of YF infection, including 30,000 deaths, occur annually, with 90% of cases occurring in Africa [2]. The safest strategy for preventing YF infection is still vaccination because there is currently no drug that is effective against YF virus infection. In the last 70 years, more than 500 million people around the world have been vaccinated with the YF 17D/17DD virus-attenuated vaccines with a remarkable record of safety and efficacy [3]. Attenuated YF virus vaccines generate both long-lasting neutralizing antibodies and T-cell responses [4, 5]. However, despite several improvements in the manufacturing process and quality control, severe side-effects resulting from vaccination continue to be reported [6–9]. In some cases, vaccination was associated with increased severity of symptoms [10] and on rare occasions with fatal reactions [11, 12]. In view of this, the development of alternative vaccination strategies, such as DNA-based vaccines encoding specific virus sequences, has been considered [13–16].
The YF virus genome consists of a single-stranded, positive-sense RNA molecule of ~10.8 kb, flanked by a 5’ cap and a 3’ non-polyadenylated terminal loop structure. It expresses three genes for structural proteins (capsid—C, pre-membrane/membrane—pM/M, and envelope—E) and seven genes that code for non-structural (NS) proteins (NS1, NS2a, NS2b, NS3, NS4a, NS4b, and NS5). Coexpression of flavivirus M and E genes in mammalian cells has been demonstrated to produce virus-like particles (VLPs) containing pM/M and E proteins [17–19]. The E protein is known to be the principal virus surface protein and the main target for neutralizing antibodies. pM/M and E coexpression, as a vaccination strategy, has been described as a way of triggering neutralizing antibodies against the Japanese encephalitis [19–21], West Nile [22] and dengue viruses [17,18,23,24].
DNA vaccines express endogenous cytoplasmic antigens, which are mostly introduced to the immune system through the major histocompatibility complex (MHC) class I molecules that are mostly associated with cellular cytotoxic responses and often fail to elicit a satisfactory humoral response, which is essential for efficient virus neutralization. Activation of CD4+ helper cells is important for the development of CD8+ responses, immunological memory [25], antibody maturation, class switching and expansion of antigen-specific B cells [26]. Several strategies have been proposed for enhancing MHC class II presentation of antigens encoded by DNA vaccines. The targeting of the MHC II compartment with other flavivirus E antigens has been shown to enhance neutralizing antibody production in immunized mice [17, 22, 24] and in non-human primates (Raviprakash personal communication at 2004 ASTMH meeting, http://www.astmh.org/meeting_archives.htm).
One of the main strategies for targeting the MHC II compartment with DNA-encoded antigens is based on the expression of the antigen fused to the lysosomal-associated membrane protein 1 (LAMP-1), a protein primarily found in the outer membrane of lysosomes [27]. The chimeric antigens expressed by DNA formulations in the context of type I trans-membrane LAMP are directed to compartments rich in MHC II, called the MHC II compartment (MIIC), which is where the peptide-MHC II complexes are formed [28, 29]. Other LAMP/antigen chimeric strategies, such as LAMP/HIV Gag [25, 26, 30, 31] and LAMP/dengue virus 2 pM/M-E [17, 24] antigens, have been shown to target the MIIC and were found to elicit enhanced immune responses compared with vaccines encoding unmodified native antigens.
This study investigated T-cell and humoral immune responses to the envelope of YF virus in C57Bl/6 and BALB/c mice immunized with DNA formulations expressing the full-length YF envelope protein, either as a wild-type or fused to LAMP. Responses in the mice were compared with the results obtained with standard immunization using the YF 17DD vaccine. We also evaluated the ability of DNA vaccines to provide protection against a lethal challenge. We show that although the YF 17DD vaccine produced higher neutralizing antibody titers, both DNA vaccine constructs encoding the entire E protein were also able to protect the mice against lethal challenge.
The attenuated 17DD human yellow fever vaccine was obtained from Bio-Manguinhos, a unit of the Oswaldo Cruz Foundation (FIOCRUZ, Rio de Janeiro, Brazil). The vaccine was reconstituted in chilled PBS, kept in an ice bath, and used for mouse immunizations within 4 hours of reconstitution. VERO and 293 cells were obtained from the ATCC (Rockville, MD, USA) and were grown according to the supplier’s instructions in a DMEM medium (Invitrogen) containing 10% fetal bovine serum (Gibco), 1% penicillin/streptomycin (Gibco) and 1% L-glutamine (Sigma). YF virus strain 17DD was propagated in Vero cells at 37°C in 5% CO2 to a titer of 106 plaque-forming units (PFUs) per ml. The polyclonal anti-YF hyperimmune serum used in immunofluorescence assays was obtained from mice immunized with the YF 17DD virus-attenuated vaccine in our laboratory. Secondary antibodies were purchased either from Jackson Immunoresearch Laboratories (Bar Harbor, ME, USA) or Molecular Probes (Seattle, WA, USA).
A set of 120 peptides of 15 amino acids each (15-mers), overlapping by 11 amino acids (15x11) and comprising the entire length of the envelope protein of the YF 17DD virus (NCBI GenBank accession number U17066), was synthesized using Schafer-N (Copenhagen, Denmark). The peptides were HPLC-purified to 80% purity or greater, with the exception of a few peptides that could not be purified and were used as crude extracts. The identity of each peptide was confirmed via mass spectrometry, and the amount of purified peptide was precisely measured. Stock solutions of all peptides were prepared via dilution in water when possible, or in a solution of 10 to 100% DMSO, to a final concentration of 20 mg/mL and were stored at −20°C. For the ELISPOT assays, the peptides were used at a 10 μg/mL final concentration. The highest DMSO concentration in the ELISPOT experiments was 0.05%.
The full YF genome, used as template to design primers for p/YFE and pL/YFE amplification, is deposited in NCBI’s GenBank under accession number NC 002031. The Kyte-Doolittle hydropathy plot analyzed this sequence to identify the capsid ER translocation signal and the predicted envelope trans-membrane domain of the YF genome. The wild-type pM/M-E amplicon starts with the ER capsid signal and ends with the envelope trans-membrane domain. To generate the pL/YFE construct, we designed a reverse primer that hybridizes to the YF genome just upstream of the envelope trans-membrane domain to replace it with the human membrane anchor and cytoplasmic domains of LAMP (Fig 1). The DNA pM/M-E sequence was amplified from the YF 17DD infectious clone using specific primers that incorporated an ATG start site in the context of the Kozak sequence and a translational stop codon. PCR amplification was performed using the proofreading TGO DNA polymerase (Roche, Indianapolis, IN, USA) and 0.6 μM of each primer. The amplicon was inserted into the p43.2 vector between the XhoI and NotI cleavage sites to generate the p/YFE construct. The pL/YFE construct, however, was obtained in two steps. First, the pM/M-E sequence was amplified using a reverse primer that hybridized upstream of the trans-membrane domain of the YF envelope protein. Then, the PCR product was inserted into the p43.2 vector between the NheI and XhoI sites, generating an intermediate construct (p/YFEINT), ready to receive the membrane anchor and cytoplasmic domains of LAMP. Second, LAMP was amplified from the p43.2-Gag/LAMP vector and was inserted into p/YFEINT, between the XhoI and XbaI sites, to generate the pL/YFE construct. Both the p/YFE and pL/YFE constructs were checked by sequencing; among the 2,061 nucleotides of the pM/M-E wild-type construction (p/YFE) that encodes 687 amino acids, two nonsynonymous mutations were found. An alanine (A) was replaced with a valine (V) at position 250, and a serine (S) was replaced with an aspartic acid (D) at position 349. Given that both mutations were also found at the same locations in the 644-residue sequence of the pL/YFE construction, they are very likely present in the YF 17DD infectious clone that was used as a DNA template. Regardless of the source of the two mutations, both mutations were deemed to be irrelevant for our vaccine studies as the E protein has several B-cell and T-cell preserved epitopes distributed along its sequence.
293 cells were plated onto cover slips and transfected with p/YFE, pL/YFE or empty p43.2 vectors, using Lipofectamine 2000 (Invitrogen Life Technologies). For Western blotting, transfections were carried out in 6-well tissue culture plates with 10 μg of each plasmid and 40 μl of Lipofectamine 2000, whereas transfections for fluorescence assays were carried out in 24-well tissue culture plates with 2.5 μg of each plasmid and 10 μl of Lipofectamine 2000, both in accordance with the manufacturer’s instructions. Vero cell extracts infected with the YF 17DD virus strain were used as a positive control. After 48 hours, both transfected and infected cell extracts were processed.
For Western blot analysis, cell extracts were resuspended in 2x Laemmli denaturing protein sample buffer, fractionated in 12.5% SDS-PAGE and transferred to a polyvinylidene difluoride (PVDF) membrane. After blocking with 5% milk/0.05% PBS-Tween 20, membranes were incubated for 1 hour with the appropriate primary polyclonal antibodies (anti-YFV hyperimmune rabbit serum, previously produced in our laboratory) diluted 1:500 in 1% milk/0.1% PBS-Tween 20. Membranes were washed 3 times with 1x PBS for 10 minutes/wash and incubated for 1 hour with 1:5,000 goat anti-rabbit IgG antibody conjugated with horseradish peroxidase (Jackson Immunoresearch Laboratories). The Western blot reactions were detected using enhanced chemiluminescence (ECL) reactions (Millipore). For fluorescence assays, cell extracts were fixed in 100% methanol at—20°C for 5 minutes, blocked with 1% BSA/PBS solution for 30 minutes, and incubated with an anti-YFV hyperimmune mouse antibody diluted 1:200 for 1 hour, followed by a 1-hour incubation with secondary antibody diluted 1:500 (Alexa 488-conjugated goat anti-mouse, Molecular Probes, Seattle, WA, USA). The cover slips were then mounted on glass slides using ProLong Gold (Molecular Probes, Seattle, WA, USA) and observed through a confocal microscope. The images were acquired using a Leica SPII-AOBS confocal microscope (Leica Microsystem, Hm) with a 63× oil immersion objective NA 1.3. The Alexa 488 fluorochrome was excited using an ArKr laser at 488 nm. The digital image was acquired using Leica software in a 24-bit RGB format with a 1024 × 1024 pixel area. Fields were chosen for imaging based on the spread and morphology of the cells.
Female BALB/c (H2d) and C57Bl/6 mice (H2b), aged 6 to 8 weeks (Charles River, Kingston, NY, USA), were used for the ELISPOT assays. They were housed in micro-isolator cages under specific pathogen-free conditions and handled in accordance with the Johns Hopkins Institutional Animal Care and Use Committee (IACUC) protocol number MO05M336. The animals were immunized at days zero and 21 and used for the experiments seven to ten days after the last immunization. For the neutralization and protection assays, three-week-old female BALB/c and C57Bl/6 mice were obtained from the Oswaldo Cruz Foundation Breeding Center (Rio de Janeiro, Brazil) and were housed at the Experimental Animal Laboratory (Oswaldo Cruz Foundation, Rio de Janeiro, Brazil) under specific pathogen-free conditions and handled in accordance with the Oswaldo Cruz Foundation Commission for Ethical Animal Use (CEAU) protocol number P0112-02. The animals were immunized at days zero, 30 and 45, and sera were collected via a cut in the tail vein a day before every immunization. For both protocols, the animals were immunized subcutaneously at the base of the tail with either the YF 17DD vaccine at 104 PFUs/50 μl, the DNA constructs at 50 μg/50 μl, or 50 μl of PBS as a negative control.
ELISPOT assays were performed to quantify IFN-gamma spot-forming cells (SFCs) generated via DNA construct immunization. Seven to 10 days after the last immunization, the mice were sacrificed and their spleens were removed. Splenocytes were isolated using standard methods, and single-cell suspensions, depleted of red blood cells, were prepared from freshly isolated splenocytes in culture medium (RPMI 1640 medium supplemented with 10% v/v fetal bovine serum, 100 units/ml penicillin/streptomycin, 2 mM L-glutamine, 50 μM 2-mercaptoethanol and 1 M HEPES buffer). IFN-gamma ELISPOT assays were performed in accordance with the manufacturer’s instructions (BD-Biosciences, San Diego, CA, USA). First, the ELISPOT plates were coated with anti-IFN-gamma antibody at 5 μg/ml and incubated at 4°C overnight. The plates were blocked with RPMI 1640 medium containing 10% FBS for 2 h at room temperature, and total splenocytes (1×106 cells/well) from immunized mice were then added. The cells were cultured at 37°C in 5% CO2 with culture medium alone (RPMI 1640 medium supplemented with 5% v/v fetal calf serum, 100 units/ml penicillin/streptomycin, and 2 mM L-glutamine) or with culture medium in the presence of concanavalin A (2.5 μg/ml; Sigma), 109 PFUs/mL of inactivated YF virus as a positive control (strain 17DD), or individual 15-mers from the envelope protein of the YF 17DD virus at 1 μg/ml. After 16 h of culture, the plates were washed and incubated with biotinylated anti-IFN-gamma for 2 h at room temperature, followed by HRP-conjugated avidin for 1 h at room temperature. Reactions were developed with AEC substrate (Calbiochem-Novabiochem Corporation, San Diego, CA, USA). The quantification of spot-forming cells (SFCs) was carried out using the Immunospot Series Analyzer ELISPOT reader (Cellular Technologies Ltd (CTL), Shaker Heights, OH, USA) with the aid of Immunospot software 3.0 (Cellular Technologies Ltd). The data are represented as the number of SFCs/106. The results were considered positive if the number of SFCs was greater than 20 and higher than the background (culture with medium alone) plus three standard deviations. The results are presented after subtraction of the background.
Plaque reduction neutralization tests (PRNT) were carried out using VERO cells seeded at a density of 62,500 cells/cm2 in 96-well microplates, as previously described [32]. The PRNT tests for the detection of anti-YF nAb were performed after two-fold serial dilutions of serum (1/5 to 1/640) on microtiter plates and incubation with 30 PFUs of the YF 17DD challenge virus strain in each well. After incubation at 37°C in a 5% CO2 atmosphere for 1 h, 50 μl of Vero cell suspensions (4×104/well) in medium 199 (Invitrogen) was added, and the plates were incubated at 37°C for 3 h. The medium was then discarded and the cells were overlaid with 100 μl of medium containing 3.5% carboxymethylcellulose. After 6–7 days of incubation at 37°C in 5% CO2, cell monolayers were fixed with formalin and stained with crystal violet so plaques could be counted. Standard sera of known antibody content in terms of International Units (IU) were included in each set of tests. The log10 dilution of the test and standard sera, which reduced the plaque numbers by 50% relative to the virus control, were determined via interpolation. The mean antibody content at the 50% end point of the standard was then calculated and added to the log10 end point for each sample to give log10 mIU/ml. Plaque neutralization titers were calculated as the highest dilution of antibody able to reduce 50% of the plaques from input virus. The lower limit of detection of the assay was 84.5 mIU/mL.
Groups of three-week-old BALB/c and C57Bl/6 mice immunized three times with either the YF 17DD vaccine or the DNA constructs were inoculated intra-cerebrally with 30 μl of M199 medium containing 105 PFUs of the YF 17DD virus 15 days after the last immunization. The animals were monitored for 21 days and deaths were recorded. Moribund animals were sacrificed by exposure to CO2.
Comparisons between ELISPOT and neutralization assay results were made using an unpaired T-test. The mean survival times in each group of mice were compared using a one-way analysis of variance (ANOVA) and the Kruskal-Wallis non-parametric test. Statistical tests and graphs were performed and produced using GraphPad Prism version 4.0 (GraphPad Software, San Diego, CA, USA; www.graphpad.com).
293 cells were transfected with the p/YFE and pL/YFE expression vectors encoding the pM/M-E and pM/M-E/LAMP proteins, respectively. Validation of plasmid protein expression and cellular steady-state localization was carried out using Western blot and immunofluorescence analyses. Both the E and E/LAMP proteins were stained using polyclonal anti-YF hyperimmune serum. Western blotting detected specific bands for the E (p/YFE) and E/LAMP (pL/YFE) proteins, as well as the wild-type (YF virus) E protein (Fig 2). Immunofluorescence assays showed the characteristic reticular membrane distribution (associated with the typical cellular trafficking of the viral envelope protein) in p/YFE-transfected 293 cells expressing the wild-type E protein (Fig 3A). By contrast, the E/LAMP chimeric protein from pL/YFE-transfected cells showed the typical punctuated lysosomal-like distribution of endogenous LAMP (Fig 3B). The figure represents several independent assays where the expression and cellular steady-state localization of both the E and E/LAMP proteins were considered to be invariable.
It is known that some mouse strains with distinct genetic backgrounds, when exposed to the same antigens, can polarize towards T-helper-1 (Th1) or T-helper-2 (Th2) responses. Some strains are more prone to produce Th1 responses, whereas others are more prone to polarize towards Th2. BALB/c and C57Bl/6 mice strains have been known to produce this type of distinct T-helper responses in several models and, thus, we selected these two strains to investigate how they would respond to our vaccines. An optimized YF 17DD vaccine immunization protocol and IFN-gamma ELISPOT assay conditions, which were previously described for BALB/c (H2d: Dd, Kd, Ld, I-Ad, and I-Ed) mice [33], were used to characterize the T-cell responses to peptides of the YF 17DD virus proteins in C57Bl/6 (H2b: Db, Kb and I-Ab) mice. The first round of experiments with total splenocytes led to the identification of 11 antigenic 15-mer peptides from the YF envelope protein. The subsequent experiments were performed using splenocytes depleted of CD4+ or CD8+ lymphocytes, which lead to the identification of epitopes presented by MHC class I or II, respectively. The depletion typically removed >95% of the targeted population, as assessed via flow cytometry. The CD4-depleted splenocytes, which correspond to the CD8+ lymphocyte response, reacted to seven peptides, whereas the CD8-depleted splenocytes, which correspond to the CD4+ lymphocyte response, were able to respond to five 15-mer peptides (Table 1). The results of quantitative (IFN-gamma SFCs/106) epitope mapping for the YF envelope protein in 17 DD immunized C57Bl/6 mice are shown in Table 2. Splenocytes from naïve animals did not react to any of the peptides tested.
The T-cell responses of H2b and H2d mouse strains induced by immunization with the p/YFE and pL/YFE plasmids were evaluated, and the results were compared with the responses observed for the YF 17DD vaccine immunization. Immunization with the p/YFE or pL/YFE plasmids generated a vigorous T-cell response in C57Bl/6 mice. Both plasmids, in addition to bringing about a T-cell response pattern similar to that produced by the YF 17DD vaccine, were able to elicit a significantly higher number of IFN-gamma SFCs (>200 IFN-gamma SFCs/106 splenocytes) for many immunogenic peptides of the YF envelope protein (Table 3; p<0.05) than was immunization with the attenuated virus vaccine.
Interestingly, p/YFE was able to generate a considerable response to the E413–427 and E417–431 peptides, which were not present after YF 17DD immunization and were very scarce after pL/YFE immunization (Table 3). Remarkably. p/YFE also brought about a stronger response to peptide E1–15 than both the YF 17DD and pL/YFE vaccines.
In BALB/c immunized mice, both the p/YFE and pL/YFE DNA constructs generated an immune response very similar to that obtained with the YF 17DD vaccine, eliciting almost the same immunogenic determinants (Table 3). The only considerable exception was the lack of response in YF-17DD-immunized mice to peptide E329–343, which contains a previously characterized MHC class I epitope (CD8+ response) (Table 4). Both the p/YFE and pL/YFE plasmids also produced a significantly higher number of T cells specific to the immunodominant E57–71 and E61–75 peptides (p<0.05), which contain MHC class I and class II epitopes for the H2d mouse strain. The number of IFN-gamma SFCs was similar for all remaining positive peptides. Groups of mice from both strains immunized with either the empty plasmid or a plasmid expressing only the LAMP protein did not react to any of the peptides tested.
The protection provided by the YF vaccine is mainly attributed to the neutralizing antibody (nAb) response generated after vaccination. Because the presence of nAb is a hallmark of protection, we evaluated the humoral response of C57Bl/6 and BALB/c mice after immunization with the DNA constructs and compared them with the levels of nAb obtained after immunization with the YF 17DD vaccine. To investigate the kinetics of nAb responses, the animals were immunized at days zero, 30 and 45 and bled 15 days after each immunization.
The YF 17DD vaccine was able to produce very high levels of nAb in C57Bl/6 and BALB/c mice after the first immunization (day 15). nAb levels in C57Bl/6 mice seemed to reached a plateau after the second immunization (day 45) and increased slightly after the third immunization (day 60), whereas BALB/c mice showed increasing levels of nAb after the second (day 45) and third (day 60) immunization with the YF 17DD vaccine. The levels of nAb observed in C57Bl/6 mice (≥9,664.0 mIU/mL; obtained at the highest dilution tested) were approximately 20% higher compared with the levels observed in BALB/c mice (7,500±780.1 mIU/mL) (Fig 4). In C57Bl/6 mice, both plasmids expressing the whole envelope protein (p/YFE and pL/YFE) were able to produce significant levels of nAb (p<0.0039 and p<0.002, respectively) after three immunizations compared with the empty vector control. The DNA plasmid pL/YFE, expressing the chimeric E-LAMP protein, led the BALB/c mice to produce higher titers of nAb after the second immunization compared to the p/YFE plasmid. The levels of nAb titers increased after the third immunization and were significantly higher (p<0.045) than those of the control groups immunized with empty vector or PBS (Fig 4).
On average, the pL/YFE DNA immunization elicited nAb titers 7-fold greater than the p/YFE DNA immunization. Compared with the 17DD attenuated virus vaccine, the nAb titers produced by the pL/YFE DNA vaccine were approximately 3.5-fold lower. The fact that these DNA vaccines produced these levels of nAb may still be considered significant.
Intra-cerebral challenge with the YF 17DD virus in mice is a useful model for evaluating the protection provided by vaccine candidates [34]. We evaluated our DNA constructs using the immunization/challenge model by injecting 105 PFUs of the YF 17DD virus intra-cerebrally into DNA-immunized C57Bl/6 and BALB/c mice. The animals were immunized three times at days 0, 30 and 45 and were challenged 15 days after the last immunization. As previously reported [35], immunization with YF 17DD vaccine was able to protect both C57Bl/6 and BALB/c mouse strains against the intra-cerebral challenge. In a similar fashion, immunization with both DNA constructs expressing the full-length YF envelope protein was able to fully protect both mouse strains from the lethal challenge. The majority of mice immunized with PBS or empty vector died 10 to 14 days after the challenge assay (Table 5 and Fig 5).
YF infection continues to be a worldwide problem, especially in tropical areas [11], but this may change as the world continues to be affected by climate change. Despite the high efficiency of commercially available YF vaccines, 17D and 17DD, there are a few reports of rare but fatal side-effects after vaccination [9, 11, 12]. Furthermore, these vaccines are not recommended for infants, pregnant women, immunodeficient subjects, or those allergic to the egg components present in the vaccine formulations [36]. In light of these factors, there is reason to pursue complementary or alternative YF vaccine strategies that could replace the use of the virus-attenuated vaccine version.
Although no DNA vaccines have yet been approved for human use, they represent potential candidates to replace live/attenuated vaccine formulations because they are considered safer. DNA formulations can be easily manipulated, do not require a cold-chain for distribution and eliminate the infectious nature of live/attenuated agents. They also allow the manipulation of immunogens to provide the immune system with the desired epitopes and signals while avoiding the use of unnecessary or potentially harmful antigens or epitopes [37, 38].
Previous studies have described the development of DNA vaccines against flaviviruses based on the expression of the pM/M-E virus sequence cloned in-frame with the LAMP sequence [17, 22, 24]. This approach showed that the chimeric protein, driven by the cytoplasmic sequence of LAMP, was targeted to LAMP-containing organelles, which also co-localized with MHC-class-II-rich intracellular compartments [25]. The immunofluorescence microscopy study of our plasmid expressing the YF pM/M-E in-frame with LAMP (pL/YFE) produced findings similar to those of previous studies and suggests that the presence of LAMP was indeed able to lead the chimeric protein to lysosomes; in contrast, the expression of pM/M-E without LAMP (p/YFE) resulted in a reticular membrane distribution. We also investigated the immunogenicity of these two plasmids as DNA vaccines against YF virus infection. The performance of our DNA constructs was compared with the successful human YF 17DD vaccine, which is a better positive control than the inactivated virus emulsified in CFA, for example [17, 22], which is used when an approved vaccine is not available.
We first carried out epitope mapping of the E protein, comparing the 17DD vaccine with the p/YFE and pL/YFE DNA constructs. We observed that the epitope profile repertoire recognized by the T cells of mice immunized with the DNA constructs was very similar to that of mice immunized with the standard YF vaccine. Moreover, the majority of the T-cell responses in the DNA-immunized mice showed higher numbers of IFN-gamma SFCs compared with the numbers observed for the 17DD vaccine. The immunization of C57Bl/6 mice with the DNA constructs expressing the whole envelope protein resulted in recognition of three extra peptides that were not produced by the YF 17DD vaccine. The T-cell response against the peptide E169–183 generated by p/YFE and pL/YFE was significant, and the responses to the peptides E413–427 and E417–431 were even higher in mice immunized with p/YFE compared with mice immunized with pL/YFE.
It seems that DNA immunization was able to lead to the presentation of some additional epitopes that are not normally induced by the YF 17DD vaccine. It is possible that different antigen-presenting cells (APCs) processed different epitopes, according to the source of the antigen, i.e., attenuated virus or DNA plasmids. However, the lack of response to the E329–343 15-mer observed in BALB/c mice immunized with DNA seemed to be partially because the same cells were able to respond to the minimum epitope within that sequence, as seen against the 9-mer E330–338. Our results also expanded the C57Bl/6 (H2b) epitope mapping for the envelope protein of the YF virus. Sequences E1–15 and E233–247 were previously described as containing CD8 and CD4 epitopes, respectively [39]; however, we were able to identify several new epitopes, six for CD4 and four for CD8 (Table 2).
The presence of anti-YF nAb is a recognized hallmark of protection against YF infection. A dose of 104 PFUs of YF 17DD virus was potent enough to produce a high concentration of nAb after a single immunization in both mouse strains. The plasmid expressing the chimeric E-LAMP protein (pL/YFE) was able to produce significantly higher concentrations of anti-YF nAb after three immunizations in both mouse strains compared with the controls or p/YFE; however, the levels of nAb were considerably lower compared with the YF 17DD immunization. The p/YFE plasmid, expressing only the YF E protein, failed to generate high levels of anti-YF nAb in BALB/c mice and produced only a modest increase of anti-YF nAb in C57Bl/6 mice. These data are in accordance with previous reports demonstrating that the expression of chimeric proteins in-frame with LAMP lead to an improvement in B-cell responses [17, 22, 30]. Others have used extended immunization protocols to produce higher levels of antibody [17]. Although we have not tested this hypothesis here, it is interesting to speculate that extra DNA immunizations could further increase the levels of anti-YF nAb observed.
In addition to nAb, complement-fixing antibodies have also been described as a protective mechanism against YF [40, 41]. In fact, it has been shown in an animal model that expression of the YF NS1 protein in the vaccinia virus could partially protect mice from an intracranial challenge [42], most likely through a mechanism involving complement-fixing antibodies. We cannot rule out the hypothesis that immunization with the YF envelope protein, as a DNA plasmid, could lead to the presentation of B-cell epitopes different from those found after immunization with the attenuated YF vaccine. Moreover, these epitopes could be sites for neutralizing and complement-fixing antibodies. Thus, the characterization of the B-cell epitopes in the context of DNA vaccines could potentially become a relevant parameter for comparing DNA vaccines to their virus-attenuated counterparts.
To further explore the protection provided by DNA immunization, we challenged immunized mice with an intracranial injection of the 17DD virus. This in vivo protection assay enables the evaluation of how effectively a vaccine candidate can prevent the encephalitis caused by infection with the YF virus, and it has been extensively used [35, 41–44]. Both DNA constructs, p/YFE and pL/YFE, were able to protect both mouse strains from an intracranial challenge. These two plasmids were able to promote a very similar profile of T-cell epitope recognition compared to the YF 17DD vaccine. However, only the pL/YFE was able also to produce significant levels of anti-YF nAb. It is possible that the p/YFE plasmid, which did not raise appreciable levels of anti-YF nAb, led to the protection of the challenged mice through complementary mechanisms in the presence of low levels of nAb. It is also possible that strong T-cell responses were able to mediate protection in this system. T cells may play a role in protection from encephalitis caused by flaviviruses; it has been demonstrated that the depletion of CD4+ and/or CD8+ lymphocytes leads to a decrease in the protection offered by an experimental vaccine expressing the dengue envelope protein in the context of the YF virus [39]. The possibility that these DNA vaccines provide T-cell mediated protection is very interesting and we are planning to investigate this in more detail.
The results reported here are very encouraging, and we are confident that this vaccine candidate is worth further investigation in more relevant animal models, specifically in a non-human primate challenge model. It is interesting that even with lower neutralizing antibody titers the DNA vaccine was still capable of protection, suggesting an important role for T-cell mediated protection. In further studies, we plan to dissect in more details the mechanisms of protection provided by these DNA vaccines. Another critical point is the duration of the protection, and this also needs to be addressed in more relevant animal models. In summary, this research shows that expression of the envelope protein in-frame with the cytoplasmic targeting sequence of LAMP led to high levels of anti-YF nAb and produced a strong T-cell response. The possibility of generating a protective anti-YF response through a DNA vaccine may provide a safer alternative to the attenuated YF virus vaccine and should be further investigated.
|
10.1371/journal.pcbi.1000119 | Intrinsic Gain Modulation and Adaptive Neural Coding | In many cases, the computation of a neural system can be reduced to a receptive field, or a set of linear filters, and a thresholding function, or gain curve, which determines the firing probability; this is known as a linear/nonlinear model. In some forms of sensory adaptation, these linear filters and gain curve adjust very rapidly to changes in the variance of a randomly varying driving input. An apparently similar but previously unrelated issue is the observation of gain control by background noise in cortical neurons: the slope of the firing rate versus current (f-I) curve changes with the variance of background random input. Here, we show a direct correspondence between these two observations by relating variance-dependent changes in the gain of f-I curves to characteristics of the changing empirical linear/nonlinear model obtained by sampling. In the case that the underlying system is fixed, we derive relationships relating the change of the gain with respect to both mean and variance with the receptive fields derived from reverse correlation on a white noise stimulus. Using two conductance-based model neurons that display distinct gain modulation properties through a simple change in parameters, we show that coding properties of both these models quantitatively satisfy the predicted relationships. Our results describe how both variance-dependent gain modulation and adaptive neural computation result from intrinsic nonlinearity.
| Many neurons are known to achieve a wide dynamic range by adaptively changing their computational input/output function according to the input statistics. These adaptive changes can be very rapid, and it has been suggested that a component of this adaptation could be purely input-driven: even a fixed neural system can show apparent adaptive behavior since inputs with different statistics interact with the nonlinearity of the system in different ways. In this paper, we show how a single neuron's intrinsic computational function can dictate such input-driven changes in its response to varying input statistics, which begets a relationship between two different characterizations of neural function—in terms of mean firing rate and in terms of generating precise spike timing. We then apply our results to two biophysically defined model neurons, which have significantly different response patterns to inputs with various statistics. Our model of intrinsic adaptation explains their behaviors well. Contrary to the picture that neurons carry out a stereotyped computation on their inputs, our results show that even in the simplest cases they have simple yet effective mechanisms by which they can adapt to their input. Adaptation to stimulus statistics, therefore, is built into the most basic single neuron computations.
| An f-I curve, defined as the mean firing rate in response to a stationary mean current input, is one of the simplest ways to characterize how a neuron transforms a stimulus into a spike train output as a function of the magnitude of a single stimulus parameter. Recently, the dependence of f-I curves on other input statistics such as the variance has been examined: the slope of the f-I curve, or gain, is modulated in diverse ways in response to different intensities of added noise [1]–[4]. This enables multiplicative control of the neuronal gain by the level of background synaptic activity [1]: changing the level of the background synaptic activity is equivalent to changing the variance of the noisy balanced excitatory and inhibitory input current to the soma, which modulates the gain of the f-I curve. It has been demonstrated that such somatic gain modulation, combined with saturation in the dendrites, can lead to multiplicative gain control in a single neuron by background inputs [5]. From a computational perspective, the sensitivity of the firing rate to mean or variance can be thought of as distinguishing the neuron's function as either an integrator (greater sensitivity to the mean) or a differentiator/coincidence detector (greater sensitivity to fluctuations, as quantified by the variance) [3],[6],[7].
An alternative method of characterizing a neuron's input-to-output transformation is through a linear/nonlinear (LN) cascade model [8],[9]. These models comprise a set of linear filters or receptive field that selects particular features from the input; the filter output is transformed by a nonlinear threshold stage into a time-varying firing rate. Spike-triggered covariance analysis [10],[11] reconstructs a model with multiple features from a neuron's input/output data. It has been widely employed to characterize both neural systems [12]–[15] and single neurons or neuron models subject to current or conductance inputs [16]–[19].
Generally, results of reverse correlation analysis may depend on the statistics of the stimulus used to sample the model [15], [19]–[25]. While some of the dependence on stimulus statistics in the response of a neuron or neural system may reflect underlying plasticity, in some cases, the rapid timescale of the changes suggests the action of intrinsic nonlinearities in systems with fixed parameters [16], [19], [25]–[29], which changes the effective computation of a neuron.
Our goal here is to unify the f-I curve description of variance-dependent adaptive computation with that given by the LN model: we present analytical results showing that the variance-dependent modulation of the firing rate is closely related to adaptive changes in the recovered LN model if a fixed underlying model is assumed. When the model relies only on a single feature, we find that such a system can show only a single type of gain modulation, which accompanies an interesting asymptotic scaling behavior. With multiple features, the model can show more diverse adaptive behaviors, exemplified by two conductance-based models that we will study.
Recently, Higgs et al. [3] and Arsiero et al. [4] identified different forms of variance-dependent change in the f-I curves of various neuron types in avian brainstem and in cortex. Depending on the type, neurons can have either increasing or decreasing gain in the f-I curve with increasing variance. These papers linked the phenomenon to mechanisms underlying spike rate adaptation, such as slow afterhyperpolarization (sAHP) currents and slow sodium channel inactivation. We recently showed [7] that a standard Hodgkin–Huxley (HH) neuron model, lacking spike rate adaptation, can show two different types of variance-dependent gain modulation simply by tuning the maximal conductance parameters of the model. These differences in gain modulation correspond to two different regimes in the space of conductance parameters. In one regime, which includes the standard parameters, a neuron periodically fires to a sufficiently large constant input current. In the other regime, a neuron never fires to a constant input regardless of its magnitude, but responds only to rapid fluctuations. This rarely discussed property has been termed class 3 excitability [30],[31]. Higgs et al. [3] proposed that the type of gain modulation classifies the neuron as an integrator or differentiator.
Here, we examine two models that show these different forms of variance-dependent gain modulation without spike rate adaptation, and study the resulting LN models sampled with different stimulus statistics. We show that these fixed models generate variance-dependent gain modulation, and that this gain modulation is well predicted by aspects of the LN models derived from white noise stimulation. The two models are both based on the HH [32] active currents; one model is the standard HH model, and the other (HHLS) has lower Na+ and higher K+ conductances. The HHLS model is a class 3 neuron and responds only to a rapidly changing input. For this reason, the HHLS model can be thought of as behaving more like a differentiator than an integrator [3],[7].
Figure 1 shows the different gain modulation behaviors of the HH and HHLS conductance-based models. For the HH model, Figure 1A, the f-I curves in the presence of noise are similar to the noiseless case except that they are increasingly smoothed at the threshold. In contrast, Figure 1C shows that the f-I curves of the HHLS model never converge toward each other as the noise level increases. This case resembles that of layer 5 pyramidal neurons in rat medial prefrontal cortex [4], as well as nucleus laminaris (NL) neurons in the chick auditory brainstem and some pyramidal neurons in layer 2/3 of rat neocortex [3]. While for these layer 2/3 neurons, there is evidence that this change in f-I curve slope may be related to the sAHP current [3], at steady state this effect can be obtained in general by tuning the maximal conductances without introducing any mechanism for spike rate adaptation [7].
For a system described by an LN model with a single feature, we derive an equation relating the slopes of the firing rate with respect to stimulus mean and variance. We then consider gain modulation in a system with multiple relevant features and derive a series of equations relating gain change to properties of the spike-triggered average and spike-triggered covariance. Throughout, we assume that the underlying system is fixed, and that its parameter settings do not depend on stimulus statistics. For example, if the model has a single exponential filter with a time constant τ, we assume that τ does not change with the stimulus mean (I0) or variance (σ2). However, this does not mean that the model shows a single response pattern regardless of the statistical structure of stimuli. The sampled LN description of a nonlinear system with fixed parameters—even when the underlying model is an LN model [25]—can show interaction with the input statistics, leading to different LN model descriptions for different input parameters [19], [25], [27]–[29]. We refer to this as intrinsic adaptation.
An LN model is composed of its relevant features {εμ(t)} (μ = 1,2,…,n)), which act as linear filters on an incoming stimulus, and a probability to spike given the filtered stimulus, P(spike|filtered stimulus). For a Gaussian white noise stimulus with mean I0 and variance σ2, the firing rate is(1)where is the time-integrated filter and x is the mean-subtracted noise stimulus filtered by the n relevant features. p(x) is an n-dimensional Gaussian distribution with variance σ2. We refer to the Materials and Methods section for a more detailed account of the model.
For a one-dimensional model n = 1, Equation 1 can be rewritten with change of variables(2)Since p(x) is Gaussian, it is also the kernel or Green's function of a diffusion equation in terms of (x,σ2) and therefore so is p(x−I0ε̅) in terms of (I0,σ2). In other words, we haveNow operating with on both sides of the equation, p(x−I0ε̅) is the only term on the left hand side of Equation 2 that depends on (I0,σ2) and therefore the right hand side of Equation 2 vanishes. Thus one finds(3)The boundary condition is given by evaluating Equation 2 as σ2→0; here the Gaussian distribution becomes a delta functionand the boundary condition is given by the zero-noise f-I curve. Thus, when a model depends only on a single feature, ε(t), the f-I curve with a noisy input is given by a simple diffusion-like equation, Equation 3, with a single parameter, the time integrated filter, , determining the diffusion constant 1/2ε̅2.
Equation 3 states that the variance-dependent change in the firing rate is simply determined by the curvature of the f-I curve. Thus, a one-dimensional system displays only a single type of noise-induced gain modulation: as in diffusion, an f-I curve is gradually smoothed and flattened as the variance increases. Given a boundary condition, such as an f-I curve for a particular variance, the family of f-I relations can be reconstructed up to a scale factor by solving Equation 3. For example, one can predict how the neuron would respond to a noise stimulus based on its output in the absence of noise. Note that the solution of Equation 3 generalizes a classical result [33] based on a binary nonlinearity to a simple closed form which applies to any type of nonlinearity.
Figure 2A and 2B show a solution of Equation 3. While this one-dimensional model is based on the simplest and most general assumptions, it provides insights into the structure of variance-dependent gain modulation. The boundary condition is an f-I curve with no noise, f = (I+0.1)1/2 for I>0 and f = 0 for I≤0, which imitates the general behavior of many dynamical neuron models around rheobase [34]–[36]. Compared with the HH conductance-based model, Equation 3 captures qualitative characteristics of the HH f-I curve despite differences due to the increased complexity of the HH model over a 1D LN model: in Figure 2A and 2B, there is a positive curvature (second derivative of firing rate with respect to current) of the f-I curve below rheobase related to the increase of the firing rate with increasing variance. In contrast, the behavior of the HHLS model cannot be described by Equation 3. Even though the f-I curves in Figure 1C mostly have negative curvature, the firing rate keeps increasing with variance, implying that the HHLS model cannot be described by a one-dimensional LN model.
We also compared Equation 3 with the f-I curves from two commonly used simple neuron models, the leaky integrate-and-fire (LIF) model (Figure 2C), and a similar model with minimal nonlinearity, the quadratic integrate-and-fire (QIF) model [37],[38] (Figure 2D). The f-I curves of the two models are similar but have subtle differences: in the LIF model, firing rate never decreases with noise, even though parameters were chosen to induce a large negative curvature, as shown analytically in Text S1. The QIF model behavior is much more similar to the 1D LN model, marked by a slight decrease in firing rate at large I0. From this perspective, the QIF is a simpler model in terms of the LN description despite the dynamical nonlinearity.
It is interesting to note that for one-dimensional models, the gain modulation given by Equation 3 depends only on the boundary condition, which implicitly describes how an input with a given mean samples the nonlinearity, but not explicitly on the details of filters or nonlinearity. An ideal differentiator, where firing rate is independent of the stimulus mean, is realized only when the filter has zero integral, ε̅ = 0. This is also the criterion that would be satisfied if the filter itself were ideally differentiating. We will return to the relationship between the LN model functional description and that of the f-I curves in the Discussion.
Here we examine gain modulation in the case of a system with multiple relevant features. In this case, one cannot derive a single simple equation such as Equation 3. Instead, we derive relationships between the characteristics of f(I0,σ) curves and quantities calculated using white noise analysis.
Fixed multidimensional models can display far more complex response patterns to different stimulus statistics than one-dimensional models, because linear components in the model can now interact nonlinearly [29]. For example, in white noise analysis, as the stimulus variance increases, the distribution of the filtered stimuli also expands and probes different regions of the nonlinear threshold structure of the model. This induces a variance-dependent rotation among the filters recovered through sampling by white noise analysis, and the corresponding changes in the spike-triggered average, spike-triggered covariance, and the sampled nonlinearity [19].
Here, we relate parameters of the changing spike-triggered average and spike-triggered covariance description to the form of the f-I curves. The relationships are derived by taking derivatives of each side of Equation 1 with respect to I0 and σ2 (see Materials and Methods section). The first order in I0 establishes the relationship between the STA and the gain of the f-I curve with respect to the mean(4)The second order leads to a relationship between the second derivative of the f-I curve and the covariance matrix(5)The gain with respect to the variance is(6)where
Equations 4–6 show how the nonlinear gain of an f-I curve with respect to input mean and variance is related to intrinsic adaptation as observed through changes in the STA and STC. Note that Equations 4–6 apply to one-dimensional LN models as well. In that case, the STA has the same shape as the feature in the model, and only its magnitude varies according to the overlap integral, Equation 1, between the nonlinearity of the model and the prior stimulus. This is the same for the STC, and thus Equations 4–6 are not independent. This leads to a single form of variance gain modulation, given by Equation 3. However, in a multidimensional model, changing the stimulus mean shifts the nonlinearity in a single direction, , while increasing the variance expands the prior in every direction in the stimulus space. Therefore, the overlap integral can show more diverse behaviors.
We now examine whether the gain modulation behaviors we have described can be captured by a multi-dimensional LN model. We tested this by computing f-I curves, spike-triggered averages and the spike-triggered covariance matrices for the noise-driven HH and HHLS models for a range of input statistics. Figure 3A, B, and C show the result of fitting simulation data from the HH (left) and HHLS (right) model to Equations 4, 5, and 6, respectively. The linear relationships are quite clear in Figure 3A and 3C which show the gains with respect to mean and variance. Figure 3B involves the curvature of f-I curves, which is more difficult to calculate accurately, and shows larger errors. In every case, goodness of fit is p<1.3×10−6 and p<5.8×10−6 for the HH and HHLS where the upper bounds of p-values are given by the case of Equation 5, corresponding to Figure 3B. These results show that intrinsic adaptation of the LN model predicts the form of noise-induced gain modulation for these models.
Here we discuss a consequence of intrinsic adaptation for neuronal encoding of mean and variance information for a one-dimensional model. In this case, Equation 3 completely specifies intrinsic adaptation, and therefore we will focus on this case.
Our first observation is that Equation 3 is invariant under the simultaneous rescaling of the mean and standard deviation, I0→αI0, σ→ασ, where α is an arbitrary positive number. This invariance is preserved if the solution is also a function of only a dimensionless variable I0/σ, which would represent a signal-to-noise ratio if we describe the neuron's input/output function in terms of an f-I curve at a fixed noise level σ. Note that this situation is analogous to the Weber–Fechner [39],[40] and Fitts' law [41], which states that perception tends to depend on only dimensionless variables that are invariant under scaling of the absolute magnitude of stimulus [42]. However, the invariance of Equation 3 under the scaling of a stimulus does not necessarily lead to the invariance of a firing rate solution. By rewriting Equation 2 in terms of the “rescaled” variables, y = x/σ and μ = I0/σ, we get(7)where f0(I) = P(spike|Iε̅) is an f-I curve with no noise. Thus, the scaling of f(I0,σ2) with standard deviation depends on the boundary condition, f0(I), which in principle can be any arbitrary function.
Nevertheless, in practice, the f-I curves of many dynamical neurons are not completely arbitrary but can share a simple scaling property, at least asymptotically. For example, in the QIF and many other neuron models, the f-I curve with no noise asymptotically follows a power law f0∼(I0−Ic)1/2 around the rheobase Ic [34]–[36]. In general, if f0(I)∝Iα asymptotically in such a regime, from Equation 7, the firing rate is asymptotically factorized into a σ dependent and μ = I0/σ dependent part as(8)In other words, I0/σ becomes an intermediate asymptotic of the f-I curves [43].
To test to what extent this scaling relationship holds in the models we have considered, we calculated the rescaled relative gain of the f-I curves, which we define as (σ/f) ∂f/∂I0 = σ ∂ log f/∂I0; the rescaled relative gain of Equation 8 depends only on μ = I0/σ, not on σ. Thus, if the rescaling strictly holds, this becomes a single-valued function of the signal-to-noise ratio, I0/σ, regardless of the noise level σ.
We find evidence for this form of variance rescaling in the QIF, LIF, and HH models. Figure 4 shows the rescaled gains evaluated from the simulated data. The QIF and HH case, Figure 4B and 4D, match well with the solution of Equation 3, Figure 4A. In the LIF case, Figure 4C, the relative gain shows deviations at low variance, but it approaches a variance-independent limit at large σ. We also present an analytic account in Text S1. On the other hand, in Figure 4E, the HHLS model does not exhibit this form of asymptotic scaling at all. The role of the signal-to-noise ratio, I0/σ, in the HHLS model appears to be quite distinct from the other models. In summary, Equation 3 predicts that one-dimensional LN models will have the tendency to decrease gain with increasing noise level. However, if the f-I curve of a neuron is power-law-like, the resulting gain modulation will be such that the neuron's sensitivity to mean stimulus change at various noise levels is governed only by the signal-to-noise ratio.
In this paper, we have obtained analytical relationships between noise-dependent gain modulation of f-I curves and properties of the sampled linear/nonlinear model. We have shown that gain control arises as a simple consequence of the nonlinearity of the LN model, even with no changes in any underlying parameters.
For a system described by an LN model with only one relevant feature, a simple single-parameter diffusion relationship relates the f-I curves at different variances, where the role of the diffusion coefficient is taken by the integral of the STA. This form strictly limits the possible forms of gain modulation that may be manifested by such a system. The result qualitatively describes the variance dependent gain modulation of different neuron models such as the LIF, QIF, and standard HH neuron models. Models based on dynamical spike generation, such as QIF, showed better agreement with this result than the LIF model. The QIF model case is a good example of how a nonlinear dynamical system can be mapped onto an LN model description [19],[44]. The QIF model has a single dynamical equation whose subthreshold dynamics are captured approximately by a linear kernel, which takes the role of the feature; one can then determine a threshold which acts as a binary decision boundary for spiking. Thus, it is reasonable that the QIF model and the one-dimensional LN model show a similar response pattern to a noisy input. When the system has multiple relevant features, we obtain equations relating the gain with respect to the input mean and the input variance to parameters of the STA and STC. We verified these results using HH neurons displaying two different forms of noise-induced gain control.
Previous work has related different gain control behaviors to a neuron's function as an integrator or a differentiator [3],[7]. From an LN model perspective, the neuron's function is defined by specific properties of the filter or filters ε(t). An integrating filter would consist of entirely positive weights; for leaky integrators these weights will decay at large negative times. A differentiating filter implements a local subtraction of the stimulus, and so should consist of a bimodal form where the positive weights approximately cancel the negative weights.
In general, characterizations of neural function by LN model and by f-I curves are quite distinct. The f-I approach we have discussed here describes the encoding of stationary statistical properties of the stimulus by time-averaged firing rate, while the LN model describes the encoding of specific input fluctuations by single spikes, generally under a particular choice of stimulus statistics. Indeed, the LN characterization can change with the driving stimulus distribution, even, in principle, from an integrator to a differentiator. Thus, a model may, for example, act as a differentiator on short timescales but as an integrator on longer timescales. For systems whose LN approximation varies with mean and variance, the neuron's effective computation changes with stimulus statistics, and so does the information that is represented. One might then ask how the system can decode the represented information. It has been proposed that statistics of the spike train might provide the information required to decode slower-varying stimulus parameters [22],[45]. The possibility of distinguishing between responses to different stimulus statistics using the firing rate alone depends on the properties of the f-I curves.
The primary focus of this work is the restricted problem of single neurons responding to driving currents, where the integrated synaptic current in an in vivo-like condition is approximated to be a (filtered) Gaussian white noise [46]–[50]. However, our derivations can apply to arbitrary neural systems driven by white noise inputs, if f-I curves are interpreted as tuning functions with respect to the mean stimulus parameter. Given the generality of our results for neural systems, it would be interesting to test our results in cases where firing is driven by an external stimulus. A good candidate would be retinal ganglion cells, which are well-described by LN-type models [9], [14], [51]–[53], show adaptation to stimulus statistics on multiple timescales [23],[54] and display a variety of dimensionalities in their feature space [14].
A limitation of the tests we have performed here is a restriction to the low firing rate regime where spike-triggered reverse correlation captures most of the dependence of firing probability on the stimulus. The effects of interspike interaction can be significant [16],[17],[55] and models with spike history feedback have been developed to account for this [44],[51],[56],[57]. We have not investigated how spike history effects would impact our results.
Although evidence suggests that gain modulation by noise may be enhanced by slow afterhyperpolarization currents underlying spike frequency adaptation [3], these slow currents are not required to generate gain enhancement in simple neuron models [7], [19], [25]–[29]. While one may generate diverse types of noise-induced gain modulation only by modifying the mechanism of generating a spike independent of spike history [7], in realistic situations, slow adaptation currents are present and will affect neural responses over many timescales [58]–[60]. In principle, it is possible to extend our result to include these effects: f-I curves under conditions of spike frequency adaptation have been already discussed [61]–[63] and can be compared to LN models with spike history feedback. However, our goal here was to demonstrate the effects that can occur independent of slow adaptation currents and before such currents have acted to shift neuronal coding properties.
The suggestive form of our result for one-dimensional LN models led us to look for a representation of neuronal output that is invariant under change in the input noise level. Our motivation is based on a simple principle of dimensional analysis: the gains of the f-I curves with noise may be asymptotically described by a signal-to-noise ratio, a dimensionless variable depending only on the stimulus itself. We showed that this may occur if the f-I curve with no noise obeys asymptotic power-law properties. Such a property has been determined to arise both from the bifurcation patterns of spike generation [31],[34],[35] and due to spike rate adaptation [61]. This relationship implies that the gain of the firing rate as a function of the mean should scale inversely with the standard deviation. Scaling of the gain of the nonlinear decision function with the stimulus standard deviation has been observed to some degree in a number of neural systems [10], [15], [22]–[25], [29], [64]–[67]. Such scaling guarantees maximal transmission of information [10],[22]. As we and others have proposed, a static model might suffice to explain this phenomenon [25],[27], although in some cases slow adaptation currents are known to contribute [65],[66].
In summary, we have presented theoretically derived relationships between the variance-dependent gain modulation of f-I curves and intrinsic adaptation in neural coding. In real neural systems, any type of gain modulation likely results from many different mechanisms, possibly involving long-time scale dynamics. Our results show that observed forms of gain modulation may be a result of a pre-existing static nonlinearity that reacts to changes in the stimulus statistics robustly and almost instantaneously.
We used two single compartmental models with Hodgkin–Huxley (HH) active currents. The first one is an HH model with standard parameters while the second model (HHLS) has a lower Na+ and higher K+ maximal conductance. The voltage changes are described by [32]and the activation variables m, n, and h behave according towhereThe voltage V is in millivolts (mV).
For the HH model, the conductance parameters are g̅K = 36 mS/cm2 and g̅Na = 120 mS/cm2. The HHLS model has g̅K = 41 mS/cm2 and g̅Na = 79 mS/cm2. All other parameters are common to both models. The leak conductance is g̅L = 0.3 mS/cm2 and the membrane capacitance per area C is 1 μF/cm2. The reversal potentials are EL = −54.3 mV, ENa = 50 mV, and EK = −77 mV. The membrane area is 10−3 cm2, so that a current density of 1 μA/cm2 corresponds to a current of 1 nA.
All simulations of these models were done with the NEURON simulation environment [68]. Gaussian white noise currents with various means and variances are generated with an update rate of 5 kHz (dt = 0.2 ms) and delivered into the model via current clamp. For the f-I curves, we simulated 4 min of input for each mean and variance pair. The whole procedure was repeated five times to estimate the variance of the f-I relationship, σrepeat.
We ran another set of simulations for reverse correlation analysis and collected about 100,000 spikes for each stimulus condition. The means and variances of the Gaussian noisy stimuli were chosen such that the mean firing rate did not exceed 10 Hz, and we selected eight means and seven variances for the HH model, and nine means and four variances for the HHLS model.
In addition to the conductance-based model, we investigated the behavior of two heuristic model neurons driven by a noisy current input. Each model consists of a single dynamical equation describing voltage fluctuations of the form
The first model is a leaky integrate-and-fire (LIF) model [69],[70], for which L(V) = −gL(V−EL). We used the parameters gL = 2, EL = 0, and C = 1. Since this choice of L(V) cannot generate a spike, we additionally imposed a spiking threshold Vth = 1 and reset voltage Vreset = −3.
The second is a quadratic integrate-and-fire (QIF) model [31],[37],[38], for which L(V) = gL(V−EL)(V−Vth)/ΔV where ΔV = Vth−EL>0. We used gL = 0.5, EL = 0, Vth = 0.1, and C = 1. In this model, the voltage V can increase without bound; such a trajectory is defined to be a spike if it crosses Vspike = 5. After spiking, the system is reset to Vreset = 0.
These two models are simulated using a fourth-order Runge–Kutta integration method with an integration time step of dt = 0.01. The input current I(t) was Gaussian white noise, updated at each time step, with a range of means and variances. The f-I curves were obtained from 1,000 s of stimulation for each (mean,variance) condition. We then compared the f-I curves from these models with the relationship derived in the Results section, Equation 5. A numerical solution of the partial differential equation was obtained using a PDE solver in Mathematica (Wolfram Research, Inc.).
We use the linear/nonlinear (LN) cascade model framework to describe a neuron's input/output relation. We will focus on the dependence of the firing rate of a fixed LN model on the mean and variance of a Gaussian white noise input.
We will take the driving input to be I(t) = I0+ξ(t) where I0 is the mean and ξ(t) is a Gaussian white noise with variance σ2 and zero mean. The linear part of the model selects, by linear filtering, a subset of the possible stimuli probed by I(t). That subset is expressed as n relevant features {εμ(t)}, (μ = 1,2,…,n). Interpreted as vectors, the components of any stimulus that are relevant to changing the firing rate can be expressed in terms of projections onto these features. The firing rate of the model for a given temporal sequence I(t) depends only on s, the input filtered by the n relevant features. Thus the firing rate from the given stimulus depends on the convolution of the input with all n features and can be written as P(spike|s = I0ε̅+x) whereSince I(t) is white noise with stationary statistics, the projections xμ can be taken to be stationary random variables chosen from a Gaussian distribution at each t.
Given the filtered stimulus, a nonlinear decision function P(spike|I0ε̅+x) generates the instantaneous time-varying firing rate. For a specified model and stimulus statistics, the mean firing rate f(I0,σ2) = P(spike) is simply(9)where
Equation 9 describes an f-I curve of the model in the presence of added noise with variance σ2. The slope or gain of the firing rate with respect to mean or variance can be computed if P(spike|I0ε̅+x) is known. However, the gains can be also obtained in terms of the first and second moments of P(spike|I0ε̅+x), which can be measured directly by reverse correlation analysis.
We used spike-triggered reverse correlation to probe the computation of the model neurons through an LN model. We collected about 100,000 spikes and corresponding ensembles of spike triggered stimulus histories in a 30 ms long time window preceding each spike.
From the spike-triggered input ensembles, we calculated spike-triggered averages (STAs) and spike-triggered covariances (STCs). The STA is simply the average of the set of stimuli that led to spikes subtracted from the mean of the “prior” stimulus distribution, the distribution of all stimuli independent of spiking output(10)Therefore, one may consider only the noise part of the zero mean stimulus.
When computing the STC, the prior's covariance is subtracted(11)
In calculating the slope and curvature of the f-I curves, we used 6–10 degree polynomial fitting of the f-I curves, where in any single case, the lowest degree was used which provided both a good fit and smoothness. From the fitting procedure, we obtained the standard deviation of the residuals, σfit. This was repeated five times for f-I curves computed using different noise samples, and from this we computed σrepeat, the standard deviation of each computed slope and curvature. We estimated the total error of our calculation as σtotal = (σrepeat2+σfit2)1/2. In practice, σrepeat was always greater than σfit by an order of magnitude. This σtotal was used for the error bars in Figure 3.
To evaluate the goodness of fit in Figure 3, we used the Pearson χ2 test by using the reduced χ2 statisticwhere O and E represent the right and left hand sides of Equations 4–6, respectively. From this, the p-values are estimated from the cumulative density function of the χ2 distribution, Q(χ2/k,k). The degree of freedom is k = 54 and k = 34 for the HH and HHLS, respectively.
We first present two key identities: the first one, which depends on the form of s having additive mean and noise components, is a change of variables for the gradient of P(spike|x+I0ε̅)(12)Secondly, when x is a Gaussian random variable with zero mean and variance σ2, by using integration by parts in can be seen that any function F(x) satisfies(13)
Then, we first take derivatives of both sides of Equation 9 (or equivalently Equation 1), by I0 and σ2, and apply Equations 12 and 13. The first order in I0 is(14)The second order is given by(15)where δμν is a Kronecker delta symbol. The gain with respect to variance is(16)
Now, we show how the right hand sides of Equations 14–16 correspond to the STA and the STC. Given a Gaussian white noise signal ξ(t), we can split it as ξ = ξ∥+ξ⊥, where ξ∥ belongs to the space spanned by our basis features {εμ}, and therefore relevant to spiking. ξ⊥ is the orthogonal or irrelevant part. ξ∥ can be written asAgain, x is a Gaussian variable from a distribution Equation 9.
The STA issince ξ⊥ is irrelevant and does not make any contribution. Here we use Bayes theoremAs in Equation 9, P(s = x+I0ε̅) = p(x), and therefore the STA becomesComparing this result with Equation 14, we obtain Equation 4.
A similar calculation for the second order [19] showsThis result, combined with Equations 15 and 16, leads to Equations 5 and 6, respectively. |
10.1371/journal.pcbi.1005437 | Stimulus-specific adaptation in a recurrent network model of primary auditory cortex | Stimulus-specific adaptation (SSA) occurs when neurons decrease their responses to frequently-presented (standard) stimuli but not, or not as much, to other, rare (deviant) stimuli. SSA is present in all mammalian species in which it has been tested as well as in birds. SSA confers short-term memory to neuronal responses, and may lie upstream of the generation of mismatch negativity (MMN), an important human event-related potential. Previously published models of SSA mostly rely on synaptic depression of the feedforward, thalamocortical input. Here we study SSA in a recurrent neural network model of primary auditory cortex. When the recurrent, intracortical synapses display synaptic depression, the network generates population spikes (PSs). SSA occurs in this network when deviants elicit a PS but standards do not, and we demarcate the regions in parameter space that allow SSA. While SSA based on PSs does not require feedforward depression, we identify feedforward depression as a mechanism for expanding the range of parameters that support SSA. We provide predictions for experiments that could help differentiate between SSA due to synaptic depression of feedforward connections and SSA due to synaptic depression of recurrent connections. Similar to experimental data, the magnitude of SSA in the model depends on the frequency difference between deviant and standard, probability of the deviant, inter-stimulus interval and input amplitude. In contrast to models based on feedforward depression, our model shows true deviance sensitivity as found in experiments.
| We present a possible mechanism for the way auditory cortex emphasizes stimuli that are deviant within a regular, repetitive sequence. This enhancement is strong and widespread in auditory cortex, but not in its major thalamic input, the ventral division of the medial geniculate body. In contrast with previous models, which are based on depression of the synapses that convey the input to the cortex, here the network structure and the known dynamics of intracortical synapses play a key role. The model accounts better than previous models for available experimental data, and provides testable predictions that differentiate it from feedforward models. It is a useful starting point for studying the circuit mechanisms that underlie cortical responses to unexpected stimuli.
| Stimulus-specific adaptation (SSA) is the decrease in responses to a repeating stimulus (standard) that does not generalize to another, rarely-occurring stimulus (deviant). SSA is a robust and widespread finding in the auditory system. It has been demonstrated in single neurons in primary auditory cortex (A1) of anesthetized cats [1], and in single-unit, multi-unit and local field potential (LFP) recordings in A1 of awake [2] and anesthetized rats [3,4]. Although SSA is present in the inferior colliculus [5–7] and the auditory thalamus [8–11], it is mostly confined to the non-lemniscal pathway [8,9] and is thus likely generated de novo in A1, whose major thalamic input is from the lemniscal part of the medial geniculate body. SSA in A1 has true deviance sensitivity: The response to a rare stimulus presented within a sequence composed mostly of the same standard tone (which violates the expectation for yet another repeat of the standard) is larger than the response to the same rare stimulus when presented within a sequence composed of many different tones (that doesn't generate strong expectations for any stimulus) although the level of sensory adaptation in the multi-tone sequence may be lower [4,12,13]. SSA is therefore an appealing test case for linking high-level concepts such as deviance sensitivity with mechanistic models.
SSA shares many similarities with mismatch negativity (MMN) [14], an event-related potential evoked by deviant stimuli that has been investigated extensively in humans [15]. While it is clear that SSA is not the direct neuronal correlate of MMN [16,17], it may be one of the stages leading to MMN. In fact, human midlatency potentials correspond better with SSA temporally. Like SSA, midlatency potentials show deviance sensitivity [18]. It is therefore tempting to hypothesize that cortical SSA is the direct neural correlate of deviance sensitivity in midlatency potentials.
There exist a number of models for deviance sensitivity, mostly in the context of the MMN. While some models are based on sensory adaptation [19,20], others postulate an explicit prediction mechanism [21]. These models, however, are not very realistic at the single-neuron and local network level. In models based on sensory adaptation, changes in membrane excitability, which are a prominent mechanism of neural adaptation [22,23], cannot easily explain SSA as they would affect similarly the responses to all stimuli. Therefore, SSA models have been primarily based on synaptic adaptation, which can be stimulus-specific.
SSA may arise from synaptic depression in the thalamocortical (ThC) input to A1, as studied e.g. by Lee and Sherman [24]. Indeed, computational models based on purely feed-forward connectivity show SSA [4,25,26]. However, these models provide predictions that are inconsistent with some of the experimental data [4,12]. Here we study an alternative mechanism: SSA may arise from the heavily recurrent cortical connectivity and the depression of the local, intracortical synapses in A1 [27]. We use a modified version of a recurrent neural network model with synaptic depression [28], which reproduced important properties of A1 responses, including frequency tuning, forward masking, and lateral inhibition. An important feature of this model is the generation of synchronized firing events called population spikes (PSs). The existence of PSs in A1 has support from intracellular [29,30] and multi-unit recordings [31–33], and from network [34] and single-cell calcium imaging studies [35]. We study the basic properties of SSA and its robustness using both simulations and mathematical analysis, and provide predictions for experimental testing of the role of intracortical synaptic depression in cortical SSA.
We model primary auditory cortex (A1) as a neural network with multiple cortical columns (Fig 1A). The crucial property of our model is the use of synaptic depression in all of the excitatory connections, both intracortical and thalamocortical. A similar model has been used successfully to model many properties of auditory cortex [28]. Here we show that this model also exhibits stimulus-specific adaptation (SSA). Our model network often responds to sensory stimulation by generating population spikes (PSs). PSs are events in which a large number of the neurons in a column fire a spike within a few milliseconds, and are characteristic of recurrent networks with synaptic depression [28,36,37]. A PS is generated when there are enough synaptic resources in the network to allow a spike in one neuron to evoke spikes in its post-synaptic targets. It is a regenerative event in which a large fraction of the neurons in the network is rapidly recruited to a brief period of general firing. In a rate model such as the one studied here, PSs are expressed as transient increases in the mean firing-rate within a column (Fig 1B, top). The expected number of spikes fired by each individual neuron may not be large–in fact, it is typically about 1 spike for each PS in the simulations shown here. However, the fact that many neurons fire this spike within a short period of time causes a depletion of the synaptic resources (denoted later by x) within the column, followed by a gradual recovery with a time-constant τrec (Fig 1B, bottom). Consistent with their regenerative nature, PSs are characterized by a sharp threshold (Fig 1C, left). Below threshold, PSs are not evoked. Above threshold, PSs are always evoked, although stronger input amplitude produces PSs with shorter latency and somewhat higher peak firing- rate.
Each column in our model consists of an excitatory and an inhibitory population. The inhibitory neurons do not get a direct thalamic input, contrary to the known physiology. The reason is technical: the behavior of the model does not change much when adding a low amount of direct thalamic input to the inhibitory neurons. However, when adding a substantial amount of thalamic input, as found in experiments, the sensory responses are completely quenched, because rate models cannot easily produce the lag between excitation and inhibition consequent to the additional synaptic delay between inhibitory and excitatory neurons [38]. In order to produce this delay, we kept only the feedback inhibition. Indeed, in this configuration, the firing rate of the inhibitory population, I, lags behind the excitatory population’s firing rate, E (Fig 1B, and cf. also [28]) and acts to both restrain the amplitude of the PS and increase its threshold (Fig 1C, right).
Importantly, once PSs are initiated, they can propagate from one column to the next (Fig 1D). The response latency in the next columns can be tens of milliseconds longer than in the column where the PS originates, as has also been observed both in this model [28] and experimentally [39]. PSs propagate over a substantial part of the model network. The extent of the propagation is limited by the extent of the sensory input, which has to be large enough in order to prime a column to allow PS propagation into it. Specifically, each column in our model corresponds to about 1/3 octave along the tonotopic gradient of rat A1 and sensory input consisting of a single tone activates a region covering about 2 octaves, in line with typical tuning-curve widths in rat A1 at suprathreshold levels (see Materials and Methods; cf. [40]). Under these conditions, the evoked PS typically propagated into 3 columns on each side of the column where it initiated, making the population response more or less co-localized with the thalamic input (as found experimentally e.g. by Li et al. [41]).
Because of synaptic depletion, the recurrent connections in the column are effectively weaker following a PS. This may prevent subsequent stimuli from generating another PS (cf. Fig 1E). The adaptation of columns in our model is fast, with the population response failing to generate another PS already upon the second stimulus. Examples of fast adaptation in in-vivo recordings may be found in [42,43]. Depression in the recurrent connections is thus the basic mechanism of adaptation to repetitive stimulation in this model. Following a PS, the depressed column enters an essentially refractory period, during which the threshold for a subsequent PS is increased (Fig 1F), decreasing slowly back to normal (Fig 1G). During a refractory period, even when a stimulus evokes a response, the PSs are smaller in amplitude than the initial PS. Fig 1H shows how the amplitude of the response to the second stimulus (in spike counts) varies with the input amplitude of the second stimulus and the time since the previous stimulation. Responses qualify as a second PS if their spike count is close to or higher than 1.
It is important to bear in mind that sparse, asynchronous firing can still occur during the period following a PS. Evoked activity is also possible during this period, but the few neurons activated will usually not be able to recruit the network for another PS. This stems from insufficient synaptic resources, as even a small depletion can greatly affect the ability of the network to generate a PS.
SSA is commonly demonstrated using two tones that evoke similar responses in the recording site [1] and presenting them repetitively within an oddball protocol, where one tone occurs with high probability (standard) and the other occurs with low probability (deviant; cf. Fig 2A). Hence, we analyzed SSA in the column midway between the “standard column” and the “deviant column” (the columns whose best frequencies were the standard and deviant frequencies, respectively), referred to as the “middle column”. As evident from Fig 2A, SSA occurred in the model because standard tones often could not produce PSs in the middle column, yet the deviant tones could.
SSA in this model looks somewhat different when considered from the point of view of the single neuron (averaging many single-trial responses) and from the point of view of the single-trial network activity (averaging across many neurons), a dissociation that has been studied experimentally in auditory cortex (e.g. [32]). Fig 2A shows the firing-rates of selected neurons in a single block of the oddball protocol consisting of 100 stimuli. In each single trial, neurons showed one of two response types. The first was a small transient increase in firing-rate (Fig 2B, left, showing the responses of all neurons in trial no. 22). The other is the signature of a PS, a concerted sharp increase in firing-rate (Fig 2B, right, showing the responses of all neurons to trial no. 24) during which each neuron fired on average about one spike. Population spikes occurred mostly (but not exclusively) in response to deviant tones, as will be discussed in more detail below.
The responses of single neurons across many presentations of the same stimulus could show substantial variability. The responses of a neuron to the standard tones (e.g. neuron no. 75. Fig 2D, top, shows responses from 500 trials, including those shown in Fig 2A) mostly consisted of small increases in firing-rate but included also a few large responses that occurred when standards did evoke a PS (light blue traces in Fig 2C, left; average response in blue). The responses of the same neuron to the deviant tones were mostly large, although they also included a few small responses, corresponding to failures of a deviant to evoke a PS (e.g. trial no. 4 in Fig 2A; Fig 2C, right, light red traces; average response in red).
In conclusion, sensory responses may occur in a single neuron whether or not a population spike occurred. The average response of a neuron to both standards and deviants is composed of trials with PSs, with a large and consistent response across neurons, and trials with no PSs. The difference between standard and deviant responses has to do with the probability of such events: PSs are likely to occur in response to deviant tones, with a few failures. On the other hand, standard responses show a high probability of failures, but the successful PS responses to the standard are similar in their magnitude to those evoked by the deviant. This situation is also found in experimental results [12,43].
To illustrate the mechanisms of SSA in the model, Fig 3A–3C show the firing-rates and time-course of synaptic resources not only in the middle column but also in its neighbors. PSs left a strong depletion of resources in their wake (Fig 3C). Both the deviant and standard tones initiated PSs in their respective columns. However, standard tones initiated PSs much less often in the standard column, and even those that were initiated mostly failed to propagate into the middle column. Importantly, as discussed above, even when a stimulus failed to evoke a PS, sensory responses to that stimulus still occurred at least in some of the neurons in the column.
PSs initiated in the standard column were of two types. Occasionally a PS could be evoked in the standard column during a train of standard tones. This occurred due to gradual recovery of resources from one tone to the next (Fig 3D and 3E, left panels). Such PSs failed to propagate outside the standard column because of their small amplitude. The other type consisted of PSs evoked in the standard column by the first standard following a deviant (Fig 3D, right panel). These PSs occurred because the deviant tone did not deplete the resources in the standard column as much as another standard tone would have done (Fig 3E, right panel). This was primarily due to the adaptation in the standard column, which prevented the PS evoked by the deviant from propagating into it, resulting in little depletion of resources during the presentation of the deviant. In consequence, following the presentation of a deviant, the available resources in the standard column were higher than following the presentation of a standard tone, sometimes allowing PS generation in response to the next standard tone. However, in these circumstances the propagation of PSs from the standard column into the middle column generally failed, due to depletion of the resources in the middle column by the recent deviant-evoked PS. Also, although somewhat larger than the PSs of the first type, PSs of the second type were still of relatively small amplitude and thus limited in their ability to propagate away from the column in which they were initiated (cf. Fig 3D, right, red and blue traces, and also Fig 3B within the green frame for the different ranges of propagation of PSs evoked in the standard and deviant columns).
SSA in the model was not related to preference of the middle column for one tone or another. This was verified by presenting the two tones with equal probabilities of occurrence, as well as switching their roles as standard and deviant (Fig 3F, top row, displays the average responses of these three conditions). Each tone evoked an average response depending mostly on its probability. When the two tones were presented with a probability of 50%, both evoked similar average response. These results reproduce the phenomenon seen in electrophysiological recordings, where responses depended on tone probability (Fig 3F, bottom row; data from Taaseh et al. [4]): Responses to a tone in the deviant condition were stronger than responses to the same tone in the equal condition, and those were in turn stronger than responses to the same tone when standard.
True deviance sensitivity requires more than just larger responses to rare stimuli: it is necessary that responses to deviant tones depend also on the identity of the other stimuli in the sequence. The extent of deviance sensitivity in our model was assessed using control protocols similar to those used in actual experiments [4,12]. These protocols were adapted from human studies [e.g. 44], and included two types of sequences (see Fig 4A and the Materials and Methods section):
These three controls, together with the oddball and Equal protocols, gave rise to 6 conditions in which tones were tested (Fig 4A). Since our model includes some random heterogeneity in the tuning-curves within each column, we ran all the protocols on 12 networks with different randomizations of the tuning curves. The Common-contrast SSA Indexes (CSIs, defined in Eq 9) of the single neurons within the middle column of each network were all positive (Fig 4B, left). Their distribution was bimodal, related to the background input received by each neuron (see the Materials and Methods section for full description of the background inputs). Neurons that received strong, positive background input tended to respond strongly to standard tones and accordingly had relatively low CSI. Neurons that received negative background input and no sensory input had almost no standard responses but participated in the population spikes evoked by deviants, and accordingly had a high CSI. The CSI calculated from the mean firing-rate responses, averaged over all the neurons of the middle column, was slightly different from the mean single-neuron CSI (CSI = 0.632 based on the mean firing-rate vs. 0.679 ± 0.204 from single-neuron responses, marked by an arrow and a gray line in Fig 4B, right, respectively). Throughout the paper we calculate the CSI from the mean responses of all neurons in a column (rather than calculate CSIs of each neuron and then average). A histogram of the mean firing-rate CSI is shown in Fig 4B, right. Importantly, this histogram is very narrow (CSI = 0.643 ± 0.007, mean ± standard deviation). Our 12 networks therefore showed a similar level of SSA, despite the different randomizations of tuning-curves and the wide CSI distribution in the individual neurons of the middle column of each network.
Average responses of the middle column to tones in all 6 conditions are plotted in Fig 4C. An example of multi-unit activity (MUA) responses in rat auditory cortex to tones presented in these conditions is displayed in Fig 4D for comparison. These data were collected by Taaseh et al. [4], and are representative of the responses of both multiunit clusters and single neurons in rat A1, as shown by Hershenhoren et al. [12].
In the model, as in the experimental data, three conditions tended to give rise to relatively small responses: Standard, Equal and Diverse Narrow. In these conditions, the tones had high probability (Standard and Equal) or were packed in a narrow frequency band, causing substantial cross-frequency adaptation (Diverse Narrow). Three conditions evoked large responses: the Deviant Alone condition evoked the largest responses, as expected; and most importantly, responses to the Deviant condition were larger than those evoked by the same tone presented as part of the Diverse Broad sequence (cf. scatters in Fig 4E; paired t-test: Deviant > Diverse Broad, t = 6.4, df = 11, p = 5.3x10-5 for f1 and t = 4.9, df = 11, p = 5.1x10-4 for f2). Thus, the model passes the accepted test for true deviance sensitivity. Significantly, these simulation results are in line with the findings in rat auditory cortex by Taaseh et al. [4] (cf. Fig 4D) and Hershenhoren et al. [12], who showed this pattern of responses in about half of their neurons.
Why were the responses in the Deviant condition larger than in the Diverse Broad condition? This requires consideration of the responses in the deviant column in these two conditions. In the Deviant condition, the standard tone provided some weak input to the deviant column, causing some adaptation and reducing the average deviant response relative to the Deviant Alone condition. However, since most standard presentations did not evoke PSs in the deviant column, the adaptation caused in this column by standard tones remained small (cf. the response to deviant tones in Fig 3A relative to the response to the first standard tone, which should be similar to the Deviant Alone condition for f2). In contrast, in the Diverse Broad condition the probability of a tone to evoke a PS in its column was much higher, due to the wide distribution of tone frequencies in the sequence (Fig 4F–4H). These PSs propagated across the network according to its recent history, and some reached the deviant column and produced PSs in it. In consequence, PSs occurred in the deviant column at a rate that was higher than during an oddball protocol. This left the deviant column generally more adapted in the Diverse Broad condition than in the oddball protocol, and less able to generate responses that could propagate into the middle column. As an example, notice the PS generated in column 14, just before time t = 5 s in Fig 4G. This PS propagated into the deviant column and left it too adapted, such that it could not generate a PS in response to the next presentation of its own best frequency, f2 = 12, two stimuli later. Based on our results we conclude that the regularity of standard presentations in the oddball protocol effectively resulted in less adaptation in the deviant column compared to the Diverse Broad condition, and consequently the deviant responses in the middle column were stronger than Diverse Broad responses.
Our model reproduces the experimental dependences of SSA on several parameters: The probability of deviant occurrence, P; the frequency separation, Δf (defined here as Δf = f2 –f1); inter-stimulus interval (ISI); and input amplitude A.
Smaller probabilities of deviant occurrence produced larger firing-rate responses to the deviant tone (Fig 5A). This resulted in increased CSI, as in Ulanovsky et al. [1]. In addition, SSA was present only within a limited range of input amplitudes (Fig 5B and 5C, left-hand panels), which was about 60% wider at the lower deviant probability (Fig 5B, left). As the firing-rates in thalamocortical fibers are related to sound intensity, the decreased SSA with increased input amplitude in the model may correspond to the experimental findings of Taaseh [45], who reported a decrease in CSI for increased sound level. Their findings, based on LFP recordings in rat auditory cortex, are re-plotted here in Fig 5D (left). A general decrease in CSI for increased sound level was also reported recently by Nieto-Diego and Malmierca [42], who recorded MUA responses across different fields of rat auditory cortex.
SSA was non-monotonic not only as a function of input amplitude, but also as a function of inter-stimulus interval (Fig 5B and 5C, right-hand panels). Here, the range of inter-stimulus intervals allowing SSA was about 16% wider at lower deviant probabilities (Fig 5B, right). This is also in line with experimental results [1,4,12]; an example for the experimental dependence of CSI on ISI is provided in Fig 5D (right), which re-plots the results of intracellular recordings by Hershenhoren et al. [12].
Increasing the frequency separation between standard and deviant, Δf, usually produced an increase in CSI, extending the ranges of input amplitudes and inter-stimulus intervals that allowed SSA (Fig 5C). In general, SSA was present only for rather short ISIs (< 1 s), except for the case of Δf = 6 that showed SSA for ISIs up to nearly 2 s, similar to the results of Ulanovsky et al. [1] at their largest Δf, as well as Hershenhoren et al. [12]. The effect of increasing the frequency separation on CSI in columns other than the middle column is presented below, as part of the experimental predictions of the model.
Our model showed hyperacuity, which is another important feature of SSA in A1 [1]: SSA was present for Δf values substantially smaller than the width of the tuning curve. We first tested for hyperacuity by choosing f1 and f2 such that Δf was 10 times smaller than the input tuning-curve width. This setting put the two tones near the peak response frequency of the middle column, so that this column was adapted by both the standard and the deviant. Consequently, it couldn’t support the initiation of population spikes and did not show any SSA (Fig 5E, light gray trace). This result emphasizes the fact that SSA in our model depends on the propagation of PSs between columns, rather than depression of thalamocortical synapses in the middle columns. As an alternative method for testing hyperacuity, we increased the frequency resolution of the model. This was achieved by increasing the width of the input tuning-curves. For these simulations, the width of the input tuning curve was set to 20 times the frequency difference between two nearby columns. In consequence, the distance between the two columns used for standard and deviant, Δf = 2, corresponded to 1/10 of the thalamocortical tuning-curve width. Such a setting corresponds to finer columnar organization, i.e. shorter-distance connections within the cortex. SSA was present in this paradigm, albeit with lower CSI values and in a narrower range of ISIs compared to the standard network (Fig 5E, dark gray vs. black traces).
SSA is a non-monotonic function of most parameters of the simulation (cf. Fig 5B and 5C). This property is due to the patterns of PSs evoked by the two tones in the oddball protocol. We identified four different regimes of PS initiation: (i) No stimulus was able to evoke a PS. This “No PS” regime occurred for low A values, when no stimulus was strong enough to elicit a PS (Fig 6A), as well as for short ISIs when synaptic resources were too depleted for PS initiation and propagation. It was also found at small values of U, the fraction of resources used upon synaptic activation, which effectively reduced synaptic strength, and for long recovery time-constants of the synaptic resources, which kept the synaptic weights dynamically small. This regime showed no SSA, and was named “No PS” since it was the only regime that had no PSs, except a single PS that sometimes occurred at the beginning of a protocol. (ii) Deviant stimuli evoked PSs in the middle column whereas standards did not. This “Selective” regime (Fig 6B) is the one that showed strong SSA, as illustrated in Fig 3. (iii) A Periodic regime, in which the standard tone evoked PSs in its column, and these invaded the middle column only once every few stimuli (Fig 6C). PS responses to the deviant were not always successful because the middle column, and sometimes the deviant column as well, were adapted by the PSs initiated in the standard column. (iv) Each and every stimulus evoked a PS, regardless of its identity (Fig 6D). This “Reliable” regime occurred for example at high input amplitudes and long ISIs. Regimes (ii) and (iii) are similar to the phenomenon of cycle skipping in excitable systems, where periodic excitation gives rise to responses in some but not all cycles.
Fig 6E–6H illustrate the association of strong SSA with the Selective regime. CSI is plotted against selected parameters, as in Fig 5, and the different response patterns, identified from the simulation traces, are superimposed in color. In addition to the stimulus parameters, A and ISI (Fig 6E and 6F, black curves), the same regimes and the same dependence of CSI on parameters occurred when varying dynamical parameters of the local intracortical connections such as U, the fraction of synapse utilization, and τrec, the recovery time-constant of synaptic resources. Very short τrec gave rise to yet another pattern of activity, in which PSs were generated spontaneously. CSI was generally high only within the Selective regime, although there was also moderate SSA on the margins of the Periodic regime.
As mentioned above, one of the modifications we made to the model of [28] was to introduce heterogeneity in the tuning curves within each column, following experimental findings in mouse A1 [46,47]. We use the sensitivity of SSA to stimulus parameters to assess the effect of this feature of the network (Fig 6E and 6F). The same parameters were tested for a heterogeneous network, as was used throughout this work (black) and for one with homogeneous columns, where all the input-receiving neurons in each column have the same best frequency (gray; the regimes marked in color on the plot area are for the heterogeneous network). In the homogeneous network, SSA existed in a similar range of amplitudes and in a range of ISIs that included shorter values compared to the heterogeneous network. Thus, the basic mechanism of SSA was unaffected by the heterogeneity, which primarily contributed to the existence of SSA at longer ISIs.
It seems therefore that SSA exists in this model only within a relatively narrow range of parameters. To verify this claim generally, we searched exhaustively through the space of possible parameters of the stimulus sequence (amplitude and ISI) and through the space of possible network parameters (U and τrec). Fig 7A presents the CSI for different combinations of A and ISI, with the different response regimes delineated on the map. Strong SSA is mostly confined to the Selective regime. The existence region of SSA has the shape of a narrow, curved band on the A-ISI plane. It consists of two “branches”, one spanning a wide range of ISIs but confined to low input amplitudes, and the other confined to short ISIs and spanning a wide range of input amplitudes. Fig 7B shows the CSI map and the regimes obtained in a homogeneous network, where all inputs to a column share the same best frequency. Comparison with Fig 7A reveals that network heterogeneity resulted in a slightly expanded low-amplitude, long ISI branch of the existence region for SSA.
We studied the existence region for other parameters as well. For example, when keeping the stimulus parameters fixed (A = 5 Spikes/s, ISI = 350 ms) and varying the synaptic parameters U and τrec (Fig 7C), SSA was also confined to a narrow band of parameters (Fig 7C). These calculations extend the results of Fig 6G and 6H. Here, the existence region for SSA was somewhat wider when both U and τrec were large.
The existence region found for SSA in our simulations can be viewed as a consequence of the difference between responses of the deviant and standard columns, which are affected differentially by adaptation. SSA exists when the slightly-adapted deviant column generates PSs while the strongly-adapted standard column doesn’t. Consider the responses of the deviant and standard columns as a function of any of the parameters of the model. Such a relationship is called here the ‘response function’. Adaptation modifies the response function, and its net effect can be heuristically modeled by a modification of the relevant model parameter. For example, adaptation as a function of sound level can be described by effectively reducing the sound level of the input. Operations on the input of the response function fall into two major classes [48], illustrated in Fig 8A: (i) Subtraction, which shifts the response function along the parameter axis while preserving its slope (left); and (ii) division, which scales the slope of the response curve (right).
In Fig 8B, the responses of the middle column, which is where the SSA was evaluated, to standard and deviant stimuli (blue and red) are displayed as a function of log(ISI) (left) and log(A) (right). These responses are largely inherited from the standard and deviant columns (where the PSs are generated; light blue and light red), although responses to deviants were attenuated more strongly in the middle column (red vs. light red) compared to the responses to standards (blue vs. light blue). The responses to standards showed a sigmoidal, monotonic dependence on both parameters. The response curves to the deviants were shifted to smaller values, and also somewhat distorted: for both parameters, there was a non-monotonic region corresponding to mid-range standard responses. Comparing these results of the simulation with the schematic curves in Fig 8A, we see that adaptation acts as a subtractive operation along the logarithmic ISI scale, and at least within the low-A monotonic region of the deviant responses, it is a divisive operation along the log(A) scale (Fig 8B).
While the divisive operation on stimulus amplitude doesn’t have an obvious explanation, the subtractive effect along the log(ISI) scale is easily explained by considering the effective ISI. On average, this was ISI/(1 – P) in the standard column, where P is the probability of occurrence of the deviant tone, but ISI/P in the deviant column. On the logarithmic scale, these become shifts of –log(1 – P) and –log(P) with respect to a column stimulated repetitively (P = 1). Interestingly, these shifts are the expressions for the objective surprise associated with the standard and deviant tone, respectively. Thus, at least under these conditions, the model roughly implements a predictive coder [49].
Since the existence region for SSA is rather narrow, we were interested in mechanisms that may increase its extent. We show here that the depression of the thalamocortical synapse is one such mechanism, whose inclusion in the model increases the range of parameters producing large CSI (Fig 7D).
Changing Us, the fraction of resource utilization in the ThC synapses, produced a pattern similar to changing A or U: Low values gave rise to the No-PS regime and higher values to the Selective and Periodic regimes (Fig 9A). In contrast, changing the value of τsrec, the time-constant of recovery of the ThC synapses, did not abolish the Selective PS responses (Fig 9B). The only effect of increasing τsrec was a slight increase in CSI. Robustness to changes in τsrec is highlighted by the fact that tested values spanned over two orders of magnitude.
The dynamics of ThC depression in our model (Eq 5 in the Materials and Methods section) can be treated analytically. When presenting a sequence of identical, repetitive stimuli, the mean fraction of available resources in the ThC synapses (denoted by z) is periodic with a period that equals the inter-stimulus interval (Fig 9C, bottom). This is approximately what happens during a long sequence of standard stimuli. Importantly, treating the system as an iterated map reveals that z assumes the same value at the onset of each stimulus (onset z value) and another value at the offset of each stimulus (offset z value). These values can be derived analytically under some approximations (Eqs 17 and 18; resulting values shown as dashed lines in Fig 9C, bottom). Importantly, the analytical derivation further reveals that this is the only steady-state behavior of the map, so that the resources of the ThC synapses cannot show oscillations with longer periods.
Thus, ThC depression cannot account for the more interesting dynamical phenomena in the network. For example, in Fig 9C the population spikes within the standard column occur once in every three stimulus cycles, while the dynamical parameters of the ThC depression only follow the faster periodicity of a single stimulus cycle. Similarly, ThC depression alone cannot produce the phenomenon we observe in the Periodic regime, in which PSs may occur in the middle column on some but not all stimulus presentations (Fig 6C). In both cases, Periodic PS responses to trains of standard stimuli resulted from the dynamics of the synaptic resources of the intracortical connections (denoted respectively by x and y). Indeed, while z follows the rate of stimulus presentation, x and y generally follow the slower rate of the population spikes.
We use the onset and offset z calculated analytically (Eqs 17 and 18) to examine the effects of ThC depression within the SSA Region. The SSA Region found in our simulations consists of two “branches”, in which ThC depression plays different roles (Fig 9D).
The first branch, at low stimulus amplitude and spanning a wide range of ISIs, has high values of both offset and onset z. The equations show that the low-level stimuli that gave rise to this branch in the simulations hardly deplete ThC resources, and the moderate ISIs allow recovery of virtually all the depleted resources. Thus, ThC depression has little effect on network responses in this branch, explaining the robustness of SSA with respect to τsrec found in our simulations (Fig 9A).
The second branch lies in the short-ISI range and spans a wide range of stimulus amplitudes. In this branch, the equations show that stimuli cause significant ThC depletion (low offset z) and little resource recovery (low onset z). This combination causes a strong effective attenuation of stimulus amplitude, explaining why in the simulations the CSI values were quite uniform (Fig 7A)–high-amplitude stimulation was effectively attenuated to input strength that was comparable to that of low-amplitude stimulation. The analytical treatment suggests therefore that this branch is formed by “stretching” the short-ISI, medium-amplitude region into the high-amplitude range. We note that both branches of the SSA Region have a similar, rather negligible amount of resources recovered during the ISI (Fig 9D, right-hand plot).
To summarize, the main effect of thalamocortical depression in our model is to weaken the dependence of CSI on some model parameters. This is illustrated in Fig 9 for input amplitude–at least at short enough ISIs, ThC depression adaptively reduces stimulus amplitude and therefore brings the activity pattern into the selective regime. The same analysis is also true for the effect of the fraction of resource utilization in the thalamocortical synapses (Us) on SSA, which was found to resemble that of stimulus amplitude (cf. Figs 6E and 9A): At the low-level, long-ISI region Us indeed strongly affected SSA. However, within the short-ISI branch, SSA should be only weakly dependent on Us, which effectively acts as a modifier of stimulus amplitude.
Thus, ThC depression has an important role in shaping the SSA Region itself, expanding the parameter region formed by intracortical depression alone. This finding was confirmed by our simulations of the same network as the one used for Fig 7A, except that ThC depression was removed. We tested this network on the same range of stimulus parameters (Fig 7D): The Selective regime without ThC depression was narrower than that of the standard network, and while it showed strong SSA in the low-A branch, SSA was weak and existed for a more limited set of parameters on the short-ISI branch.
Our simulations of a network with multiple columns allowed us to study model responses to the oddball protocol in columns other than the middle column. For these columns, the two tones presented were usually located on the same side of the peak of the tuning curve (Fig 10A and 10B).
We found that in columns whose best frequency corresponded exactly to either the standard or the deviant, SSA was considerably weaker than in the middle column. This is because adaptation to the best frequency as the standard prevented PS responses to deviants from entering the column, while responses to the best frequency when it served as deviant were attenuated by cross-frequency adaptation. It should be noted that when Δf is smaller than 2, the smallest value shown in Fig 10B, the middle column shows weak SSA as well (cf. Fig 5E).
In contrast, in columns with a best frequency either lower than f1 or higher than f2, SSA could be even stronger than in the middle column. This resulted from large deviant responses in one of the protocols. For example, columns whose best frequency was lower than the low frequency of the oddball sequence (f1) showed large responses to f1 as deviant but not to f2 as deviant. Responses to f2 as deviant in these columns were small because PSs evoked in the f2 column had to travel through the strongly adapted f1 column before reaching the low-best frequency columns. The CSI in some of the low-best frequency columns was even higher than in the middle column because deviant responses to f1 suffered less cross-frequency adaptation from the standard, which was farther away on the frequency scale. The opposite occurred on the other side of the network.
The results described here correspond to playing an oddball protocol with f1 and f2 not centered on the best frequency of the recording site. Such a paradigm has not been systematically studied in cortical recordings.
The model makes clear predictions for the effect of tone duration on SSA. Fig 10C presents a map of CSI values as a function of tone duration and the offset-to-onset interval. The latter quantity is highly relevant for the dynamical behavior of the network, as it is the time allowed for recovery of synaptic resources. High CSI values in Fig 10C are limited to shorter tone durations and offset-to-onset intervals, and SSA was always abolished at long tone durations. However, the reason that SSA disappeared at long tone durations depended on the offset-to-onset interval: At long offset-to-onset intervals, longer tone durations produced more frequent PS responses (Periodic and Reliable regimes). This was due to the ongoing recovery of intracortical resources, showing that thalamocortical depression did not play an important role at these intervals (cf. the high onset z for long ISIs in Fig 9D). In contrast, at short offset-to-onset intervals, SSA was abolished at long durations because no PSs were elicited. We attribute this to the accumulating depletion of thalamocortical resources from one stimulus to the next, which rendered them too low for eliciting PS responses even in the deviant column.
We note that at intervals around 300 ms, used in most of the simulations above, SSA was quite robust to changes in duration but not to changes in the interval. Indeed, Hershenhoren et al. [12] recently reported that in rats, while CSI is significant for tone duration of 30 ms and offset-to-onset intervals of 270 ms, it becomes rather weak at intervals of 670 ms and 1170 ms (cf. Fig 5D, right).
We showed here that a network with recurrent dynamics and synaptic depression is capable of producing stimulus-specific adaptation. In particular, the network demonstrates true deviance sensitivity, in contrast with competing models based on feedforward synaptic depression [4,25] but in accordance with experimental results [4,12].
SSA in our model is first and foremost a network phenomenon that does not require thalamocortical depression. It depends on intrinsic cortical dynamics that produce population spikes and allow their lateral propagation across the network. The existence of population spikes in A1 has been inferred from the distribution of EPSP amplitudes [29,30], multi-electrode recordings [31–33], and network [34] and single-cell calcium transients [35]. In-vitro recordings also show population events occurring in A1 upon stimulation of the thalamus [50]. The evidence suggests that large ensembles of neurons may be active during such events: Bathelier et al. [35], for example, used 2-photon calcium imaging with single-cell resolution and found that the population events dominating the activity in A1 included a large fraction (over half) of the imaged neurons.
The PSs in our model are of slightly faster rise-time and shorter duration relative to evoked population bursts recorded in rat A1 [31–33]. While the underlying mechanisms have not been identified, modeling work [36,37] has suggested that population spikes may be a consequence of the dynamics of a network with depressing synapses. Loebel et al. [28] further demonstrated that such networks can be used to provide a unified account for many response properties of neurons in auditory cortex, including forward masking, lateral inhibition and hypersensitive locking suppression. Here we showed that such a network can produce SSA as well.
The model we presented here, similar to [28], should be taken as a simplified representation of A1. The use of a rate-model to represent single neurons in A1 was introduced by Loebel et al. [28], based on a previous work that showed the equivalence of such a model to a firing-rate description of a recurrent network with synaptic depression in terms of its activity [36]. The model is stripped of many details such as layers, feedforward inhibition, cell types and numbers, accurate connectivity probabilities, or realistic connectivity profiles. The simplifications made it possible to focus on a minimal number of basic features and demonstrate that they are sufficient to support cortical SSA. Nevertheless, the model keeps some crucial features of the actual rodent auditory cortex.
In cortex, connection strengths and probabilities decay smoothly with spatial distance [27]. In the model, this organization was discretized into columns. In rat A1, the tonotopic gradient spans about 2 mm and covers about 7 octaves [40]; in mouse A1 the tonotopic axis is somewhat shorter. Since the model had 21 columns, each corresponded to about 1/3-1/2 octave. Tuning curves in the model have a half-maximum width of 5 columns, or 1–2 octaves. This is similar to curves measured in cortical recordings [40]. Based on the above numbers, the usual frequency separation of our simulations (Δf = 2) corresponds to about 2/3 of an octave, or slightly above the typical frequency separation used in experiments, which is 0.53 octave (the 44% of [4,12] and 0.37 of [1]). Thus, the main conditions of the simulations reflected the typical experimental conditions used to study SSA.
The connection strengths within each column were set to prevent spontaneous PS activity (see Materials and Methods). While it is difficult to compare between synaptic strengths in a rate-model and electrophysiologically-measured quantities, our tests of single-neuron activation in the model showed that connections between excitatory neurons within the same column were weaker than those reported in-vitro in mouse A1. On the other hand, connections were more probable [27], since our network had full connectivity within each column (see Materials and Methods). Synaptic depression parameters used here are on the same order as those reported in A1 and used in modeling its activity (U = 0.55, τrec = 500 ms in [51], Us = 0.8, τsrec = 1 s in [52]).
Finally, following [28], our model features purely feedback inhibition, where inhibitory neurons do not receive sensory input. Feedforward inhibition is known to have an important role in A1, arriving at a typical delay of a few milliseconds following the excitatory (presumably thalamocortical) currents [38], presumably due to the additional synapse on the way. Adding an explicit delay would significantly complicate our rate model (see e.g. [53] for an implementation of such a delay). We chose not to implement a synaptic delay, instead emulating the effect of delayed inhibition by feedback inhibition. Indeed, in simulations in which inhibitory neurons also received sensory input, the resulting responses were similar to those of the main simulations as long as the direct sensory input to inhibitory neurons was relatively weak. With stronger inputs to the inhibitory neurons, responses were abolished due to the simultaneous arrival of excitatory and inhibitory inputs onto the neurons in the network.
The absence of direct feedforward inhibition should not affect our conclusions regarding the role of synaptic depression in shaping SSA, based on the following reasons: (i) The first evoked spikes in the thalamo-recipient neurons are already sufficient to trigger the population spike, with feedforward inhbition having its effect only following these initial spikes [38]. Furthermore, once a population spike is triggered, its propagation involves feedforward inhibition, since the intercolumnar connections target both excitatory and inhibitory neurons. Therefore, feedforward inhibition should not prevent the initiation of population spikes. (ii) Synaptic depression would come into play once the first spike is fired, and is the main factor that prevents spiking over the longer time-scale, i.e. for inter-stimulus intervals longer than 100 ms [54]. (iii) The feedforward inhibition itself undergoes SSA, thus scaling with the excitatory responses and at most modulating the SSA, not generating it [55]. Some insight into the effects of feedforward inhibition on SSA may perhaps be gained from the model of Schiff and Reyes [52]. Their results suggest that at the low stimulation rates used in the oddball protocol, feedforward inhibition may balance the effects of thalamocortical depression and result in a relatively constant net drive to the excitatory neurons. Such an effect would emphasize the role of depression of the recurrent synapses in generating SSA (cf. Fig 7D, which shows the extent of the SSA region without thalamocortical depression).
The main weakness of our model is its sensitivity to stimulus and network parameters (Figs 6, 7 and 9A). The existence range of SSA in parameter space is expanded by mechanisms such as depression of the thalamocortical synapses and, to a lesser extent, heterogeneity of the tuning curves. However, even with the action of the above mechanisms, SSA in our model is not a robust phenomenon. The sensitivity of SSA to network and stimulation parameters stems from its existence in the intersection region of the requirements of PS response to the deviant tone but no PS response to the standard tone (as illustrated by the response curves in Fig 9). The extent of this region is determined by the constraints posed by the frequency-tuning of columns and by the lateral propagation: Propagation should be strong enough to deliver deviant responses into the middle column, yet weak enough to prevent most of the successful standard responses from doing so.
For the model to be a viable account of cortical SSA it is necessary to assume the existence of tight regulation of the cortical network, which drives it into the SSA Region. One possible locus of such regulation may be the narrow dynamic range of firing-rates in thalamocortical fibers [9] and/or thalamocortical EPSPs in layer-4 input neurons [56], which in our model would both correspond to a limited range of input amplitudes. Sources of regulation might include neuromodulatory control, a mechanism that could conceivably drive A1 between states favoring SSA and states favoring other computational tasks. Cholinergic input, for example, may control important parameters such as the utilization parameters for sensory and recurrent synapses [57–59].
Two other classes of models for SSA have been studied previously. The first class of models is based on depression of the feedforward synapses [4,12,25,26]. Thalamocortical depression was not essential for the existence of SSA in our model, so that the mechanism studied here is indeed different from the SSA that depends on feedforward depression. The second class of models consists of networks with recurrent connections. The emergence of SSA from synaptic depression in the local, cortical connections has also been shown in a recent computational work by May et al. [20]. Their model is different from ours in that they ignored the single neurons, modeling the interaction between cortical columns. Their analysis is not detailed enough to understand whether the basic mechanisms that produce SSA in their model are the same as in the model presented here. Furthermore, they provide only limited information on the dependence of SSA on stimulus and network parameters. A different network model of SSA-like responses has been developed by Wacongne et al. [21]. Their model depends on synaptic dynamics, although not on synaptic depression. However, their model has been developed in order to test a specific model of predictive coding, and some parts of it (e.g. the short-term memory component) are hand-crafted to have the required properties. Thus, it is different from the network model we describe here, which uses generic network mechanisms to produce SSA. In consequence, all further comparison with existing models considers only feedforward models of SSA.
Our model correctly accounts for a number of properties of SSA in rat auditory cortex. SSA increases with frequency separation between the tones and with decreasing probability of the deviant [4,12]. SSA decreases at high sound levels [45]. Similarly, SSA weakens as ISI is increased [4,12]. The model suggests that SSA should also decrease at low amplitudes and very short ISIs. The low-amplitude and short-ISI ranges are presumably overlooked in experimental studies because of the difficulty to elicit responses with such parameters. All of these properties are also correctly predicted by other models of SSA [25,26]. On the other hand, in the model presented here, single-trial responses to standard tones show occasional successes, consistent with a recent intracellular study of SSA [12], where the largest responses to the standard tone were as large as those in the Deviant and Deviant Alone conditions. The existence of occasional strong responses to the standard tone may also find support in the spike-count distributions in [43], where successful standard responses were similar in strength to the deviant responses. Feedforward models cannot easily account for such findings.
Importantly, the model has true deviance sensitivity, in which responses to deviants within a regular background were stronger than to deviants within an irregular background (“many standards”, Jacobsen and Schroger [44]). This phenomenon could not be explained by a feedforward model of adaptation in narrow frequency channels [4], at least not when frequencies were close enough to produce cross-frequency adaptation. Mill et al. [25] also showed a small preference for true deviants over the “deviant-among-many-standards”, but only at large frequency separations (equivalent to separations larger than our Δf = 2). At these frequency separations, the deviant and standard were far enough apart that the standard in the oddball sequence did not induce strong cross-frequency adaptation of the deviant responses, while the many-standards condition did [25]. In contrast, in our model the many-standards (our Diverse Broad) condition generically gives rise to cross-frequency adaptation that is stronger than in the oddball sequence. The many-standards sequence, with its extensive propagation of population spikes across the network, caused more activity in the deviant column than the oddball sequence, and accordingly also more adaptation. Consequently, the deviant column was able to produce stronger responses to its best-frequency tones during the oddball sequence than during the many-standards sequence. Neuronal responses in rat auditory cortex indeed often show true deviance sensitivity [12].
In addition to modeling existing data, the model makes predictions that have not yet been tested in cortex. In our model, the CSI depends on the positions of the two tones relative to the column’s best frequency, with a minimum CSI when one of the tones is equal to the best frequency. To our best knowledge, this prediction has not been tested systematically in cortical recordings. Interestingly, such conditions have been tested in the inferior colliculus by Duque et al. [60]. Their results are qualitatively different from our prediction for cortical SSA in that they do not show a decreased CSI for f1 or f2 equal to the best frequency; rather, they are consistent with purely feedforward depression and the characteristic asymmetry of tuning curves in the auditory system. Bäuerle et al. [11] have also looked at SSA with frequency pairs not centered on the site’s best frequency, in gerbil ventral MGB. Most of these results favor the feedforward depression mechanism. Thus, in subcortical stations, feedforward depression may be the dominant mechanism shaping SSA.
Our model predicts that SSA should be found within a limited range of stimulus durations (cf. Fig 10C, the longer-duration “branch”). Significantly, our prediction contradicts that of a model based on feedforward depression [26], which suggests that SSA should become stronger when stimulus duration is extended. We know of no study that checked systematically the dependence of SSA on stimulus duration. However, there is a tendency to use relatively short stimuli in SSA experiments in rats (30 ms, [3,4,12]; 75 ms [5,9,60]), suggesting that indeed longer stimuli result in less SSA.
SSA has been demonstrated in somatosensory and in visual cortices. Mechanisms similar to those studied here may be at play in the generation of SSA in barrel cortex. In rat barrel cortex, SSA has been reported for whisker identity as well as for velocity and direction of whisker deflection [61]. The responses to non-principal whiskers are probably mediated by lateral connections [62]. The similar amplitudes of deviant and ‘many-standards' early responses in barrel cortex [61] are better explained by adaptation of intracortical synapses, as in our model, rather than adaptation in the thalamocortical synapses only [4], since the latter mechanism predicts higher responses in the ‘many-standards’ condition (see above for comparison of these two types of models).
In visual cortex, studies in cats [63] and monkeys [64] have found that complex cells adapt to a specific orientation while generally retaining their responses to other orientations. Such cells, therefore, show SSA to orientation. In contrast with auditory and somatosensory cortices, in visual cortex this phenomenon can be explained by a purely feedforward mechanism, with adaptation in simple cells (e.g. due to fatigue) shaping the tuning curves of complex cells in a stimulus-specific way (as suggested in [64]). Such a mechanism would be in line with the “cascading” adaptation characteristic of the visual system, e.g. in the way spatial adaptation in V1 appears to arise from integration of LGN responses [65] and motion adaptation in MT can be explained by broadly-tuned adaptation of their input V1 neurons [66,67]. Recently, true deviance sensitivity has been demonstrated in visual cortex using both extracellular recordings and calcium imaging, but the difference between deviant and many-standards responses occurs late [68]. This difference may challenge the purely feedforward explanation of adaptation in visual cortex, even if only indicating a late top-down modulation. We note that consideration of recurrent connectivity as essential to explaining V1 phenomena was emphasized in recent modeling work by Chariker et al. [69], although not in the context of adaptation.
Importantly, the mechanism generating SSA in our model generalizes to any stimulus or modality, if the tonotopic columns are replaced with populations coding for different stimuli or features. These populations need not be spatially segregated, but should have (i) stronger intra- vs. inter-population connection strengths and (ii) different sensitivities to afferent inputs. Thus, activity-dependent synaptic depression combined with heterogeneity in synaptic strengths leads to SSA for any stimuli encoded by the connectivity. In higher order areas of the auditory system, this can lead to SSA for complex stimuli.
We model primary auditory cortex (A1), using multiple cortical columns (Fig 1A). Our model (described in detail below) closely follows Loebel et al. [28], which was based on the rate model of Wilson and Cowan [70] except that the basic units represented single neurons rather than local populations of neurons. A previous work [36] showed the equivalence of this rate model and a network of integrate & fire units in terms of the spontaneous and evoked activities, and importantly in the generation of population spikes. Our main modifications to the model of Loebel et al. [28] were the introduction of (i) synaptic depression in thalamocortical synapses and (ii) heterogeneity of tuning curves within each column, as suggested by imaging studies of mouse A1 [46,47]. We note that although cortical SSA was mostly reported in the cat [1] and rat [2,4,12,16], the physiological properties of A1, e.g. the heterogeneity of tuning curves on which we rely here, are better described in mice. All simulations, data analysis and calculations were performed using MATLAB (MathWorks).
Each column in our model is a fully-connected recurrent network: all neurons within the column are connected to all others. Our simulations show that population spikes occur also in a network with connection probabilities as found in vitro [27]. In general, even with a connection probability of around 0.1, as reported by Levy and Reyes [27] for connections between pyramidal neurons, the shortest path between any two neurons is on average less than 2 synapses (see e.g. [71]), and recruitment of the neuronal population into the population spike is rapid. Each column represents an iso-frequency band along the tonotopic axis of A1, covering about 1/3 of an octave, and consisting of NE excitatory and NI inhibitory neurons. Each neuron is described by two dynamic variables: firing-rate, denoted by EiQ for excitatory and IlQ for inhibitory neurons (Q specifies the column index and i or l the index of the neuron within the column); and the amount of resources available at each of the synapses made by this neuron, relative to its full resources, a fraction between 0 and 1 denoted xiQ for excitatory and ylQ for inhibitory neurons. Every neuron in the network receives external input, denoted eiE,Q for excitatory and elI,Q for inhibitory neurons. External inputs are uniformly distributed within a specific range of firing-rates, i.e. each neuron receives input that depends linearly on its index within the column. All neurons within each column receive input from the excitatory populations of the nearest- and second-nearest neighboring columns, with connection strengths that decrease according to inter-column distance. Most excitatory neurons in the network also receive sensory input through a set of thalamocortical (ThC) synapses, each mediating a different tone frequency f and having its own dynamic variable, ziQ,f, representing the fraction of synaptic resources available at this synapse.
The network dynamics are defined by the following equations, following Loebel et al. [28] with the modifications described above:
τEdEiQdt=−EiQ+(1−τrefEEiQ)⋅g(∑R=−22JEE|R|NE∑j=1NEUxjQ+REiQ+R+JEINI∑j=1NIUyjQIjQ+eiE,Q+∑f=1MUsziQ,fsfTiQ,f)
(1)
τIdIlQdt=−IlQ+(1−τrefIIlQ)⋅g(∑R=−22JIE|R|NE∑j=1NEEjQ+R+JIINI∑j=1NIIjQ+elI,Q)
(2)
dxiQdt=1−xiQτrecE−UxiQEiQ
(3)
dylQdt=1−ylQτrecI−UylQIlQ
(4)
dziQ,fdt=1−ziQ,fτrecs−UsziQ,fsfTiQ,f
(5)
The units of our rate equations (Eqs 1 and 2) and rate variables (E, I, e, s) are 1/s. In the text and figures we specify the rate as Spikes/s for clarity.
J represents synaptic efficacy and has values depending on the type of connection and the distance over which it is made: EE denotes an excitatory-to-excitatory connection, IE an excitatory-to-inhibitory connection, II an inhibitory-to-inhibitory connection and EI an inhibitory-to-excitatory connection; the distance of the connection, in columns, is specified by a superscript R (R = 0 represents intra-columnar connections). U represents the fraction of utilization, i.e. how much of the available synaptic resources is utilized upon an action potential reaching the presynaptic terminal. Us is the fraction of utilization of the ThC synapses. τE and τI are the membrane time-constants of excitatory and inhibitory neurons, respectively. τref is the refractory period of excitatory or inhibitory neurons, as specified by the superscript E or I. τrec represents the time-constant for recovery of available synaptic resources in excitatory, inhibitory or ThC synapses, as specified by the superscript (E, I or s, respectively).
Eqs (1) and (2) describe the dynamical behavior of the excitatory and inhibitory neurons, respectively. These are rate equations–they describe the output rates of the neurons as a non-linear gain function of their summed inputs. The excitatory neurons receive intracortical excitation and inhibition (first two terms of the right-hand side of Eq 1) as well as direct excitatory thalamic input (the last term of the right-hand side of Eq 1). For simplicity, inhibitory neurons receive only intracortical excitation and inhibition, with no direct thalamocortical input. The lack of direct thalamo-cortical input to the inhibitory neurons in the model is discussed in length and justified in the main paper.
Most importantly for the special behavior of this model, the synapses impinging on excitatory neurons show depression. The model of synaptic depression implemented here consists of resource depletion followed by exponential recovery (Eqs 3–5). Each input class on the excitatory neurons has its own separate dynamical process of resource depletion and recovery.
The gain function used in the firing-rate equations is defined as:
g(w)={0w<0w0≤w<EmaxEmaxw≥Emax
(6)
Where w is the sum of inputs to a neuron, as detailed in Eqs 1 & 2. In all of our simulations, Εmax was set to 300 spikes/s. The implementation of a maximum firing-rate follows [28]. This value for Εmax is close to the effective maximum firing-rate that is set by the refractory period (τref) when its value is set to 3 ms (as we do here for both excitatory and inhibitory neurons; see list of values below).
Our model neurons sample the frequency axis in channels (f) that are ordered from 1 to M. In almost all types of protocols, M was equal to the number of columns and so f corresponds to Q (i.e. there is one channel per cortical column). The exceptions to this were the Diverse Narrow protocol, which required denser frequency channels, and the tests for hyperacuity (see below and in the Results section).
The input mediated by each ThC fiber is represented by a firing-rate variable sf. Its magnitude at each time-point depends on the maximum amplitude of the stimulus (A, measured in spikes/s) and the temporal envelope of stimuli presented in that channel (ξf(t), a fraction between 0 and 1):
sf=ξf(t)⋅A
(7)
In all the stimuli presented in this work, ξf(t) had the form of a trapezoid pulse: a 5-ms ascending linear ramp from 0 to 1, a period of constant amplitude, and a 5-ms linear ramp descending back to 0, all adding up to the nominal stimulus duration.
TiQ,f specifies the relative amplitude at which an excitatory neuron receives input from frequency channel f compared to its best frequency (BF). The values of TiQ,f over all channels make up the tuning curve of the neuron’s thalamic input. Each neuron was assigned a triangular-shaped input tuning curve. This shape was chosen in order to avoid having input to all columns upon each sensory stimulation, as was the case with the tuning curves used by Loebel et al. [28] that decayed exponentially from their peak at the best frequency. The width of the tuning curve is determined by a localization parameter λ, which scales the tuning curve according to the frequency separation between the columns. The equation for the input tuning curve of a typical neuron of column Q, i.e. with best frequency fBF = Q, is:
TtypicalQ,f={0f<Q−λ1−(Q−f)/λQ−λ≤f<Q1−(f−Q)/λQ≤f<Q+λ0f≥Q+λ
(8)
The value of λ = 5 used in our simulations confers upon the neurons in our model tuning widths of about 2 octaves (cf. Fig 1D), similar to the tuning of suprathreshold activity in rat A1 [40].
Recently, it has been shown that neurons within a local population of mouse A1 have heterogeneous best frequencies [46,47]. In line with these results, each column in our model contained neurons with several different best frequencies: Approximately 1/16, 1/8, 1/8 and 1/16 of the neurons in each column had their tuning-curves shifted by -2, -1, +1 and +2, respectively, relative to the typical fBF = Q. Neurons with shifted input tuning-curves were selected randomly (Fig 1A illustrates the resulting tonotopy; the distribution is adjusted for illustration purposes). The ThC input to all neurons that had a firing-rate of 0 spikes/s when there was no sound input (“non-active” neurons) was zero, as proposed by Loebel et al. [28]. Calcium imaging studies in mice [47,72] suggest that the heterogeneity of tuning-curves implemented in our model is intermediate between those of layer 4 and layers 2/3 in mouse A1. We note that the randomization of tuning-curves was not done in order to generate independent samples, but in order to better model the physiology. We do not compare between networks with different randomized tuning-curves. Rather, almost all of our analyses were made on column averages. Our simulations of networks with homogeneous vs. heterogeneous columns showed that heterogeneity of the tuning curves slightly extended the parameter range that allowed stimulus-specific adaptation, but its effect was very small (cf. Fig 7B).
All simulations of this model used networks with 21 columns. The values we used for the different parameters were:
NE,NI=100;λ=5;U=0.5;Us=0.7τE,τI=1×10−3s;τrefE,τrefI=3×10−3s;τrecE,τrecI=0.8s;τrecs=0.3sJEE0=6;JIE0=0.5;JEI=−4;JII=−0.5;JEE1=4.5×10−2;JIE1=3.5×10−3;JEE2=1.5×10−2,JIE2=1.5×10−3
Values of e for both populations were distributed uniformly between -10 and 10 spikes/s. The recovery time constants for both excitatory and inhibitory neurons are referred to as τrec, since their values were identical in all of our simulations.
The values above are the same as in Loebel et al. [28]. Some were based on experimental studies, while others were tuned to obtain approximate balance of excitation and inhibition, a spontaneous firing-rate of a few spikes/s, and no occurrence of spontaneous PSs. All simulations were run using a time step dt = 1 x 10−4 s.
Stimuli are called “tones” here because they were set to simulate pure-tone stimulation. Each was presented in one frequency channel f. Since the behavior of the model in response to sequences of stimuli was of major interest here, tones were presented in sequences consisting of 100 stimuli. The number of different tones was determined by the protocol type [4,12]. In the oddball protocol there were 2 tones, the lower at frequency f1 and the higher at frequency f2, with a frequency separation Δf = f2 –f1. These tones were centered on the best frequency of the middle column of the network (column 11), using Δf = 2 unless specified otherwise. The deviant tone had a probability of occurrence P (P = 10% unless specified otherwise). The other tone, called the standard, occurred at a probability of 1 – P. Tone sequences were generated by a random permutation of a block of standard tones followed by a block of deviant tones, in a way that allowed deviants to occur close apart, or even one after another. Responses to the two conditions were calculated as averages over all occurrences, including deviant responses that were weaker due to recent deviants (cf. Fig 6B). This randomization is in line with the practice in experimental studies by our group. We do not treat the short-term effects of deviants on one another. For experimental results of this effect, see [43,73]. The oddball protocol was always run twice, first with f2 as deviant and f1 as standard and then with their roles reversed (Deviant f2 and Deviant f1; Fig 3F). The Equal protocol was similar to the oddball but had P = 50%. In the Deviant Alone protocol, all presentations of the standard tone were replaced with silent trials. We also used two sequences of “deviant-among-many-standards”, in which we presented 10 tones spaced evenly along the frequency scale (including f1 and f2), each occurring at a probability of 10%. The Diverse Narrow protocol had two tones lower than f1, four between f1 and f2, and two higher than f2. This required a denser sampling of the frequency axis, as the spacing between tones had to be Δf/5 (cf. Fig 4A). In the Diverse Broad protocol we used 10 tones with a spacing of Δf = 2, which is the same as in the oddball protocol. Four tones were below f1 and four were above f2 (cf. Fig 4A).
Individual stimuli had a duration of 50 ms, including onset and offset linear ramps of 5 ms. Inter-stimulus interval (ISI) was defined as the time between the onsets of consecutive stimuli. We used ISI = 350 ms, A = 5 spikes/s unless specified otherwise.
Responses to stimuli were generally evaluated based on the mean firing-rate in the middle column, E, unless explicitly stated otherwise. In Figs 1C, 3B, 3D, 4G, 8B, 9C, 10A and 10B we present and analyze responses of other columns as well. For each stimulus, a spike-count was calculated by correcting E to the baseline (defined as the average of E during the 5 ms before stimulus onset) and then integrating it from stimulus onset to 45 ms after offset. Since this is a firing-rate model that has no sources of noise, a period as short as 5 ms is valid as a baseline. Its advantage is that it allowed us to calculate of the response over 45 ms even for the shortest ISIs used. The response to each tone was defined as the average spike-count for all presentations of that tone. In experimental studies of SSA the baseline is usually not subtracted (but see [12]). However, it should have little effect over the response sizes and strength of the SSA, since in general our simulations showed low spontaneous rates and very large responses.
We quantified the strength of SSA in a way that is becoming standard [1,2,4,8,9,12,16,60,61]. The responses in both conditions of the oddball protocol (Deviant f2, Deviant f1) were calculated and used to compute the Common-Contrast SSA Index (CSI):
CSI=d(f1)+d(f2)−s(f1)−s(f2)d(f1)+d(f2)+s(f1)+s(f2)
(9)
Where d(f1) and s(f1) (d(f2) and s(f2)) are the responses to f1 (f2) when it was used as deviant and standard, respectively.
The dynamics of resources in each thalamocortical synapse (Eq 5) is a one-dimensional system, independent of the other variables in the network. The amount of resources at stimulus onsets or offsets during a train of standard stimuli can be treated as an iterated map, given that the stimulus duration and inter-stimulus intervals are constant throughout the stimulus protocol. For simplicity, we rewrite Eq 5 as:
dziQ,fdt=1τrecs−(1τrecs+UssfTiQ,f)ziQ,f
(10)
When no stimulus is presented, the solution of this equation is:
ziQ,f(t)=1−[1−ziQ,f(toffset)]exp(−t−toffsetτrecs)
(11)
Where toffset is the time when the last stimulus ended. For the period before the first stimulus in a protocol, we assume the system is in its steady-state and therefore ziQ,f = 1 for all synapses. Under stimulus presentation, the solution of Eq 10 is:
ziQ,f(t)=11+τrecsUssfTiQ,f−[ziQ,f(tonset)−11+τrecsUssfTiQ,f]exp[−(1+τrecsUssfTiQ,f)t−tonsetτrecs]
(12)
Where tonset is the onset of the current stimulus. Since we are interested only in the fractions of synaptic resources at stimulus onsets or offsets, we substitute the relevant time periods and get the following relations between the values of z for consecutive onset and offset times:
zonset=1−kisi(1−zoffset)
(13)
zoffset=z˜ss−kdur(zonset−z˜ss)
(14)
Where:
zonset≡ziQ,f(tonset);zoffset≡ziQ,f(toffset);z˜ss≡11+τrecsUssfTiQ,f;kisi≡exp(−tonset−toffsetτrecs);kdur≡exp(−toffset−tonsetτrecsz˜ss)
Substituting Eq 13 in Eq 14, and vice versa, gives the two iterated maps:
zonset(n+1)=1−kisi[1−z˜ss(1−kdur)]+kdurkisizonset(n)
(15)
zoffset(n+1)=z˜ss(1−kdur)+kdur(1−kisi)+kdurkisizoffset(n)
(16)
Which have the joint stable fixed points:
z*onset=1−kisi[1−z˜ss(1−kdur)]1−kdurkisi
(17)
z*offset=z˜ss(1−kdur)+kdur(1−kisi)1−kdurkisi
(18)
Eqs 17 and 18 are the expressions for the onset and offset values of z reached after enough identical stimuli were presented. These values were used in Fig 9C and 9D, as the approximate steady-state resources in synapses conveying the standard stimuli from the thalamus to the standard column (the column whose best frequency is the standard stimulus). They are correct when all neurons in the column have the same best frequency, when the stimulus sequence consists of standard tones only, and when stimulation is on-off only (no ramp in the stimulus envelope, i.e. ξf(t) consists only of square pulses).
The MATLAB scripts for running our model and the data replotted from [4,12,45] are available online as supporting information.
|
10.1371/journal.ppat.1003930 | The Secreted Triose Phosphate Isomerase of Brugia malayi Is Required to Sustain Microfilaria Production In Vivo | Human lymphatic filariasis is a major tropical disease transmitted through mosquito vectors which take up microfilarial larvae from the blood of infected subjects. Microfilariae are produced by long-lived adult parasites, which also release a suite of excretory-secretory products that have recently been subject to in-depth proteomic analysis. Surprisingly, the most abundant secreted protein of adult Brugia malayi is triose phosphate isomerase (TPI), a glycolytic enzyme usually associated with the cytosol. We now show that while TPI is a prominent target of the antibody response to infection, there is little antibody-mediated inhibition of catalytic activity by polyclonal sera. We generated a panel of twenty-three anti-TPI monoclonal antibodies and found only two were able to block TPI enzymatic activity. Immunisation of jirds with B. malayi TPI, or mice with the homologous protein from the rodent filaria Litomosoides sigmodontis, failed to induce neutralising antibodies or protective immunity. In contrast, passive transfer of neutralising monoclonal antibody to mice prior to implantation with adult B. malayi resulted in 60–70% reductions in microfilarial levels in vivo and both oocyte and microfilarial production by individual adult females. The loss of fecundity was accompanied by reduced IFNγ expression by CD4+ T cells and a higher proportion of macrophages at the site of infection. Thus, enzymatically active TPI plays an important role in the transmission cycle of B. malayi filarial parasites and is identified as a potential target for immunological and pharmacological intervention against filarial infections.
| Triose phosphate isomerase (TPI) is a ubiquitous and highly conserved enzyme in intracellular glucose metabolism. Surprisingly, the human lymphatic filariai nematode parasite Brugia malayi, releases TPI into the extracellular environment, suggesting a role in helminth survival in the mammalian host. We first established that B. malayi-infected humans and rodents generate TPI-specific serum antibody responses, confirming presentation of this protein to the host immune system. However, immunisation of rodents with B. malayi TPI did not induce protection against infection. Furthermore, TPI from a related parasite, Litomosoides sigmodontis, did not induce protective immunity in mice. Notably, antibodies from infected hosts did not neutralise the enzymatic activity of TPI. We then generated twenty-three anti-TPI monoclonal antibodies, of which only two inhibited enzymatic activity. Transfer of neutralising antibody to mice prior to B. malayi infection effected a 69.5% reduction in microfilarial levels in vivo and a 60% reduction in microfilariae produced by individual adult female parasites. Corresponding shifts in the host immune response included reduced Th1 cytokine production and enhanced macrophage numbers. Enzymatically active TPI therefore promotes production of the transmission stage of B. malayi filarial parasites and represents a rational target for new vaccine and drug development to protect against filarial infections.
| Continued survival of parasitic helminths within their mammalian host requires that they neutralise potentially protective immune responses, generate energy and reproduce. Filarial nematodes are particularly long-lived, tissue-dwelling parasites which evade immunity and maintain transmission over many years [1]. Over 100 million people are infected with lymphatic filariae, such as Brugia malayi, and no vaccine is available for human use [2], [3]. Transmission occurs when blood-borne microfilarial larvae are taken up by a mosquito vector, generating infective third-stage larvae which enter humans on a subsequent blood-meal. Hence, any immunological means of blocking microfilarial release would interrupt transmission.
As extracellular pathogens, the interaction of live parasites with both the host and each other is likely to occur through a combination of excretory/secretory (ES) products and surface molecules [4], [5]. Given the presumed involvement of ES molecules in a range of processes essential for successful parasitism, they represent attractive vaccine and drug targets. Because of this, we and others have taken a proteomic approach to characterise the complex mixture of proteins secreted by the human filarial nematode Brugia malayi (B. malayi ES, BES) [6]–[9]. This revealed that the most abundant ES protein of adult B. malayi is the glycolytic enzyme triose phosphate isomerase (Bm-TPI, EC 5.3.1.1), predominantly from female worms. Detailed analysis of the secretions of all life cycle stages has revealed that TPI is also released by moulting L3 larvae early in infection [8].
TPI catalyses the interconversion of the triose phosphates glyceraldehyde 3-phosphate and dihydroxyacetone phosphate, an essential step in glycolysis and gluconeogenesis [10]. Whilst TPI has been detected in the ES products of other worms, such as the cercariae and eggs of Schistosoma mansoni [11], [12] and adult Haemonchus contortus [13], the levels appear low compared to the large amounts released by adult B. malayi [6]–[8]. Furthermore, there is little or no secretion of other glycolytic enzymes, implying that TPI is selectively secreted through an active process, rather than simply leaching from compromised worms. This was supported by the demonstration that TPI is approximately 20-fold enriched in BES compared to a soluble extract of adult B. malayi [6].
However, it is unclear why Brugia and other nematodes secrete TPI given that glycolysis occurs in the cytosol. In this regard, several reports indicate that TPI is a multifunctional protein. For instance, TPI binds to the intracellular tail of the integrin αIIb in platelets and may regulate integrin signalling [14]. TPI can also function as an extracellular adherence molecule, and in this way mediates the interaction of the fungal pathogen Paracoccidiodes brasiliensis with both host epithelial cells and the extracellular matrix proteins laminin and fibronectin [15]. Similarly, surface associated TPI of Staphylococcus aureus contains a lectin activity that can bind fungal sugars and promote bacterial adherence to, and subsequent killing of, Cryptococcus neoformans [16], [17]. Additionally, studies of human TPI deficiency have shown that exogenous TPI can complement TPI-deficient cells, suggesting that the secreted enzyme may be taken up in a functional form by surrounding cells [18], [19].
Glycolytic enzymes, including TPI, have also been tested as targets of protective immunity. Thus, a neutralizing monoclonal antibody to S. mansoni TPI can confer up 40–50% reduction in worm burden in mice [20] and DNA vaccination with S. japonicum TPI reduces worm and egg burdens in experimentally infected pigs and water buffalo [21], [22]. Successful vaccination with schistosome TPI is consistent with its induction of IFNγ, a cytokine associated with protective immunity against the larval schistosomula [23]. Even in the Th2-dominated environment that develops following schistosome egg production in mice, TPI preferentially stimulates Th1 cytokines [24]. S. mansoni TPI can also induce Th1 differentiation by T cells from unexposed humans [25]. In other helminth species, certain glycolytic enzymes have been similarly tested as vaccine candidates: for example, Onchocerca volvulus fructose 1,6-biphosphate aldolase is strongly recognised by antibodies from exposed but uninfected subjects, and can elicit a 50% reduction in larval survival in vaccinated mice [26]. However, studies on B. malayi have to date focussed solely on the biochemical properties of glycolytic enzymes with a view to development of new pharmacological inhibitors [27], [28].
Since glycolysis plays a key role in filarial worm energy metabolism [29]–[31], coupled with the unusually high level of secretion of TPI, we investigated the role of Bm-TPI in B. malayi infection. We confirmed that Bm-TPI is highly preferentially secreted, enzymatically active, and an antibody target in both infected mice and humans. Whilst vaccination with filarial TPI failed to confer protection against challenge infection with B. malayi or Litomosoides sigmodontis, antibody-mediated neutralisation of Bm-TPI shows it is required for the optimal survival of microfilariae within the mammalian host. As such, this parasite enzyme represents a novel and rational target for intervention by immunological or pharmacological means.
To study the role of filarial triose phosphate isomerases, we first cloned the cDNA encoding this enzyme for expression as recombinant proteins. Full-length Bm-TPI cDNA was amplified by PCR from mixed adult B. malayi cDNA, cloned and confirmed as identical to the annotated B. malayi gene (Bm1_29130 [32]). The encoded 247-aa protein lacks a signal sequence, and has a predicted molecular weight of 27,097 Da. Sequence analysis shows a high degree of amino acid conservation with human (61% identity), S. mansoni (58%) and C. elegans (76%) proteins, including the AYEPVWAIGTG active loop and catalytic E165 (corresponding to E166 in human TPI; [33]), as well as the other active site residues, N10 (human N12), K12 (K14) and H94 (H96) (Fig. 1 A).
Recombinant Bm-TPI was expressed in bacteria and purified by nickel resin affinity chromatography, appearing as a single band of approx. 28 kDa by SDS-PAGE and a dominant molecular species by mass spectrometry of 28,030 (data not shown). Functional activity of recombinant Bm-TPI was confirmed by enzymatic assay, in which it displayed typical Michaelis-Menten kinetics indistinguishable from rabbit TPI with a Vmax of 715 U/mg and a Km of 1.8 mM (Fig. 1 B). The activity of Bm-TPI was compared to the homologous enzyme from the mouse filarial parasite Litomosoides sigmodontis [34]. Ls-TPI has 94% (233/247) amino acid identity to Bm-TPI (Fig. 1A), and following cloning of the corresponding cDNA and bacterial expression, recombinant protein showed similar enzyme kinetics to both Bm-TPI and rabbit TPI (Fig. 1 B).
Previous proteomic studies have indicated Bm-TPI is amongst the most abundant proteins secreted by adult female B. malayi [6]–[8]. Preferential secretion was confirmed by Western blot using a polyclonal antiserum raised against rBm-TPI, which showed significant enrichment of Bm-TPI in BES compared to somatic extracts of adult worms, L3 larvae and microfilariae (Fig. 2 A). Native secreted Bm-TPI was shown to be enzymatically active by comparing BES with varying amounts of recombinant Bm-TPI. This revealed that each μg of BES had equivalent enzymatic activity to 370±86 ng recombinant protein (Fig. 2 B). Enzymatic activity in BES was abolished by heat denaturation (Fig. 2 B).
Immunohistochemistry of sections of adult male and female B. malayi showed ubiquitous somatic expression pattern expected of a glycolytic enzyme, but provided no clues as to the source of secreted Bm-TPI by adult females (Fig. 2 C). Additionally, surface staining of intact whole worms was not seen, indicating that Bm-TPI is not shed from the cuticle of the parasite (data not shown).
An important question is whether the prominent expression of Bm-TPI results in a strong antibody response in infected patients. We analysed serum samples from a cohort of B. malayi-exposed residents of Rengat, Sumatra, Indonesia that were classified into presumed uninfected subjects (“endemic normals”), asymptomatic microfilarial carriers, and patients with chronic filarial pathology who are generally amicrofilaraemic [35], [36]. Individuals within each exposed group were found with positive IgG responses against Bm-TPI compared to sera from unexposed UK residents (Fig. 3 A). However, a greater proportion of infected individuals suffering from lymphatic pathology were seropositive (76%) compared to asymptomatic microfilaremics (48%) and endemic normals (42%), and the majority of strong responders were within the filarial pathology group. In contrast, no antibody reactivity was detected against mammalian TPI (using rabbit TPI which has 245/249 amino acid identity with the human protein) (Fig. 3 B). An isotype analysis of anti-Bm-TPI antibodies showed that reactivity was confined to the IgG1 and IgG4 isotypes (Fig. 3 C–F); notably IgG4 levels were higher to Bm-TPI in the pathology group, although the Mf+ individuals display far higher IgG4 levels to total B. malayi somatic antigens [35]. As we had previously detected little anti-Bm-TPI antibody reactivity using 2-D Western blots [6], the high level of reactivity found by ELISA indicated that the epitopes are predominantly conformational and denatured by SDS-PAGE electrophoresis, a conclusion supported by the lack of immunoreactivity in the vast majority of individuals to heat-treated Bm-TPI (Fig. S1 A–D).
We next assessed whether vaccination of Mongolian jirds (Meriones unguiculatus), which are fully permissive to B. malayi infection [37], with Bm-TPI would generate protective immunity against challenge infection. In an initial experiment, animals were vaccinated three times with either Bm-TPI or control protein (BSA) in alum. Animals were infected intraperitoneally, with serum antibody titres and worm burdens being determined at 8 weeks post-challenge. All infected animals made strong IgG1 responses against a somatic extract of B. malayi adults (data not shown), but prior immunization with Bm-TPI induced >10 times higher IgG1 titres against this protein (Fig. 4 A). Despite this potent antibody response, and as shown in Fig. 4 B, there was no significant reduction in worm burdens at 8 weeks of infection in the Bm-TPI-immunized jirds (44±6.5, vs BSA, 55±7.0, p = 0.290). In a further experiment, we reasoned that a longer duration of infection might be required to see any protective effects induced by vaccination with a largely adult-specific secretory product. As such, jirds were immunised with Bm-TPI or BSA as before, challenged and assessed 21 weeks later. Vaccination again induced high titers of anti-TPI IgG1 (Fig. 4 C), but failed to provide any protection, and indeed both adult worm (Fig. 4 D) and peritoneal microfilariae numbers (Fig. 4 E) were slightly elevated compared to the BSA control.
The jird model is one that tests immunity of a rodent host to a human parasite, with parasites resident in the peritoneal cavity rather than their physiological niche of lymphatic vessels (for adult worms) and blood (for microfilariae). We therefore conducted a parallel test of protective capacity of TPI in a natural murine model of filarial infection, L. sigmodontis, which resides in the pleural cavity [34]. Mice were vaccinated three times with Ls-TPI in alum, and then challenged with L. sigmodontis L3. The results were similar to those observed with B. malayi in the jird: while specific anti-TPI antibodies were strongly boosted (Fig. 4 F), adult worm numbers were unchanged at day 70 post-challenge (Fig. 4 G) and when worm lengths were measured no differences were seen (data not shown). Moreover, circulating microfilarial numbers were not significantly diminished in immunized animals (Fig. 4 H).
One explanation for the poor level of protection induced by vaccination with Bm-TPI is the failure to generate high titres of neutralizing antibodies. Indeed, the sera from vaccinated jirds with high anti-Bm-TPI titres (Fig. 4 B) had limited ability to block Bm-TPI catalytic activity, leaving 75% of isomerase activity intact (Fig. 5 A). Similarly, polyclonal mouse serum raised to Bm-TPI effected only a slight reduction in enzyme activity (Fig. 5 B), whilst human sera from individuals strongly reactive to Bm-TPI (Fig. 3) failed to inhibit the enzyme at all (Fig. 5 C). This suggested that both immunisation and natural infection generated only limited amounts of anti-Bm-TPI antibodies directed at the active site. To confirm this, a panel of mouse monoclonal antibodies specific for Bm-TPI were generated. Only 2 of 23 (9%) anti-Bm-TPI clones were capable of blocking enzyme activity (Fig. 5 D), of which one anti-Bm-TPI mAb (clone 1.11.1, IgG1 isotype) was used in subsequent experiments.
MAb 1.11.1 was able to effectively block recombinant Bm-TPI enzymatic activity in a dose-dependent manner (Fig. 5 E), with a calculated Ki of <1 μg/ml for 100 ng Bm-TPI. In contrast, no blockade of either mammalian (Fig. 5 E), or perhaps surprisingly, L. sigmodontis TPI (Fig. 5 F) was seen. Most importantly, anti-Bm-TPI blocked native Bm-TPI as assessed by its ability to inhibit TPI activity present in BES (Fig. 5 G).
We then tested whether antibody inhibition of TPI enzymatic activity could cause parasite death in vitro. However, adult male and female B. malayi were able to survive in culture for sustained periods (≥3 days) in the presence of up to 500 μg/ml mAb clone 1.11.1 (data not shown). This suggested that whilst the antibody can inhibit secreted TPI, it cannot act directly on the parasite, and that in vitro worm survival over this period does not depend on TPI activity in the culture medium.
Next, we tested whether MAb 1.11.1, with specific neutralizing ability, would alter the course of filarial infection in vivo. Mice were implanted intraperitoneally with 10 B. malayi adults (8 female, 2 male) and treated every 1–2 days with 200 μg anti-Bm-TPI mAb or IgG1 isotype control. Transfer of mAb 1.11.1 mAb established serum anti-Bm-TPI titres 35-fold greater than develop normally in response to infection, as seen in mice given the isotype control antibody (Fig. 6 A). Furthermore, transfer of 1.11.1 mAb conferred on recipient serum the ability to effectively block TPI enzymatic activity in vitro (Fig. 6 B). Despite this, live adult worms were recovered from the peritoneal cavities of both groups of infected mice after 28 days (Fig. 6 C) with no significant differences in male or female numbers (data not shown). In contrast, numbers of peritoneal microfilariae (Mf) were significantly reduced in anti-Bm-TPI-treated mice, being 69.5% lower than in animals given isotype control (Fig. 6 D) indicating that Bm-TPI activity is required for either the release or survival of live Mf.
Using Mf obtained from non-immunised jirds, we next demonstrated that in vitro Bm-TPI blockade was unable to kill Mf (Fig. 6 E), indicating that the antibody is not directly toxic to Mf, and that neutralisation of TPI is not detrimental to this stage of the parasite. Likewise, in vitro Bm-TPI blockade did not reduce Mf production by adult females obtained from non-immunised jirds (Fig. 6 F). Instead, when we cultured adult females from the peritoneal cavity of mice following in vivo anti-Bm-TPI or isotype treatment, parasites from anti-Bm-TPI treated mice produced ∼60% fewer MF in vitro, consistent with Bm-TPI blockade compromising parasite fitness in terms of female reproductive output in vivo (Fig. 6 G). In particular, a much greater proportion of adult female worms from anti-TPI-treated jirds completely failed to release live Mf during in vitro culture (45.2% compared to 12.5% in isotype-treated controls).
Analysis of uterine contents from individual female B. malayi revealed that TPI blockade led to a significant decrease in the number of unfertilised oocytes (Fig. 6 H). Whilst oocytes were the predominant developmental stage in females obtained from isotype treated mice (Fig. S2 A), anti-Bm-TPI treatment led to the accumulation of smaller developmental stages that appeared damaged or partially degraded (Fig. S2 B). In contrast, no significant difference was observed in the number of Mf transferred directly into antibody-treated mice (Fig. 6 I) again implying that TPI blockade targets female egg production, thereby reducing subsequent release of live Mf.
Parasite mediated-immunomodulation relies on products secreted into the environment of the pathogen in vivo, and the analysis of in vitro released “excretory-secretory” (ES) proteins has provided an approximation of the spectrum of released macromolecular components. We and others have identified Bm-TPI as a dominant product secreted by adult B. malayi worms [6]–[9], and here confirm not only the preferential secretion of this enzyme by live adult worms in vitro, but show, by its antigenicity in infected mice and humans, that it is exposed to the immune system in vivo. How TPI is secreted, in the absence of a signal peptide, remains unclear and its ubiquitous expression throughout the somatic tissues of B. malayi does not provide any pointers to a particular route of secretion. Possibly, as secretion is far higher in female worms than males [7], TPI could be released along with microfilariae from the female genital tract.
Filarial TPIs, from both B. malayi and L. sigmodontis, are active enzymes with catalytic properties very similar to those of mammalian homologues. Immunologically, however, TPIs from mammals and nematodes are non-cross-reactive, and indeed we generated monoclonal antibodies capable of distinguishing between the two filarial enzymes. Hence, in human infections anti-Bm-TPI antibodies were not found to be auto-reactive with self TPI, suggesting that the pathologies of lymphatic inflammation, edema and fever are not linked to an autoimmune reaction against human TPII. In this respect, filarial infection differs from that of Trypanosoma cruzi, which induces auto-antibody production against host TPI [39].
When comparing the levels of anti-Bm-TPI between infected individuals we noted that titres are greater in cases of filarial pathology in whom circulating microfilariae are generally absent. Because of this, we tested the potential of Bm-TPI to provoke protective immunity against B. malayi in animal models. In the jird, M. unguiculatus, which is fully susceptible to infection with the mosquito-borne L3 stage, both total worm and Mf numbers were unchanged in vaccinated animals following peritoneal infection. However, vaccination did not generate high levels of neutralizing antibody in terms of the catalytic activity of the enzyme, which may be more efficient than removal of TPI by complexing and opsonization. By screening a large panel of murine monoclonals, we selected a neutralizing antibody that conferred, in a mouse peritoneal implantation model, immunity against the Mf stage.
We also noted that TPI blockade inhibits the development of eggs within the adult female worm, reflecting a loss of fitness in the parasite that may have either or both a metabolic or immunological cause. Several other regimes have been shown to limit the release of viable microfilariae by filarial parasites. In this regard, antibiotic-mediated depletion of endosymbiotic Wolbachia causes degeneration of B. malayi oocytes, embryos and microfilariae [40], [41], although in this setting parasite metabolism is likely to be more heavily compromised. An immunological cause of reduced parasite fecundity can also be illustrated in mice immunised with L. sigmodontis Mf, >70% of which fail to develop microfilaraemia, due to an inhibition of embryogenesis to the pretzel stage [42].
The selective effect on microfilarial levels recapitulates a consistent, but unexplained, feature of filarial nematode infections. In both humans and animal models, cryptic amicrofilaraemic infections can occur in which immunity appears to operate only against the microfilarial stage. For example, cats infected with B. pahangi often became Mf-negative and yet remained seropositive for circulating filarial antigen and were found to harbour live adult worms at autopsy [43]. Similarly, circulating antigen tests in humans identify a significant proportion of Mf-negative infected subjects. However, in the majority of cases, it appears that natural infection does not generate blocking antibody to TPI suggesting the possibility that the active site has in some manner evolved to minimise stimulation of neutralising antibody. The surprising finding that such antibodies do not cross-react between the highly conserved TPIs from two related filarial species reflects a further unusual property of these proteins.
To establish whether TPI enzyme activity is essential for its extracellular function, and whether only blocking antibodies can be protective, it would be desirable to test multiple panels of monoclonals and to conduct experiments with pharmacological inhibitors of filarial TPI. Future studies may also develop blocking antibodies against L. sigmodontis TPI which would permit the analysis of the biological role of TPI during the full course of infection in mice, as the murine antibodies we describe here cannot be administered to other rodent species for extended periods of time.
In addition, species-specific chemical TPI inhibitors have been described which target non-conserved amino acids, particularly at the dimer interface e.g. Trypanosoma cruzi [44], Trypanosoma brucei [45], Plasmodium falciparum [46] and Giardia lamblia [47]. Small molecule inhibitors may be superior to antibody-mediated blocking as they offer the advantage of penetration into the parasite itself, rather than just inhibition of secreted form, and are likely to be more effective against the adult worms which, as we show, survive even in the face of very high neutralising Ab titres.
The in vivo consequences of TPI neutralisation may link poor microfilarial survival with shifts in the anti-parasite immune response. Thus, the reduction in IFN-γ production by peritoneal cavity CD4+ T cells may be an indirect effect of diminished numbers of microfilariae, as this stage (unusually) induces Th1 responsiveness [48]–[50]. Similarly, the reduced eosinophilia could reflect weaker stimulation from the microfilarial stage. Interestingly, this argument would suggest that microfilariae normally dampen macrophage expansion in vivo, a point which has yet to be experimentally investigated.
The finding that TPI is also secreted by plant-parasitic nematodes [51], and the importance of glycolysis in adult B. malayi [31], are consistent with TPI release facilitating adult worm metabolism and being required for optimal fecundity and Mf production in vivo. It is also possible that the heightened frequency of alternatively-activated macrophages imposes immunological damage on adult worms, thereby reducing their ability to reproduce, and such damage may of course be more easily achieved if the target is also metabolically compromised. In any event, our data present a remarkable immunological strategy for transmission in the filarial nematode: adult females release TPI which promotes oogenesis, and so increases the number of their offspring. Hence, in the mouse model at least, TPI neutralisation inhibits this process and reduces the microfilarial burden.
BALB/c mice and Meriones unguiculatus jirds were bred in-house. B. malayi (obtained originally from TRS Laboratories, Athens, Georgia, USA) was maintained in Aedes aegypti mosquitos and jirds. Infective larvae (L3) were recovered from mosquitos 12–14 days following feeding on microfiariae- (Mf-) containing blood, as detailed previously [52]. Jirds were infected with up to 600 L3; adult worms and microfilariae were recovered from the peritoneum approximately 4 months later. The Litomosoides sigmodontis life cycle was maintained, and L3 larvae or adult worms recovered, as described previously [53]. Somatic extracts of worms (from adults, BmA; from L3, L3A and from Mf, MfA) and BES were produced as previously described [6].
All animal protocols adhered to the guidelines of the UK Home Office, complied with the Animals (Scientific Procedures) Act 1986, were approved by the University of Edinburgh Ethical Review Committee, and were performed under the authority of the UK Home Office Project Licence number 60/4105. Human serum samples were taken from archived stocks derived from a study in Rengat, Indonesia that has been previously described [54] and in which informed consent was obtained from all patients before clinical and parasitologic investigation and blood withdrawal in accordance with the guidelines of the Indonesian Department of Health and Human Services.
Total RNA was extracted from adult mixed sex B. malayi and L. sigmodontis using TRIzol (Invitrogen), and reverse transcribed with MMLV reverse transcriptase (Stratagene) using standard protocols. A partial sequence (nt 1–664) for Ls-TPI (LS00587) was obtained from NEMBASE v4 (http://www.nematodes.org/nembase4/index.shtml). The missing 3′ end was obtained by 3′ RACE using Invitrogen Gene Racer Core kit with RACE-ready cDNA and the forward primer ATG TCT CGA AAG TTT CTA GTT as previously described [55]. The resultant full-length nucleotide sequence has been submitted to the European Nucleotide Archive with the Accession Number HG329626. The following PCR primers were used for amplification; Bm-TPI forward primer CAT ATG TCG CGA AAA TTT CTT, Ls-TPI forward primer CAT ATG TCT CGA AAG TTT CTA GTT, Bm-TPI and Ls-TPI reverse primer CTC GAG ATC ACG TGC ATG AAT AAT TT (restriction sites are underlined). PCR conditions were as follows: 35 cycles of 95°C 30 sec, 60°C 30 sec, and 72°C 2 min. Reaction products were separated on 1% agarose gels, visualised using ethidium bromide, and the 750-bp amplicons excised and purified (QIAquick gel extraction, Qiagen). PCR products were cloned into pGEM-T vector (Promega) and transformed into E. coli JM109 (Promega) for overnight colony formation. Minipreps from positive colonies were sequenced. An internal Nde1 site was removed in both Bm-TPI and Ls-TPI by PCR-based site-directed mutagenesis (CATATG replaced with CATACG, a synonymous mutation, in-frame codon underlined), using 50 ng of parental plasmid, Deep Vent DNA polymerase (New England Biolabs) and the following PCR primers : Bm-TPI forward primer CCT TAT TTA TCA TAC GTT AAG GAG AAA, Bm-TPI reverse primer TTT CTC CTT AAC GTA TGA TAA ATA AGG, Ls-TPI forward primer CCA TAT TTG TCA TAC GTC AAG GAA AAA GTT, Ls-TPI reverse primer AAC TTT TTC CTT GAC GTA TGA CAA ATA TGG. PCR conditions were as follows: 18 cycles of 95°C 30 sec, 55°C 1 min, and 75°C 8 min. Reaction products were digested with Dpn1 (New England Biolabs) for 2 hours at 37°C to remove parental plasmid, purified as for plasmid preps, and then used to transform E. coli JM109, and grown as before. Coding sequences for Bm-TPI and Ls-TPI were ligated into linearised pET29c (Novagen) following digestion with Nde1 and Xho1. Protein expression was induced in BL21(DE3) cells with 1 mM IPTG for 3 hours at 37°C. Bacteria were pelleted and lysed in Bug Buster supplemented with 25 U/ml benzonase (Novagen) for 20 min at room temperature. C-terminal His-tagged proteins were purified by metal affinity chromatography using Hi-Trap chelating Hp columns on an AKTAprime (GE Healthcare). Eluted fractions containing recombinant TPI protein were pooled and dialysed into PBS. Endotoxin was removed using Detoxi-Gel Endotoxin Removing Columns (Thermo Scientific). Recombinant TPI was stored at 2 mg/ml at −80°C.
Enzymatic activity of TPI was determined in the reverse direction (conversion of DHAP to G3P) as described by Lambeir et al [56]. Standard reaction conditions were 100 mM TEA-HCl pH 7.6, 1 mM EDTA, 0.16–10 mM DHAP (1.5 mM if not indicated), 1 mM NAD+, 5 mM disodium hydrogen arsenate and 10 μg rabbit GAPDH. Reactions were initiated with 100 ng TPI in a total volume of 150 μl. Rabbit TPI and all above reagents were from Sigma. The initial reaction rate was calculated by the change in NADH absorbance at 340 nm with a Nanodrop 2000 (Thermo Scientific). Enzyme activity was calculated in units (1 U = 1 μmol substrate formation min−1 μg−1 enzyme at 25°C). Calculations were confirmed using a NADH standard curve (Sigma). To determine the level of TPI activity in BES, reactions were initiated with 500 ng of BES or heat denatured BES (95°C 30 min), and compared to 25–400 ng of Bm-TPI. To assess the inhibitory potential of serum from immunised animals, enzymatic assays were performed in the presence of 10% test or control serum. Enzyme blockade by monoclonal antibodies was assessed in the range 0.5–50 μg/ml, and an isotype control mouse IgG1 MOPC31C was used where indicated. In these instances, enzymatic activity was calculated as percentage activity compared to the relevant control.
Recombinant Bm-TPI, rabbit TPI (Sigma), BSA or BmA were coated (1 μg/ml in 0.06 M carbonate buffer pH 9.6) onto Maxisorb 96 well Immunoplates (Nunc) overnight at 4°C. Plates were blocked in 2% BSA (mouse and human sera) or 1% casein (jird sera) in tris-buffered saline/0.05% tween 20 (TBST) for 2 hours at 37°C. Human IgG responses to Bm-TPI were tested using sera (1/100 dilution) from a previously characterised Indonesian B. malayi endemic study population [57]. Individuals were classified into three clinical groups, elephantiasis (pathology) patients, asymptomatic microfilaraemics, and endemic normals. Normal human sera (NHS) were obtained from non-exposed UK residents. IgG binding was detected using peroxidase labelled anti-human IgG (1/1000, DakoCytomation), or anti-human IgG1, IgG2, IgG3 or IgG4 (1/5000, The Binding Site). Plates were developed using ABTS peroxidase substrate (KPL), and measured with an Emax microplate reader (Molecular Devices). Samples were considered positive if they exceeded the mean value plus 3 standard deviations of nonendemic human serum samples. Jird sera (1/100 dilution) antibody levels were measured using peroxidase labelled anti-mouse IgG1 (1/2000, Southern Biotech). BALB/c anti-Bm-TPI titres were determined with doubling dilutions of sera (1/50 onwards) and anti-mouse Ig peroxidase (1/2000, DakoCytomation).
For polyclonal antibody production, BALB/c mice were immunised i.p. with 50 μg Bm-TPI in alum, and then boosted with 10 μg Bm-TPI i.p. in alum on days 28 and 35. Serum was recovered on day 42. For monoclonal antibody production, BALB/c mice were immunised with 50 μg Bm-TPI, and then boosted with 1 μg Bm-TPI in PBS i.v. on days 28, 29 and 30. Spleens were recovered on day 32 and fused with SP2 as before [58]. Cells were screened for Bm-TPI binding by ELISA, and positive wells were tested for Bm-TPI blocking ability in the above enzymatic assay. From this a blocking hybridoma was obtained, and cloned through two rounds of limiting dilution, resulting in anti-Bm-TPI clone 1.11.1. This was found to be an IgG1 using mouse antibody isotype kit (Isostrip, Roche). Monoclonal antibody was purified from culture supernatants using HiTrap protein G HP columns and an AKTAprime, and dialysed into PBS. Control mouse IgG1 MOPC31C was produced in the same way from cells sourced from ECACC.
BES and somatic extracts of L3, MF and mixed adult B. malayi (1 μg) were run on SDS-PAGE gels and blotted onto nitrocellulose membranes as previously described [6]. Following blocking in 5% milk powder/TBST (2 hours room temperature), membranes were probed overnight at 4°C with 1/500 mouse polyclonal anti-Bm-TPI, washed extensively in TBST and the incubated with 1/2000 rabbit anti-mouse Ig HRP (1 h, room temperature; DakoCytomation). Following further washing in TBST, blots were developed using ChemiGlow West (Alpha Innotech) and imaged using a FluorChem SP (Alpha Innotech).
Adult B. malayi were mounted in Cryo-M-Bed (Bright Instruments), frozen on dry ice, and 5 μm sections cut using a Leica CM1510S cryotstat. Air dried sections were fixed in 100% acetone (10 min), washed twice with PBS (20 min), and stained in a humidified chamber with 1/100 dilution mouse anti-Bm-TPI sera (generated as above) in 1% FCS/PBS (1 hour at room temperature). Control sections were similarly treated with naïve mouse sera. Following extensive washing in PBS, sections were incubated with 1/100 goat anti-mouse IgG TRITC (Sigma) as above, washed in PBS, then mounted with Vectashield (Vector labs). Sections were analysed with an Olympus BX50 fluorescent microscope and Openlab software (PerkinElmer).
Meriones unguiculatus jirds were immunized with 200 μg of Bm-TPI or BSA i.p. in alum adjuvant and then boosted with sub-cutaneous injections of 50 μg protein in alum at weeks 5 and 6. Jirds were challenged with 190 B. malayi L3 i.p. at week 8 post-immunisation. Infection was allowed to progress 8 weeks in the first experiment and 16 weeks in the second experiment. For Ls-TPI vaccination, BALB/c mice were immunised with 50 μg Ls-TPI or BSA in alum, and then boosted on weeks 4 and 5 with 25 μg protein in alum. Mice were infected sub-cutaneously on week 7 post-immunisation with 30 L. sigmodontis L3, and infections were terminated at week 10 post-infection.
Surgical implant of adult B. malayi (8 females and 2 males) into the peritoneal cavity of BALB/c mice was performed as previously described [50]. Mice were given 200 μg of anti-Bm-TPI clone 1.11.1 mAb or MOPC31C isotype control every 1–2 days over the course of a 14–28 day infection. Embryograms were performed on recovered female parasites exactly as [42]. For in vivo microfilariae transfer, Mf from the peritoneal cavity of infected jirds (1×105) were transferred i.v. in 200 μl RPMI1640 into BALB/c recipients, which were then injected with 200 μg antibody as above. Circulating Mf numbers were determined by tail bleed 24 hours later.
Cells were recovered from infected mice by peritoneal wash [38], washed into FACS buffer (PBS +0.5% BSA +0.05% sodium azide) Fc receptors blocked with 0.5 mg/ml rat IgG on ice for 10 minutes, then stained variously with anti-CD11b Pacific Blue (Biolegend; clone M1/70) anti-siglecF (PE or PE CF-594 conjugates, BD Pharmingen; clone E50-2440), anti-Ly6G APC-Cy7 (Biolegend; clone 1A8) and anti-F4/80 (FITC or PerCP conjugates, Biolegend; clone BM8). For macrophage alternate activation analysis, cells were fixed and permeablised (eBioscience, as per manufacturer's instructions) before intracellular staining with anti-RELMα (unlabeled rabbit polyclonal; PeproTech, followed by rabbit Ig labeling reagent; Invitrogen) and anti-Ym-1 (biotin-conjugated mouse chitinase 3-like 3; R&D, followed by streptavidin PE-Cy7; Biolegend). For intracellular cytokine staining, peritoneal cells were re-stimulated ex vivo in complete RPMI1640 media (supplemented with 10% FCS, 2 mM L-glutamine, 100 U/ml penicillin, 100 µg/ml streptomycin) with 1 μg/ml ionomycin, 500 ng/ml PMA and 10 μg/ml Brefeldin A (all Sigma) for 4 hours at 37°C. Following FcR block, cells were surface stained with anti-CD4 PerCP (clone RM4-5), fixed and permeabilised (BD Pharmingen Cytofix/Cytoperm) and then intracellular stained with anti-IFNg APC (clone XMG1.2) and anti-IL-4 PE (clone 11B11). Relevant isotype control stains were included. Alternatively, CD4+ cells were purified from the spleens of naïve BALB/c mice using MACS beads and columns (Miltenyi Biotec), according to the manufacturer's instruction, and stimulated in complete RPMI1640 for 3 days in the presence of 1 μg/ml anti-CD3 (clone 145-2C11) and 0.5 μg/ml anti-CD28 (clone 37.51) with varying amounts of recombinant Bm-TPI. Cells were then washed, resuspended in fresh media and stimulated with PMA and ionomycin in the presence of Brefeldin A as above. For in vivo transfers of ovalbumin-specific DO11.10 cells, BALB/c mice were injected i.p. with 2×10e6 DO11.10 splenocytes (equivalent to approx 4×105 CD4+ cells, data not shown). The next day, mice were given 0.5×106 dendritic cells i.p. pulsed overnight with LPS (100 ng/ml, Sigma) and ovalbumin peptide (pOVA) residues 323–339 (20 μg/ml, Invivogen), then subsequently injected on days 0, 1, 3 and 5 with 100 μg rBm-TPI or PBS control. Control mice were given PBS rather than dendritic cells. At day 7, spleens and peritoneal cells were harvested, stimulated as above and stained with biotin anti-mouse TCR DO11.10 (clone KJ1-26) followed by streptavidin-APC conjugate. Dendritic cells were generated in vitro from mouse bone marrow in the presence of GM-CSF [59]. Antibodies were from Biolegend unless stated.
Statistical significance was determined using Prism 6 (Graphpad Software Inc.). For comparison between two groups, unpaired Student's t-test or Mann-Whitney U-test was used dependent on data normality. Multiple comparisons used one-way ANOVA followed by Tukey's test.
|
10.1371/journal.pgen.1002435 | The Drosophila melanogaster Seminal Fluid Protease “Seminase” Regulates Proteolytic and Post-Mating Reproductive Processes | Proteases and protease inhibitors have been identified in the ejaculates of animal taxa ranging from invertebrates to mammals and form a major protein class among Drosophila melanogaster seminal fluid proteins (SFPs). Other than a single protease cascade in mammals that regulates seminal clot liquefaction, no proteolytic cascades (i.e. pathways with at least two proteases acting in sequence) have been identified in seminal fluids. In Drosophila, SFPs are transferred to females during mating and, together with sperm, are necessary for the many post-mating responses elicited in females. Though several SFPs are proteolytically cleaved either during or after mating, virtually nothing is known about the proteases involved in these cleavage events or the physiological consequences of proteolytic activity in the seminal fluid on the female. Here, we present evidence that a protease cascade acts in the seminal fluid of Drosophila during and after mating. Using RNAi to knock down expression of the SFP CG10586, a predicted serine protease, we show that it acts upstream of the SFP CG11864, a predicted astacin protease, to process SFPs involved in ovulation and sperm entry into storage. We also show that knockdown of CG10586 leads to lower levels of egg laying, higher rates of sexual receptivity to subsequent males, and abnormal sperm usage patterns, processes that are independent of CG11864. The long-term phenotypes of females mated to CG10586 knockdown males are similar to those of females that fail to store sex peptide, an important elicitor of long-term post-mating responses, and indicate a role for CG10586 in regulating sex peptide. These results point to an important role for proteolysis among insect SFPs and suggest that protease cascades may be a mechanism for precise temporal regulation of multiple post-mating responses in females.
| Proteases can destroy, activate, or otherwise modulate the function of other proteins. In seminal fluid, many proteins have to be activated or degraded after mating; proteolysis is an effective way to accomplish this because seminal fluid proteins act outside of the cell, where most other regulatory processes cannot be used. Despite the presence of proteases in the seminal fluid of many animals, nearly nothing is known about the kinds of processes they regulate. Here, we present evidence of a protease cascade in the seminal fluid of the fruit fly Drosophila melanogaster. This cascade involves two proteases that are activated during mating. Once in the female, the downstream protease acts on two other proteins that are important for ovulation and sperm storage. Interestingly, the protease at the top of the cascade, CG10586, is also required for other female post-mating responses, including egg laying and sperm usage, independent of the second protease. Thus, CG10586 might be a general regulatory switch used by the male to quickly activate many female responses after mating.
| Proteolysis regulators are a component of the seminal fluid of many animal taxa, including insects and other invertebrates [1]–[7], fish [8]–[10], birds [11], [12], and mammals [13]–[18]. However, the mechanisms by which seminal proteases act, and most of the processes they affect, in mated females are poorly understood.
A mechanism by which proteases may effect physiological responses is through proteolytic cascades. Because most proteases are synthesized as inactive zymogens and require the removal of a short N-terminal sequence for activation [19], a protease cascade can be rapidly set in motion without new protein synthesis. For example, in mammals a seminal protease cascade activates the protease prostate specific antigen (PSA), in order to rapidly liquefy the seminal clot formed following ejaculation (reviewed in [15]). The action of PSA is regulated, in part, by the protease inhibitor PCI (reviewed in [20]), which controls the timing and extent of liquefaction. Seminal clots are an important feature of the post-mating response in many animals [13], [21], [22].
Given the prevalence of proteolysis regulators in seminal fluid, it seems likely that they are involved in other processes whose effects may extend past the first few minutes after mating. The study of seminal fluid protease functions would benefit greatly from a genetic approach. Drosophila melanogaster provides an excellent system in which to study the roles of seminal fluid proteolytic proteins. Analysis of Drosophila seminal fluid proteins (SFPs) capitalizes on a wide range of available genetic tools, physiological and behavioral assays, and both a well-annotated genome and seminal fluid proteome. In addition, though individual SFPs, including proteases [23], are not generally well-conserved between distant taxa [24], [25], the biochemical classes into which SFPs fall are conserved between insects and mammals [21], [26], suggesting that mechanisms of action are likely to be conserved as well.
Approximately 18% of the proteins in the Drosophila ejaculate have been identified as predicted proteases or protease inhibitors [2], [27]. Mass spectrometry-based estimates indicate that the abundance of individual proteolysis regulators varies, with some being the most abundant proteins in the ejaculate (e.g. Acp62F) and others being the least abundant (e.g. CG10587) [2]. Most SFP predicted proteolysis regulators are either serine proteases or serine protease inhibitors with unknown functions [2], [28], [29], though a few other protease classes have also been identified [26], [30]–[32]. Proteolysis regulators have been identified as expressed in male reproductive tract tissues of Tribolium [7] and directly in the ejaculates of honey bees [3] and mosquitoes [33]. In crickets, a predicted trypsin-like serine protease in the ejaculate is important for inducing egg laying in mated females [5]. In the nematode Caenorhabditis elegans, a trypsin SFP has recently been reported to function in activation of male sperm [10]. Though proteases are emerging as a common SFP class in animals, there have been no studies determining whether protease cascades (i.e. proteolytic pathways that require at least two proteases in sequence) are a common regulatory mechanism for seminal fluid-mediated post-mating traits.
In Drosophila, transfer of SFPs from male to female during mating induces physiological changes in mated females [reviewed in 6]. Two of these changes are increased egg production and reduced receptivity to remating. These changes occur in two phases: short-term and long-term, both of which are necessary for optimal fertility. The short-term response (STR) occurs within 24 hours of mating and is solely dependent on the receipt of SFPs [34], including the prohormone ovulin [35], CG33943 [36], the sperm storage protein Acp36DE [37], [38], and the action of free sex peptide (SP) that is not bound to sperm [39], [40].
Long-term persistence of post-mating changes (the long-term response, or LTR) requires SP and multiple other SFPs, and the presence of sperm in storage [40]. SP binds to sperm during mating. Cleavage by an unknown trypsin protease(s) is required to release the active portion of SP from sperm within the mated female [41]. SP is gradually cleaved from stored sperm during the approximately two weeks that they remain in storage. As long as SP is released into the female, she continues to lay eggs at a high rate and is more likely to reject courting males [41]. If SP cannot be released from sperm, the LTR does not occur [41].
Fertility defects arise if SP cannot bind to sperm in the mated female, or if it cannot be released from sperm. Sperm binding by SP requires the action of at least four other SFPs: the predicted serine protease CG9997, the Cysteine Rich Secretory Protein (CRISP) CG17575, and the gene duplicate pair lectins CG1652 and CG1656 [36], [42]. These four “LTR proteins”, together with SP, function in an interdependent network to bind SP to sperm as well as to localize each other to the seminal receptacle (SR), the major sperm storage organ of the female [42]. In this network, CG9997 is cleaved into a 36-kDa protein in the male ejaculatory duct/bulb, prior to transfer to the female and is required for the normal transfer of CG1652 and CG1656. CG17575 is required to localize CG1652 and CG1656 to sperm and the SR. This final step is then required for SP to bind sperm and accumulate in the SR. If any one of the four LTR proteins is absent, SP does not bind sperm. These SP-free sperm are still stored in normal numbers, but cannot be efficiently released from storage for fertilization past the first 24 hours after mating [42], because SP is also required for sperm release [43].
In addition to SP, two SFPs involved in post-mating traits are known to be cleaved following deposition into the female. The prohormone ovulin is initially cleaved at about 10 minutes after the start of mating (ASM) [44]. Ovulin is required for a maximal ovulation rate in the first 24 hours following mating [35], [45]. Processing occurs via three cleavage events from the N-terminus of ovulin that ultimately results in the production of one major cleavage product (approx. 25 kDa) and three minor products (each 5 kDa or smaller) [44]. Ectopic expression experiments have shown that both full-length ovulin as well as two C-terminal fragments, roughly corresponding to cleavage products of ovulin, are each able to independently induce ovulation in virgin females [46]. The glycoprotein Acp36DE is also cleaved within mated females [47], starting at approximately 20 minutes ASM, as detected by Western blot [47], [48]. Acp36DE is required for efficient sperm storage [37], [48]. This protein is responsible for the conformational changes of the uterus immediately following the start of copulation, which are thought to aid the movement of sperm into the storage organs [38], [49].
A previous study of 11 SFP proteases and protease inhibitors identified CG11864 as required for processing of both ovulin and Acp36DE [32]. Though all three proteins are produced in the male accessory glands, ovulin and Acp36DE are not cleaved until several minutes after their entry into the female reproductive tract. Therefore, three possibilities exist for the regulation of ovulin and Acp36DE cleavage. CG11864 may be activated during mating, a repressor of CG11864 activity may be removed during mating, or a combination of both may occur.
CG11864 is predicted to be a member of the astacin family of metalloproteases, based on sequence similarity [32]. Astacin family proteases, like many other proteases, require removal of an N-terminal pro-peptide for activation [50]. The activity of CG11864 thus may be regulated in a similar manner. CG11864 is produced in the male accessory glands as a 33-kDa protein and is cleaved to an approximately 30-kDa form [32]. This cleavage begins in the male reproductive tract, in the ejaculatory duct and/or bulb, while CG11864 is in transit to the female during mating [32]. The size of the cleaved form of CG11864 is consistent with removal of a predicted pro-peptide from the N-terminus. We hypothesize that cleavage of CG11864 is required for its activation. If this is the case, there should be factors produced by the male that regulate the activation of CG11864. However, the previous study involving 11 SFP proteolysis regulators did not suggest their requirement for the regulation of CG11864 [32]. A recent microarray analysis by Chintapalli et al. [51] and subsequent proteomic studies by Findlay et al. [2] identified additional serine proteases in the ejaculate. We, therefore, focused on these proteases to test for roles in the activation/regulation of seminal proteolysis.
Here, we used RNAi knockdown analysis to test five male-derived serine proteases for roles in ovulin cleavage and other reproductive events. We describe the first proteolytic cascade in fly seminal fluid that is regulated by a predicted trypsin-like serine protease, CG10586. We propose to rename this enzyme seminase (gene symbol: sems). Seminase is required for cleavage, and likely activation, of CG11864. Like CG11864, seminase is produced in the accessory glands and is cleaved in the male during copulation. We show that CG11864 is not able to undergo self-cleavage in the absence of seminase. In addition to regulating CG11864 and thus its downstream SFP substrates, we show that seminase is a member of the LTR network, a CG11864-independent pathway that results in SP binding to sperm.
We tested five predicted protease SFPs for ovulin processing defects via Western blot to identify potential CG11864-interacting proteins. The tested SFPs were the predicted serine proteases ‘seminase’ (CG10586), CG10587, CG4815, CG12558, and CG32382 (sphinx2). Of these, only seminase, a predicted trypsin-type serine protease, was required for ovulin processing (Figure 1A). In addition to the results for seminase, we show for comparison the results for CG10587, which did not affect ovulin processing. The data for the other SFPs tested are not shown. Two independent insertion lines of the same RNAi construct were used to test the phenotype of seminase knockdown (see Materials and Methods); we obtained similar results with both lines. Western blotting confirmed that seminase is knocked down at least 98% by Tubulin-Gal4 driven expression of the RNAi construct in males of both lines (Figure S1A). Transcript levels of seminase were also confirmed to be knocked down by RT-PCR (Figure S1B).
Similar to the phenotype previously observed with CG11864 RNAi [32], some ovulin processing was observed in females mated to males knocked down for seminase, but ovulin was never processed fully in mates of seminase knockdown males, even at 2 hours ASM, the latest time at which ovulin can be reliably detected in female reproductive tracts when mated to controls (Figure 1A). Females mated to seminase knockdown males also failed to fully process Acp36DE at 1 hour ASM (Figure 1C), similar to CG11864 knockdown mates (Figure 1C). Even at 3 hours ASM, Acp36DE in females mated to seminase or CG11864 RNAi males had undergone only a small amount of processing relative to controls (Figure 1C).
Females mated to seminase RNAi knockdown males received CG11864 protein, but it was of the full-length molecular weight (33-kDa); the cleaved form (30-kDa) was never observed (Figure 1D). However, females who mated to control males received both full-length and cleaved CG11864 (Figure 1D). Thus, seminase is required for the predicted pro-peptide cleavage of CG11864 during mating.
Since many serine proteases are synthesized as zymogens (containing an N-terminal sequence that must be removed for activation), we tested whether seminase was also processed during or after mating. Seminase is detected as an apparent 29-kDa protein (predicted size: 28.2-kDa, excluding a predicted N-term secretion signal sequence) in the accessory glands, with no detectable expression in the testes or ejaculatory duct and bulb (Figure 2A). There was no evidence for seminase expression outside of the male accessory glands based on expression data in the FlyAtlas database [51] and our own RT-PCR (Figure S1B). We did not detect seminase protein in virgin females (Figure 2A).
During mating, an additional, lower molecular weight band (approximately 27-kDa) of seminase appeared in the male ejaculatory duct and/or bulb (Figure 2B), consistent with removal of a 2.79-kDa pro-peptide (size prediction based on an NCBI conserved domain search at http://www.ncbi.nlm.nih.gov/Structure/cdd/wrpsb.cgi?INPUT_TYPE=live&SEQUENCE=NP_649270.1). Within the mated female, seminase was further cleaved, producing a ∼16-kDa product (visible in top panel of Figure 1C) and increasing the amount of the ∼12-kDa product (Figure 2B). We first detected the ∼12-kDa product in the female at around 15 minutes ASM. However, due to the difficulty in detecting the ∼16-kDa form, we could not determine at what time ASM it is first produced.
Given that some proteases can be cleaved by their own proteolytic substrates [for example, see 52], we tested whether knockdown of CG11864 affected processing of seminase after mating. Females mated to CG11864 knockdown males showed normal processing of seminase at 30 minutes ASM (Figure 2C, top panel). As expected [32], CG11864 knockdown does prevent ovulin cleavage (Figure 2C, bottom panel), indicating a unidirectional proteolytic pathway.
The total number of eggs laid over 10 days was significantly lower in females mated to seminase knockdown males relative to their controls, and this was seen in both independent insertion lines (Line 1 Poisson regression: z = −29.56, p<0.0001, Figure 3A; Line 2 Poisson regression: z = −31.59, p<0.0001, Figure S1A). There was no difference in total number of eggs laid between females mated to CG4815 knockdown and control males (CG4815 Poisson regression: z = −0.79, p = 0.43, Figure 3A).
Eggs laid by seminase or CG4815 knockdown mates hatched in significantly larger proportions than eggs laid by control mates (Binomial regressions: Line 1: z = 10.1, p<0.001 and CG4815: z = 5.8, p<0.0001, Figure 3B; Line 2: z = 13.2, p<0.0001, Figure S1B), suggesting the egg laying defect was not accompanied by a hatchability defect (hatchability is defined as the proportion of eggs that produced adult progeny), but rather that there was a slight deleterious effect of the balancer control background on hatchability.
A repeated measures analysis of egg laying over time revealed a significant effect of male genotype on egg laying over time for both seminase lines and CG4815 (see Materials and Methods). To determine the days on which male genotype affected egg laying, data were analyzed separately for each individual day. The decrease in egg laying, relative to control, in mates of seminase knockdown males was only apparent after the first day following mating and persisted until at least 9 days post-mating (Figure 3C and Figure S1C). Females mated to seminase knockdown males laid slightly, though significantly, more eggs than females mated to control males on day 1, but only with seminase Line 1 males (Figure 3C). These results indicated that seminase only has a major role in egg laying after the first day post-mating.
Females mated to CG4815 knockdown males laid significantly fewer eggs than females mated to control males on day 1 and slightly, though significantly, more eggs than controls on day 9 (Figure 3D). Thus, CG4815 may have a short-term effect on egg laying, but no long-term effect similar to that observed with seminase knockdown mates. These results also indicate that the long-term egg laying effect of seminase is not an artifact of the VDRC strain background, which is shared by the CG4815 males.
A reduction in fecundity after the second day post-mating suggests seminase is a new member of the LTR pathway. Increased recovery of post-mating receptivity to courtship beginning after the first 24 hours post-mating is also associated with LTR defects. Therefore, we tested whether deficiency of seminase in the ejaculate also caused increased receptivity in females, relative to controls, after mating. Table 1 shows data for female receptivity at 24 hours, 2 days, and 4 days ASM to seminase knockdown or control males. Similar to previously described phenotypes of LTR SFPs [36], [42], [43], females mated to seminase knockdown males were significantly more likely to remate at 2 days and 4 days ASM than were controls. Females mated to males from seminase knockdown line 1 showed a smaller magnitude of difference in remating rate relative to their controls than did line 2. This is most likely due to a background effect in line 1 that is apparent in the control males, as the remating rate is similar in mates to knockdown males from both lines. Since the same RNAi construct is expressed in both lines, we assume that the higher remating rate for females mated to line 1 control males is due to the insertion locus of the transgene. There was no effect of CG4815 knockdown on receptivity (Table 1).
In Drosophila females, sperm are stored in two types of storage organs: the seminal receptacle (SR) and the paired spermathecae. The bulk of the sperm are stored in the SR [53]. Because LTR SFPs (in concert with SP) affect the release of sperm from storage [36],[43], we tested whether mates of seminase knockdown males also showed a defect in sperm release. Mates to seminase knockdown males stored normal numbers of sperm (Figure 4 “2 h” bars), but significantly more sperm remained in storage at 10 days ASM in mates of seminase knockdown males than in control-mated females (Figure 4A). This effect was due to a failure to release sperm from the SR (Figure 4B), as sperm numbers in the spermathecae decreased at similar rates in females mated both to control and seminase knockdown males (Figure 4C). A slight, but significant, difference in sperm release was seen in the spermathecae at 4 days ASM, but this effect was in the opposite direction from that seen in the SR and was no longer apparent by 10 days ASM (Figure 4C). Similar effects were seen with seminase knockdown Line 2 (Figure S1D).
The results above are consistent with seminase being a member of the LTR network. To determine the placement of seminase in this network, we tested whether knockdown of seminase affected the post-mating localization of SP and the three LTR proteins that localize to the SR: CG9997, CG1652, and CG1656. At 2 hours ASM, seminase was required for accumulation of SP, CG1652, and CG1656 in the SR (Figure 5A). CG9997 was not detected in the SR at 2 hours ASM, so we tested females at 1 hour ASM. Seminase was also required for accumulation of CG9997 in the SR at this time point (Figure 5A). However, seminase was not required for proper processing of CG9997 or transfer of any LTR SFPs to the female during mating (Figure 5A, RT lanes), including CG17575 (Figure 5B).
A small amount of seminase also enters the SR (Figure 5C), suggesting that it could physically interact with other LTR proteins there. However, we were unable to determine whether other LTR proteins affected seminase localization to the SR due to the extremely low seminase signal within the SR. Multiple repetitions of the experiment failed to yield consistent results. As with previous efforts to detect LTR proteins in the spermathecae [42], we were not able to detect seminase in these organs (data not shown).
To identify proteins that may interact with the predicted astacin-family protease CG11864 to process the SFPs ovulin and Acp36DE, we used RNAi to individually test five serine protease SFPs for ovulin processing defects. One of these proteins, the predicted trypsin-type serine protease ‘seminase’ (CG10586), is required for normal processing of ovulin as well as of the sperm storage protein Acp36DE. Because the phenotype of seminase knockdown was similar to that of CG11864 knockdown with respect to SFP processing, we hypothesized that both proteins might act in a single pathway. Additionally, because trypsin (serine) proteases are required for activational cleavage of some astacin-family proteases (of which CG11864 is one) [50], [54], we further hypothesized that seminase might act upstream of CG11864.
We therefore tested whether seminase regulates the cleavage, and thus activation, of CG11864. We found that seminase is required for the approximately 3-kDa mobility shift of CG11864 that is seen in the male reproductive tract very soon after mating begins, suggesting that seminase may activate CG11864 by cleaving its pro-peptide. The apparent processing of pro-CG11864 by seminase and the subsequent processing of downstream substrates is suggestive of a proteolytic cascade. No such proteolytic pathway has, to our knowledge, previously been identified in insect seminal fluid. With the identification of this pathway in Drosophila melanogaster, we have found a molecular model for dissecting proteolytic pathways involving SFPs that have consequences for fertility.
Proteolytic cascades, in their simplest form, typically have three steps [55]: 1) auto-activation of an initiator protease(s) present in low amounts and triggered by an external stimulus: 2) activation of a more abundant propagator protease(s) by the initiator protease; 3) activation of an executor protease(s) by the propagator, which will cleave the downstream substrates. In addition, the propagator may also cleave, and thereby continue to activate, the initiator. Altogether, this causes a rapid propagation of the initial external signal.
While protein abundance is not necessarily related to potency, it is intriguing that seminase (the putative initiator) is relatively scarce in the ejaculate of D. melanogaster, which is consistent with the above model. Abundance estimates are based on the normalized spectral abundance factor (NSAF) obtained by mass spectrometry on mated females [2]. NSAF is an approximate measure of the relative abundance of a protein in a complex sample. Seminase ranks at 130 out of 138 (NSAF = 1.34×10−4), with 1 being the most abundant and 138 the least [2]. CG11864 is similarly scarce (87/138; NSAF = 7.69×10×−4). This is in contrast to the much higher abundance of CG11864's substrates (ovulin: 20/138, NSAF = 1.2×10−2; Acp36DE: 19/138, NSAF = 1.21×10−2).
Also consistent with the protease cascade model, seminase is cleaved to a slightly smaller form during mating while still in the male reproductive tract. This may be an activational pro-peptide cleavage event, though this has not been directly tested. It is possible that seminase self-activates upon entering the ejaculatory duct, as is the case for many serine proteases [for example, among tissue kallikrein pathways; see 15]. Our data suggest that seminase acts as the initiator in the cascade, and CG11864 acts either as the propagator, the executor, or both.
After transfer, seminase itself undergoes additional processing in the female (after the initial pro-peptide cleavage in the male) that is not a result of CG11864 activity (Figure 2C). These cleavage products may be important for the function of seminase. On the other hand, they may simply be degradation products of seminase. However, both scenarios remain speculative.
We have shown that, in the absence of seminase, CG11864 is not cleaved to the predicted active form. However, the predicted pro-peptide cleavage site of CG11864, based on sequence threading to other astacin-family proteases [32], is not a trypsin site, as would be predicted if seminase were the only protease responsible for CG11864 activation. Interestingly, there are three trypsin cleavage sites present in the pro-peptide region of CG11864. It is possible that CG11864 is cleaved via a two-step mechanism (involving a trypsin and CG11864 itself), as is seen for the pro-peptide cleavage of Astacus astacus (crayfish) astacin, the prototype of the astacin family [54], [56]. Future studies using purified proteins in vitro will determine whether CG11864 is capable of self-cleavage and whether seminase acts to directly cleave CG11864.
Despite a severe delay in ovulin processing, knockdown of neither seminase nor CG11864 results in an egg laying defect in the first 24 hours after mating [CG11864 data reported in 32]. This result is not surprising, however, given the rather small effect on egg laying seen with a complete knockout of ovulin [45]. Additionally, ectopic expression of full-length ovulin is sufficient to induce ovulation in virgin females [46], suggesting that the additional effect of ovulin processing may be too small to detect with the current assay. It is also possible that, while seminase was knocked down to very low levels, there may still be sufficient seminase present for its role in early egg laying.
We also do not observe a defect in sperm entry into storage following seminase knockdown, as seen with knockout of Acp36DE [37]. Instead, defects were seen in sperm release from storage at later timepoints, which is not a phenotype associated with Acp36DE knockout. Acp36DE processing may be important for other functions of this protein. For example, Acp36DE is a component of the mating plug [57], but its function in the mating plug is still unknown. Further research is required to determine the consequences of loss of proteolytic processing of both ovulin and Acp36DE.
In contrast to CG11864, which seems specific to the STR, seminase has a second important activity: it is in the LTR pathway, which regulates the binding of sex peptide (SP) to sperm [36], [41]–[43] (See Figure 6 for overview). Similar to mates of SP null males, females mated to seminase knockdown males lay fewer eggs than controls over a 10 day period and also retain sperm in storage. The sperm retention phenotype is only apparent in the SR. Other LTR proteins are also known to affect sperm storage in the SR but not the spermathecae [36], though the reason for these differences is not understood. While the interaction between sperm release and egg laying is complex, over the long-term, egg laying and sperm release are independent of each other [58]. Sperm do not directly influence the release of eggs, though the presence of SP bound to sperm is required for both sperm release [43] and normal post-mating levels of egg laying [39], [40]. The failure of SP to accumulate in the SR indicates that seminase is likely required for SP to bind sperm.
The requirement for seminase in two independent post-mating pathways suggests that its activation at mating may act as a regulatory “switch” that coordinates post-mating events in Drosophila. Identification of other seminase substrates, if they exist, will allow us to determine the extent of seminase's effects as a regulatory switch for post-mating events.
In the context of evolution, SFPs represent a unique class of proteins in that they must, first and foremost, aid in successful fertilization, but are also tasked with representing the male's reproductive interests, sometimes in the face of opposing female interests [reviewed in 59]. This has the potential to set up a genetic conflict between the sexes and has been suggested to be one reason that SFPs in particular tend to be rapidly evolving [24], [25]. However, an SFP's evolutionary rate is also likely to be constrained by the need for the protein to maintain its interaction with other proteins in the seminal fluid, and/or with proteins expressed by the female. For example, seminase and CG11864 are processed first in the male, but must interact with the female environment to further process both seminase and the substrates of the proteolytic pathway. Seminase-regulated processes represent an opportunity to understand the evolution of SFP networks that contain a mixture of conserved proteins (e.g. the lectins CG1652 and CG1656 and the CRISP CG17575 [2]) and proteins under positive selection (e.g. ovulin [60] and the serine protease CG9997 [61]). Seminase itself shows no evidence of positive selection, either at the protein level [61] or at individual sites (personal communication, Geoff Findlay). However, seminase does have two very closely related SFP paralogs, CG11037 and CG10587, which show evidence for recent positive selection in the D. melanogaster lineage [61]. These three genes are clustered together in the genome of D. melanogaster and the other melanogaster subgroup species. CG10587 does not play a role in ovulin or Acp36DE processing (CG11037 has yet to be tested), suggesting that these genes arose from tandem duplications and later diverged in function, with seminase remaining as the more conserved of the paralogs.
Our data on seminase show that this member of a conserved protein class in the seminal fluid plays a vital role in reproductive success. We believe that future study of the seminase-regulated pathways in Drosophila will lead to new mechanistic and evolutionary insights related to proteolytic cascades and protein networks in seminal fluid.
The regulation and mechanism of action of seminase constitutes a new in vivo model system for studying the regulation and physiological roles of pleiotropic SFPs. Pleiotropic effects of kallikrein-related proteases involved in the liquefaction of human semen have recently been reported [62], [63]. Further understanding of the various effects of seminal fluid proteolysis in post-mating processes may have important implications for human health (e.g. the role of PSA in cancer) and fertility. Our results indicate that genetic analysis in Drosophila will be an important complement to in vitro studies in mammalian systems for understanding the role of proteolytic processing in reproduction. Future studies of the regulatory mechanisms involved in the seminase/CG11864 proteolytic pathway may generate testable hypotheses for other SFP networks, including those in mammals.
Transgenic lines carrying RNAi constructs for CG10586 (‘seminase’) and CG4815 were purchased from the Vienna Drosophila Resource Center (VDRC, http://stockcenter.vdrc.at) GD RNAi library (P element library). VDRC lines used correspond to the following transformant ID numbers: 18795, 18796 (CG10586; same construct (ID 5539), different insertions), and 15410 (CG4815). CG10586-18795 is referred to here as “Line 1” and CG10586-18796 as “Line 2”. CG10586 VDRC lines are predicted to have an off-target, CG33306. However, we found no evidence for CG33306 knockdown in either line (Figure S2B, only Line 1 shown). All RNAi knockdown and control sibling flies were produced by crossing sympUAST-SFP or VDRC virgin females to ubiquitous driver males (Tubulin-Gal4/ TM3). UAS-RNAi / Tubulin-Gal4 (non-balancer) male progeny were knocked down for the SFP of interest and the sibling UAS-RNAi/TM3 (balancer) flies were used as controls that are wildtype for seminase expression.
Matings were carried out by placing single 3–6 day old virgin females of the Canton-S strain with a single 3–6 day old virgin male in a glass vial containing a moistened square of filter paper. Matings were observed and the time at which mating began was recorded. Mating pairs with unusually short matings (<15 minutes) were discarded. Mated females were flash frozen in liquid nitrogen at the appropriate time after the start of mating (ASM) for time points less than one hour ASM and stored at −80°C until dissection. All flies were reared in standard yeast-glucose media at room temperature (23±1°C) on a 12∶12 light/dark cycle.
Generation of seminase fusion proteins and antibody purification was done following Ravi Ram et al. [64] and Cui et al. [65]. Briefly, we generated a 6×His fusion protein containing amino acids 101–200 from seminase-PA using the pDEST17 vector of the Gateway system (Invitrogen). Antibodies were generated in rabbits (Cocalico) as described previously for eight other Drosophila reproductive proteins, including CG11864 [64], and Wisp [65] except that rabbits were immunized with the 6×His-seminase fusion protein. Anti-seminase was affinity purified with a GST fusion protein of amino acids 101–200 of seminase, as described in the above references. Eluted antibodies were stored at −20°C in glycerol (1∶1), and used at a concentration of 1∶250 for Western blot analysis.
Sample preparation and Western blot analyses in Figure 1A and Figure 5A and 5B were carried out as in Ravi Ram and Wolfner [42]. Samples in other Westerns in this study were prepared similarly, except that they were separated using 5–15% gradient SDS/PAGE. Female reproductive tract samples (RT) are lower reproductive tract extracts (ovaries removed) from 4–6 mated females, unless otherwise noted.
A BCA (bicinchoninic acid) assay (Pierce BCA Protein Assay Kit, Thermo Scientific) was performed to determine the protein loading in Figure 1A. Samples identical to those used for the Western blot were prepared and protein concentration measured relative to a BSA standard, following the manufacturer's guidelines.
Number of eggs laid daily by mated females (fecundity) and number of progeny produced from those eggs (fertility) were quantified as in Ravi Ram and Wolfner [36]. Assays for the effect of seminase on fecundity and fertility were carried out three times for each independent insertion line, each time with 15–24 females measured for both knock down and control treatments. CG4815 knock down- and control-mated females were also measured for both fertility and fecundity, as a control for the VDRC background. Two assays were carried out with this line, each time with 7–12 females measured for each treatment. Hatchability was determined by dividing number of progeny by number of eggs (fertility/fecundity) [36]. Inspection of the data revealed non-significant variation in egg laying due to experimental block, so data were pooled across blocks. The effect of seminase or CG4815 knockdown on total 10-day egg laying was tested with a Poisson regression model (using the R function ‘glm()’. The statistical tests for hatchability were the same, except a Binomial regression was used.
A repeated measures analysis was performed to determine the effect of male genotype over time. This analysis was performed using a Poisson mixed-effects model with the R function ‘lmer()’ in the lme4 library. Two models were compared, a full model with day, genotype, and day-by-genotype interaction as fixed effects and female as the random effect, and a model with day as the only fixed effect. Comparison of the two models by ANOVA (R function ‘anova()’) revealed the full model was the better fit, indicating a significant effect of male genotype. To determine the statistical significance of male treatment on individual days post-mating, a Bonferroni correction for multiple tests was applied to the Poisson regressions. All plots were generated using the means and standard errors of the raw data pooled from all experiments. Statistical significance of the effect of male genotype on number of eggs laid is denoted by asterisks on the plots.
Females who had previously mated with either control or seminase knockdown males were placed with a single wildtype male of the Canton-S strain for one hour either 24 hours, 2 days, or 4 days following the initial mating and the number of copulations beginning within one hour was recorded. Receptivity response to remating was tested for seminase as in Ravi Ram and Wolfner [36]. No fewer than 10 females were analyzed for control and experimental groups at 24 hours, 2 days, or 4 days after initial mating. Data were analyzed using a Chi-squared test (R function ‘chisq.test()’ with all parameters set to default).
Sperm counts were performed as in Avila, et al. [43] at 2 h ASM, 4 days, and 10 days ASM. Sample identity was coded to avoid bias and each slide was counted twice to assess counting precision (85%–100%). SR data used are the average of the two counts. Spermathecae data are the sum of the two averages (one for each spermatheca). For 4 and 10 day post-mating samples, individual female daily egg counts were also taken. Numbers of stored sperm at 4 and 10 days ASM were significantly negatively correlated with the number of eggs laid (Figure S2). Females that laid very few eggs on day 1 (less than 2 standard deviations below the mean) were removed from the dataset as they were likely unhealthy and may have had improper sperm storage. Data were analyzed using a two-tailed Student's t-test (R function ‘t.test()’).
|
10.1371/journal.pgen.1004985 | A Rolling Circle Replication Mechanism Produces Multimeric Lariats of Mitochondrial DNA in Caenorhabditis elegans | Mitochondrial DNA (mtDNA) encodes respiratory complex subunits essential to almost all eukaryotes; hence respiratory competence requires faithful duplication of this molecule. However, the mechanism(s) of its synthesis remain hotly debated. Here we have developed Caenorhabditis elegans as a convenient animal model for the study of metazoan mtDNA synthesis. We demonstrate that C. elegans mtDNA replicates exclusively by a phage-like mechanism, in which multimeric molecules are synthesized from a circular template. In contrast to previous mammalian studies, we found that mtDNA synthesis in the C. elegans gonad produces branched-circular lariat structures with multimeric DNA tails; we were able to detect multimers up to four mtDNA genome unit lengths. Further, we did not detect elongation from a displacement-loop or analogue of 7S DNA, suggesting a clear difference from human mtDNA in regard to the site(s) of replication initiation. We also identified cruciform mtDNA species that are sensitive to cleavage by the resolvase RusA; we suggest these four-way junctions may have a role in concatemer-to-monomer resolution. Overall these results indicate that mtDNA synthesis in C. elegans does not conform to any previously documented metazoan mtDNA replication mechanism, but instead are strongly suggestive of rolling circle replication, as employed by bacteriophages. As several components of the metazoan mitochondrial DNA replisome are likely phage-derived, these findings raise the possibility that the rolling circle mtDNA replication mechanism may be ancestral among metazoans.
| Defects in the mitochondrial DNA (mtDNA) that encodes protein subunits of the respiratory complexes may cause severe metabolic disease in humans. Such defects are often caused by errors during mtDNA synthesis, motivating ongoing studies of this process. The nematode Caenorhabditis elegans has been proposed as a model for the study of mtDNA replication defects. Here we analyze the mechanism of mtDNA synthesis in the C. elegans gonad and demonstrate that it is unique among animals. Nascent worm mtDNA forms branched-circular lariat structures with concatemeric tails that we suggest would ultimately resolve into monomeric circles, the predominant molecular form identified by both transmission electron microscopy and two-dimensional gel electrophoresis. Our discovery that mtDNA replication in C. elegans does not faithfully model that in mammals is significant, because it demonstrates the breadth and evolutionary plasticity of the mechanisms that maintain this critical DNA among animals. Interestingly, the mtDNA replication mechanism within C. elegans is highly similar to that of bacteriophages, from which components of the mitochondrial DNA replisome are thought to be derived. Thus C. elegans may serve as a model for mtDNA synthesis as it occurred within ancient eukaryotes.
| Caenorhabditis elegans is a ubiquitous model animal often employed in studies of aging and metabolic disease, processes intimately associated with mitochondrial health. However, comparatively little is known of mtDNA maintenance in this organism [1,2].
Early studies of mitochondrial DNA (mtDNA) replication in mammalian cultured cells supported a unidirectional strand displacement or ‘asymmetric’ model, producing partially single-stranded-DNA (ssDNA) intermediates [3,4]. More recently, strand-coupled ‘theta’ replication has been proposed [5], and support has also amassed for a temporally asynchronous mode of replication involving provisional RNA Incorporation ThroughOut the Lagging Strand (RITOLS), a model in which expanding replication bubbles contain RNA:DNA hybrid tracts [6]. The previously described animal mtDNA replication models share two features. First, initiation of replication relies on elongation from a transcript-primed displacement loop (D-loop). Second, each successful synthesis cycle from the circular template results in only two circular daughter molecules. The previous work on mtDNA synthesis has focused primarily on mammalian species; mtDNA maintenance elsewhere in the animal lineage remains poorly understood.
MtDNA is required for nematode development beyond the early larval stages, and perturbations causing mtDNA depletion during embryonic development commonly result in a larval arrest phenotype [7]. The mitochondrial complement in somatic cells of the adult nematode appears largely to result from distribution of the approximately 100,000 maternal mtDNA molecules throughout the embryo during development, precluding a need for the mitochondrial polymerase POLG-1 during development [8]. Moreover, mtDNA copynumber tends to fall in ageing worms, suggesting minimal turnover in the somatic tissues [8]. These findings are consistent with mtDNA replication occurring primarily in the adult gonad, with the integrity and quantity of mtDNA produced reflected in the subsequent generation [8]. The confinement of ongoing mtDNA replication to the germline makes C. elegans a convenient model for studies of mitochondrial genome synthesis and mtDNA replication defects [1,9].
C. elegans mtDNA harbors two non-coding regions (NCRs), delimiting coding regions of 5.5 and 7.7 kb respectively (Fig. 1A; [10]). By analogy with the mammalian mtDNA organization, both NCRs have been proposed to play a role in C. elegans mtDNA replication, one as the first-strand origin (akin to the mammalian D-loop) and the other serving as a second-strand origin [1,7]. To test this assumption, we investigated the mechanism of C. elegans mtDNA replication in vivo and the possible function of the two NCRs therein.
To determine if either NCR could function as a replication origin, we examined mtDNA fragments containing each one of the two NCRs for origin activity using two-dimensional neutral agarose gel electrophoresis (2DNAGE) [11]. Y arcs formed by progressing forks, as well as cruciform structures, were readily apparent (Fig. 1B, ClaI/ApaI and BsrGI/ClaI). Analysis of replication intermediates (RIs) derived from restriction fragments lacking both NCRs also revealed full Y arcs and cruciforms, consistent with active replication of the entire mtDNA (Fig. 1B, ApaI/BsrGI). However, a bubble arc indicative of theta-type replication initiation was not detected from any region of the genome, even after long autoradiographic exposures (S1A–S1E Fig.).
We next considered that first-strand replication initiation might occur from more than one site in the genome. Such low frequency bubble intermediates may go undetected by fragment 2DNAGE, as only a subset of intermediates are analyzed in each experiment [4,6,11]. To address this possibility, we performed 2DNAGE analysis after digestion with restriction enzymes cutting only once in the mitochondrial genome. We reasoned that analysis of RIs spanning the complete mtDNA would pool molecules containing D-loop initiation structures along a single arc regardless of initiation site, facilitating detection. Consonant with our 2DNAGE data on sub-genomic fragments, linearization and subsequent 2DNAGE of the full-length genome demonstrated clear Y and X shaped intermediates (S1F–S1J Fig.), yet no bubble arc was observed. We therefore conclude that initiation of C. elegans mtDNA synthesis does not involve the formation of a bubble intermediate at levels detectable by blot-hybridization.
To determine if molecular signatures of replication initiation, such as skewed nucleotide composition, are present in the non-coding region of C. elegans mtDNA, we conducted a bioinformatic analysis of cumulative GC skew (S1K Fig.; [12]). Unlike the D-loop regions of human and mouse mtDNA, our analysis demonstrates that the non-coding region suggested to harbor a D-loop in C. elegans is absent of local minima or maxima, considered features of origin and termination activity respectively. This finding is consistent with the lack of classic initiation (bubble) arcs on 2D-NAGE gels of replication intermediates.
Prominent initiation intermediates have been described in analyses of mtDNA isolated from dissected human, mouse and chick tissues [5,13,14]. However, in the worm the vast majority of mtDNA replication is expected to occur specifically in the germline. Therefore, we tested whether a minor fraction of bubble initiation structures from somatic cells would become detectable when germline development was blocked. We isolated mtDNA from synchronized glp-4 mutants which exhibit deficient germline nuclei production at 25°C (allele glp-4(bn2); [15]), and compared nematode cohorts reared at permissive and non-permissive temperature by 2D-NAGE. As expected, based on the work of Bratic et al [8] replication intermediates in glp-4(bn2) animals cultured at 16°C were comparable in structure and intensity to those detected in wildtype N2 animals. In contrast, at non-permissive temperature, i.e., in “gonadless” worms, RI levels relative to total mtDNA were dramatically decreased to near the limit of detection by Southern hybridization (S2A–S2B Fig.). Bubble intermediates were not detected on exposures ranging from 1 hour to 14 days, confirming the absence of replication elongation from a D-loop from both germline and post-mitotic cells. The lack of bubble intermediates effectively excludes strand-coupled initiation from a D-loop, i.e. the theta replication mode [4].
We next investigated whether C. elegans mtDNA RI structure was consistent with the asymmetric strand-displacement or RITOLS models of mtDNA replication. Temporally asynchronous replication of the two template strands is predicted to generate partially single-stranded RIs [4,16] and references therein]. In 2DNAGE, such ssDNA regions block endonuclease cleavage, producing slow-migrating Y arcs greater than twice the unit length fragment size, and render RIs sensitive to the action of the single-strand specific nuclease S1 [4]. In contrast, RNA:DNA hybrid-containing RIs typical of the RITOLS mode can be detected based on their sensitivity to degradation by RNase H, which exposes ssDNA regions rendered sensitive to S1 nuclease [17,18]. These treatments are expected to dramatically alter the electrophoretic migration properties of RITOLS intermediates in 2D gels [4,14]; such RNA:DNA hybrid-containing intermediates represent a transient step in replication, preceding synthesis of the definitive lagging strand.
We tested for strand-displacement and/or RITOLS intermediates by systematic treatment of mtDNA fragments, collectively representing the complete mitochondrial genome, with S1 nuclease, RNase H, or RNase H followed by S1 nuclease, with subsequent analysis by 2DNAGE. For each replicate, equal amounts of purified mitochondrial nucleic acid were electrophoresed for each treatment condition on the same gel, then transferred and hybridized in parallel with the same preparation of radio-labeled probe. For all mtDNA fragments analyzed, both the Y arc and X arc (cruciform spike) persisted after treatment with either S1 or RNase H alone (Fig. 2A, see also S2C–S2H Fig.). The intensity of the Y arc hybridization signal was modestly decreased by treatment with RNase H, yet the majority of fragment RIs remained following subsequent treatment with S1 nuclease (Fig. 2A; quantification in Fig. 2B) and, importantly, were not converted to any other structure, demonstrating distinct electrophoretic migration.
These experiments indicate that C. elegans RIs lack the extensive ssDNA character expected from strand-displacement synthesis. Furthermore, slow-moving Y arcs were not observed, and depletion of the Y arc signal after treatment with RNase H and S1 was no greater than with RNase H treatment alone (Fig. 2B). Thus RNase H failed to ‘unmask’ substantial ssDNA regions, as would be expected if extensive RNA:DNA hybrid tracts were present. These data are consistent with synchronous (or very near-synchronous) replication of the two mtDNA strands independent of a D-loop, and eliminate from further consideration both asymmetric and RITOLS-mode strand-displacement replication.
The absence of theta-form, RITOLS and partially ssDNA strand-displacement intermediates led us to consider alternate DNA replication mechanisms. The detection of Y arcs, but not bubble arcs, by 2DNAGE of fragments derived from a circular template is consistent with a rolling circle replication (RCR) mechanism [19,20]. According to the RCR model, sustained elongation on a circular template produces linear DNA molecules greater than template unit-length that may become resolved to monomers in a variety of ways, or remain concatemeric linear networks [21]. A central prediction of the RCR model is the presence of “lariat” DNA forms in vivo. For C. elegans mtDNA, we hypothesized the occurrence of one-genome unit length circular templates, from which multimeric linear tails would extend. Alternatively, a second replication mode can be envisaged that would involve strand-invasion of a linear template by linear molecules, as occurs in some bacteriophages and the mtDNA of the fungus Candida albicans [22,23]. This alternative would predict Y-form RIs in the absence of bubble RIs, but not lariat structures.
To determine whether lariat molecules consistent with rolling circle intermediates were present, we directly examined C. elegans mtDNA using transmission electron microscopy (TEM). We observed both circular and branched-circular lariat molecules (Fig. 3, S1 Table). Most prominent were dsDNA circles with a mean measured length of 13.61 kb +/- .407 kb, consistent with the sequenced mitochondrial genome size of 13.794 kb [10] (Fig. 3A, B). Although C. elegans mtDNA has long been thought to be circular [10], based on its restriction map, to our knowledge this is the first evidence that non-replicating C. elegans mtDNA exists in a topologically circular form. As predicted by our 2DNAGE and bioinformatic analyses, none of these circular mtDNAs contained a visible displacement loop. Lariats with linear tails ranging from < 1 kb to 48.2 kb in length, i.e., more than three genome units, were the next most frequently observed class of molecules (Fig. 3C, D; S1 Table). The mean length of the lariat circular portion measured 13.64 kb (Fig. 3A). Fifty-six percent of lariat molecules appeared fully double-stranded at the circle-branch junction (Fig. 3E, F), though ssDNA tracts of less than ∼500 bases are not readily visible under the imaging conditions used.
While the X arc intermediates described above are intense on fragment 2DNAGE, their full structure, when undigested by restriction enzymes, is unknown. To further address the structure of the cruciform mtDNA, we isolated X-arc intermediates from a second-dimension gel on which ClaI/ApaI digested C. elegans mtDNA was fractionated. However, when spread for TEM, microfragments of agarose remained bound to the DNA, compromising the visualization of these molecules and precluding their further characterization. In non-fractionated spreads of purified mtDNA such forms would be indistinguishable from cases where two independent molecules are incidentally in contact.
Some linear molecules were observed by TEM. However, sub-genomic linear molecules were not detected by Southern blot of C. elegans mtDNA using any of the probes described in this study (S2 Table). We conclude that linear molecules on TEM grids are most likely contaminating nuclear DNA fragments, and therefore excluded them from the analysis summarized in S1 Table.
A subset of lariat molecules contained visible interspersed regions of collapsed secondary structure, considered diagnostic for ssDNA (Fig. 3G) [24]. Within individual molecules, these regions occurred at several different positions: the junction of the circular and linear tail portions of lariats, further along the linear branch only, or in both locations (Fig. 3H, I). Neither strand-displacement nor theta RIs were observed among the 1262 molecules analyzed by TEM, while lariats made up approximately 4% of mtDNA molecules, in line with previous reports of the proportion of replicating mtDNA in other species, e.g. mammalian cells and Drosophila melanogaster [3,25].
The high frequency of non-replicative circular monomers we detected by TEM suggested inter-conversion between the circular monomer and the lariat mtDNA forms. In the course of fragment 2DNAGE, we noted that the hybridization intensity of cruciform structures was most prevalent, relative to the monomer spot, in fragments harboring the 465 bp ‘major’ NCR (Fig. 1). This observation suggested that the formation of a site-specific cruciform structure could play a role in the maintenance of C. elegans mtDNA and/or the production of monomer circles [21,26].
We further addressed cruciform architecture by 2DNAGE following treatment with RusA, an Escherichia coli resolvase highly specific for Holliday junctions substrates in in vitro studies [27,28]. This analysis was performed using RusA alone or in combination with S1 nuclease. RusA treatment reduced the hybridization signal of the cruciform spike by 48% relative to the untreated controls, while S1 nuclease alone had no significant effect (Fig. 4A, B). Treatment with S1 after RusA further reduced the cruciform signal, and revealed a subclass of cruciforms resistant to both RusA and S1 that persisted after the combined treatment (Fig. 4A). On 2DNAGE these molecules formed a near-vertical spike (Fig. 4C), a migration pattern typical of hemicatenanes [29]. It has previously been reported that the collapse of adjacent four-way junctions produces resolvase-resistant hemicatenanes [29,30]. These findings are reminiscent of the RusA-resistant mtDNA cruciforms we observe; whether these X-junctional molecules are intermediates in the mechanisms of mtDNA RCR or monomer resolution awaits further investigation.of mtDNA RCR or monomer resolution awaits further investigation.
Taken together, our findings indicate that synthesis of C. elegans mtDNA proceeds by rolling circle replication. We propose that multiple replication cycles on single template circles generate lariat structures with multimer tails composed primarily, although not entirely, of dsDNA, which are subsequently resolved to monomer circles. Such conversion could potentially involve the formation and resolution of the cruciform species observed by fragment 2DNAGE (Fig. 4).
In the context of a rolling circle, each initiation event will give rise to multiple genome units, making the identification of a specific start site challenging. Neither the biochemical nor microscopic methods used here revealed a specific site of replication initiation, nor did our bioinformatic analysis identify molecular signatures thereof [12]. The data presented here do not exclude the possibility of replication initiation via site-specific nicking followed by strand invasion, or alternatively by homologous recombination. In such a case, the intensity of the cruciforms from ClaI-ApaI and MfeI-MfeI mtDNA fragment gels would be consistent with the NCR region containing a site-specific origin of such a type. It is worth noting, however, that the C. elegans NCR region contains an approximately 1 kb region of short repetitive elements with a potential to form complex secondary structures necessitating replication fork pausing and restart, or perhaps facilitate intermolecular strand exchange [10]. Our 2DNAGE analysis of mitochondrial nucleic acid isolated from the glp-4(bn2) mutant strain revealed a marked diminution or absence of both canonical replication intermediates and cruciform species (S2 Fig.). These data further imply that the cruciforms detected in wildtype animals are potentially involved in synthesis and/or resolution of mtDNA in the gonad.
The data presented herein do not directly address molecular recombination involving sequence-specific mtDNA strand exchange events in C. elegans. The occurrence of genetic recombination within or between mtDNA molecules in animal mitochondria is highly controversial [30,31], and there is currently no evidence for sequence-specific inter- or intra-molecular recombination of vertebrate mtDNA. Indeed, recent work has demonstrated that recombination is undetectable in the germline of mice segregating neutral mtDNA haplotypes, when tracked over 50 generations [32]. Sequence alterations suggestive of recombination have been detected in heteroplasmic mice carrying a deleterious allele [33], although the changes were detected at very low frequency and could plausibly have resulted from in vivo template switching during replication. Purified mtDNA both from mice and cultured mouse embryonic fibroblasts has been analyzed extensively by 2DNAGE, without the detection of prominent cruciform species such as those we infer for C. elegans [5,34]. In contrast, junctional mtDNA species have been detected by both 2DNAGE and TEM methods in human heart and brain[35,36].
Elsewhere within the metazoa, compelling evidence has accrued for recombination between maternal and paternal mtDNA haplotypes in the mussel Mytilus in which inheritance of the mitochondrial genome is doubly uniparental [37]. Moreover, PCR-based methods have also implied novel sequence organizations consistent with intramolecular recombination among short tandem repeat arrays in the mitochondrial genome of the plant parasitic nematode Meloidogyne javanica [38].
The fact that no animal mtDNA resolvases have been identified to date remains a major challenge to the mtDNA recombination concept [39]. We searched the C. elegans genome for plausible mitochondrially targeted homologues of known integrase or resolvase gene families using bioinformatic methods, without success. This raises the possibility that the resolution of rolling-circle replication intermediates to genomic monomers in C. elegans involves known proteins involved in worm mtDNA maintenance [9,36,37], which may have adopted novel molecular functions. While some mitochondrial proteins required for the maintenance of mtDNA copy number have been described in C. elegans [40–44], the functional architecture of the minimal mtDNA replisome remains to be elucidated. Future work describing the effect of manipulation of known factors on mtDNA replication intermediates by 2DNAGE could potentially reveal or exclude roles for conserved mtDNA maintenance factors in rolling circle replication.
Several of the known metazoan mtDNA maintenance factors are highly homologous to bacteriophage proteins, including the mtDNA polymerase POLG, RNA polymerase POLRMT, and TWINKLE helicase [3]. Moreover, antiviral drugs are commonly mitotoxic [3]. These observations raise the possibility that the genome of the ancestral endosymbiont may have been replicated by a phage-like RCR mechanism [3], of which the DNA replication system used in the C. elegans germline is a relic. The analogy with T7 replication is furthered by (i) the presence of sporadic ssDNA regions observed along lariat molecules (Fig. 3H), which are consistent with yet-to-be-completed and ligated gaps between lagging-strand synthesis products and (ii) looping at the lariat circle-to-tail junction (Fig. 3I) that could represent a single replisome engaged in coordinate synthesis of the leading and lagging strands, consistent with the 2DNAGE data presented here (Fig. 2A). If RCR was the ancestral mode of animal mtDNA replication, strand-displacement, RITOLS and other types of theta replication may represent taxon-specific derived mechanisms. They may represent different solutions to the challenge of maintaining genomic fidelity in the oxidative environment of the mitochondrion.
Rolling-circle replication of mtDNA has been described in the plant and fungal kingdoms [45–49]. Here we present the first report of RCR in a metazoan, furthering the ubiquity of this mechanism of mtDNA synthesis. Our findings differ from the descriptions of RCR in plants and fungi, in that the C. elegans mtDNA monomer circle remains the most common topology of non-replicative mtDNA. Among the fungi, replication of Saccharomyces cerevisiae mtDNA produces linear molecules, with circles present only transiently during replication; in contrast, mtDNA synthesis in Candida albicans is recombination driven, generating concatameric linear networks [23,50]. Intriguingly, the implied similarities between C. elegans and yeasts with respect to mtDNA replication mechanisms and the high proportion of junctional mtDNA intermediates mirror similarities in the patterns and rates of mtDNA mutation observed in these species [51]. These data raise the possibility these two phenomena are linked. Plant mitochondrial genomes are particularly complex, often consisting of a mix of branched, linear and circular topologies that may be many genome multimers in size, rendering monomer circles a rare occurrence [46]. C. elegans mtDNA topology is also distinct in one aspect from models of bacteriophage RCR, due to the apparent absence of linear mtDNA monomers as detected by Southern blot, which in phages may bear distinct (phage T7) or permuted (phage T4) ends [21,52].
It has been previously assumed that C. elegans mtDNA adheres to the strand-displacement replication mechanism in which the NCRs contain first- and second-strand origins [7,8]; here we demonstrate that this cannot be the case. Neither putative replication origin produces bubble-type intermediates that are clearly observed in multiple mammalian systems including cultured human cells [53]. Rather, junctional mtDNA species were the only detectable and identifiable structures observed specific to either of the NCRs. While the details of concatemer resolution remain to be determined, we suggest that the junctional intermediates identified here may represent termination/resolution structures, in which strand-invasion or branch migration arrest could occur in a site-specific manner, facilitated by the short repeats present in the major NCR [10]. Invasion by the unreplicated ssDNA 3’ end of the lariat tail at the major NCR sequence would create a triple-stranded structure not unlike initial events in DNA strand exchange. Subsequent migration of this junction would provide an opportunity for formation of a second four-way junction. Cleavage and resolution would then generate a gapped circular monomer and lariat tail with 3’ overhang, one genome unit-length shorter. Such a mechanism could also produce rare uni-circular multimers (see S1 Table) in which resolution does not occur at an adjacent concatemer.
This mode of resolution, while speculative, is consistent with two intriguing aspects of our results. First, a subset of Y arc RIs are sensitive to degradation by RusA (Fig. 4A), indicating that some Y-like forms in fact contain four-way junctions, as predicted by a strand-invasion model. Second, our 2DNAGE analysis of the 5 kb mtDNA region containing the major NCR demonstrated the presence of RusA-resistant cruciforms consistent with hemicatenanes, a DNA species which forms via the convergence of double Holliday junctions, or alternatively by replication fork stalling [54–56]. We did not observe molecules simultaneously involved in elongation and resolution by TEM, which would be consistent with this model. However, we note that if present in vivo, such molecules could possibly exceed 75 kb in size and therefore are likely to be fragile to DNA isolation techniques. The enzymes involved in putative site-specific resolution remain unknown. Other mechanisms can be envisaged which would exploit homologous recombination machinery documented to exist in mitochondria in at least some species[57–59], for example, involving site-specific DNA binding proteins and/or branch-migration driven by directionally acting helicases.
Whether RCR occurs elsewhere in the animal lineage remains to be explored. Unlike previous studies characterizing mtDNA RIs in organisms where both strands of mtDNA contain protein coding information, all protein-coding genes are transcribed from one C. elegans mtDNA strand (Fig. 1A), thus raising the possibility that RCR may be linked to the transcriptional architecture of the mitochondrial genome. Fortunately, the Nematoda are an excellent model system in which to test such hypotheses. The mtDNAs of many nematode species have been sequenced, revealing considerable architectural variation and enabling comparative studies in which the mtDNAs differ by gene amplification or inversion, scrambled gene orders, or translocation of the major NCR in the genome map [60]. As such, the phylum presents a powerful new model system for probing the relationship between mitochondrial genome architecture and replication mode.
C. elegans itself offers a compelling model for the study of rolling circle replication in animal cells from both mechanistic and genetic perspectives. MtDNA replication primarily occurs in the C. elegans germline, where high demand during gametogenesis likely requires efficient, yet prolific mtDNA synthesis [1]. Since RCR is the only replication mode that can be detected, it must be sufficient to meet this demand despite the small percentage of molecules replicating at any particular time; we note that mtDNA multimers resulting from RCR can potentially resolve to several copies of the mitochondrial genome (Fig. 3), in contrast to theta replication, which produces only two daughter mtDNA molecules. We anticipate that the future characterization of factors involved in this process will provide many new insights into animal mtDNA maintenance, the evolution of replication mechanisms, and possibly even the pathological derangement of mtDNA synthesis in humans.
Caenorhabditis elegans strains N2 Bristol and mutant glp-4(bn2)I were obtained from the Caenorhabditis Genetics Center (CGC; Minneapolis, USA) and maintained as described [15,61]. For wildtype N2 animals grown in liquid culture, a culture sample was removed every 24 hours for monitoring and the turbidity of the culture tested to ensure ample E. coli OP50 were present. Culture samples were visually also checked for developmentally arrested or dauered larvae that could indicate bacterial depletion—none were observed. N2 and glp-4 (bn2)I animals grown on plates were fed on a lawn of E. coli OP50 as described [62]. Nematodes were actively growing and feeding on E. coli OP50 until immediately prior to mitochondrial isolation. For preparation of intact mitochondria, nematodes were collected, dounce-homogenized and subjected to differential centrifugation to generate a crude mitochondria-enriched fraction; further enrichment was achieved by sucrose-step gradient centrifugation (1:1.3 M sucrose), with all procedures conducted at 4°C [25]. Lysis and DNA extraction protocols for C. elegans were adapted from studies in which fragile mtDNA replication intermediates have previously been successfully isolated intact, for analysis by both 2DNAGE and TEM [18,63,64]. Briefly, freshly isolated mitochondria were lysed with 1% SDS in the presence of 200 mg Proteinase K and extracted by phenol-chloroform and ethanol precipitation; mtDNA was washed in ethanol and re-suspended in Tris-EDTA pH 7.6 for further manipulation.
Treatments (restriction endonucleases, RNases, other nucleases) were executed according to manufacturer instructions; RusA treatment was as described [28].
2DNAGE, blot-transfer and probe hybridization were carried out as previously described [25,65]; see S2 Table for genomic locations and sequences of mtDNA probes. For all 2DNAGE panels, four-hour exposures are presented. For detailed information on probes see Supplemental Experimental Procedures.
Aliquots of RNase I-treated mitochondrial nucleic acid were mounted directly on parlodion-coated copper grids following the Kleinschmidt method and imaged as described [24,66]. Molecule lengths were measured in Gatan DigitalMicrograph and calibrated by measurement of a co-spread 3.5 kb pglGAP plasmid. Each mtDNA molecule was measured 3 times to obtain mean values as reported in Fig. 3.
Cumulative GC skew analysis was carried out as described [23] using a custom R script.
|
10.1371/journal.pcbi.1003640 | A Normative Theory of Forgetting: Lessons from the Fruit Fly | Recent experiments revealed that the fruit fly Drosophila melanogaster has a dedicated mechanism for forgetting: blocking the G-protein Rac leads to slower and activating Rac to faster forgetting. This active form of forgetting lacks a satisfactory functional explanation. We investigated optimal decision making for an agent adapting to a stochastic environment where a stimulus may switch between being indicative of reward or punishment. Like Drosophila, an optimal agent shows forgetting with a rate that is linked to the time scale of changes in the environment. Moreover, to reduce the odds of missing future reward, an optimal agent may trade the risk of immediate pain for information gain and thus forget faster after aversive conditioning. A simple neuronal network reproduces these features. Our theory shows that forgetting in Drosophila appears as an optimal adaptive behavior in a changing environment. This is in line with the view that forgetting is adaptive rather than a consequence of limitations of the memory system.
| The dominant perception of forgetting in science and society is that it is a nuisance in achieving better memory performance. However, recent experiments in the fruit fly show that the forgetting rate is biochemically adapted to the environment, raising doubts that slower forgetting per se is a desirable feature. Here we show that, in fact, optimal behavior in a stochastically changing environment requires a forgetting rate that is adapted to the time constant of the changes. The fruit fly behavior is compatible with the classical optimality criterion of choosing actions that maximize future rewards. A consequence of future reward maximization is that negative experiences that lead to timid behavior should be quickly forgotten in order to not miss rewarding opportunities. In economics this is called “minimization of opportunity costs”, and the fruit fly seems to care about it: punishment is forgotten faster than reward. Forgetting as a trait of optimality can further explain the different memory performances for multiple training sessions with varying inter-session intervals, as observed in a wide range of species from flies to humans. These aspects suggest to view forgetting as a dimension of adaptive behavior that is tuned to the environment to maximize subjective benefits.
| Drosophila melanogaster forgets [1], [2]. In itself this is unremarkable because forgetting as a behavioral phenomenon appears in any adaptive system of limited capacity; storing new associations will lead to interference with existing memories. Forgetting, in this sense, is just the flip side of learning. When capacity is not an issue, forgetting may nevertheless be caused by a useful mechanism: one that keeps a low memory load and thus prevents a slowdown of retrieval [3], [4]. Consequently, capacity or retrieval limitations lie at the heart of standard theories of non-pathological forgetting [5], [6], which focus on interference and decay explanations. Alternatively, forgetting has been proposed to be an adaptive strategy that has evolved in response to the demands of a changing environment [7]. It is the latter explanation that seems to apply to Drosophila where the experimental evidence suggests that the cause underlying forgetting is an active process which is modulated by the learning task and not by internal constraints of the memory system; in particular in olfactory conditioning tasks, reversal learning leads to faster forgetting [8] whereas spaced training leads to slower forgetting compared to single or massed training [9]. Further, forgetting in Drosophila seems rather idiosyncratic in that aversive conditioning is forgotten approximately twice as quickly as appetitive conditioning [10], [11].
In psychology, the term forgetting commonly refers in “to the absence of expression of previously properly acquired memory in situations that normally cause such expression.” ([6]; see also [12]). Similarly, in conditioning experiments, one speaks of forgetting, when the conditioned stimulus fails to evoke the conditioned response at some point after successful conditioning [8], [13].
In the basic protocol for behavioral studies of memory in Drosophila [1] a group of flies is placed into a tube for conditioning. There the flies are exposed to a specific odor and the exposure is paired with a reinforcer (sugar or electrical shock). Having experienced the pairing once or multiple times, the flies are removed from the conditioning tube. After a predefined delay time, the group is placed into the middle of a second, elongated tube for assessment. One side of the elongated tube is baited with the conditioned odor and, after a while, the fraction of flies is determined which exhibit the conditioned response by comparing the number of flies which are closer to the baited side of the tube with the number of flies closer to the un-baited side. The setup allows to measure memory performance (c.f. Fig. 1 D), i.e. expression of the conditioned response, as function of the delay time and of the conditioning protocol (e.g. magnitude of reinforcement, number of pairings). To check for bias in the setup, one typically in addition uses a second odor as a control which was not paired with a reinforcer.
That Drosophila has a dedicated mechanism to control forgetting was convincingly demonstrated by Shuai et al. [8] and Berry et al. [2]. Inhibition of the small G-protein Rac leads to slower decay of memory, extending it from a few hours to more than one day [8]. Conversely, elevated Rac activity leads to faster forgetting [8]. Similar results were achieved by modulation of a small subset of Dopamine neurons [2]. Stimulating these neurons leads to faster forgetting after aversive and appetitive conditioning, while silencing these neurons leads to slower forgetting [2].
Given the importance of decision making, it appears unlikely that forgetting in Drosophila is a behavioral trait which is maladaptive in an ecological sense. Hence we investigated what generic model of the environment would justify the observed forgetting and in particular the asymmetry between aversive and appetitive conditioning. For this we mathematically determined optimal decision making strategies in environments with different associations between stimulus and reinforcement.
For our model we assumed a simplified scenario where the conditioning pertains directly to the appetitive reaction. In particular, depending on the state of the environment, approaching the odor can lead to reward () or punishment () but it can also result in no reinforcement () (Fig. 1). Fleeing the odor, i.e the aversive reaction, never leads to reinforcement (). An agent (fruit fly), whose goal is to maximize reinforcement, chooses between the appetitive and aversive reaction depending on past experience. To model the non-deterministic behavior observed in the experiments we assume that the two available behavioral options involve different costs of responding. These costs of responding, however, fluctuate from trial to trial causing no bias on average. For instance, a fly which happens to find itself to the right of the group initially could well have a smaller cost of responding for staying on this side of the assessment tube on this trial. More generally, the stochastic costs of responding can be seen as incorporating all other factors that also influence the behavior but do not depend on the past experiences that involve the conditioned stimulus. The total reward received by the agent is the external reinforcement () minus the cost of responding. Our agent takes this into account in decision making, and so the costs of responding result in trial to trial fluctuation in the behavior. Whether the appetitive reaction results in depends on the state of the environment. This state changes slowly over time (according to a Markov chain, see Methods and Fig. 1A). So when the appetitive reaction results in on one trial, the same outcome is likely on an immediately subsequent trial, but as time goes by the odds increase that the appetitive reaction results in or even punishment.
If the agent knew the environmental state, the best policy would be simple: choose the appetitive (aversive) reaction if the environmental state is rewarding (punishing). Typically however, the agent does not know the actual environmental state but, at best, maintains a belief about it (see Fig. 2A and Methods). In our model, the belief consists of the probabilities , and to receive rewarding, neutral or punishing reinforcement, respectively, after selecting the appetitive reaction. Geometrically, the belief can be represented as a position in a 2-dimensional belief space that stepwise changes after the appetitive reaction and thus gaining new information about the current environmental state and otherwise drifts towards an equilibrium (forgetting), see Fig. 2B (note that, since the three probabilities sum to one, the probability of the neutral state can be computed given the probabilities of the rewarding and punishing state, i.e. ).
If e.g. a fly gets punished, the probability to be punished again on the next trial is high (initial point of red trajectory in Fig. 2B). If subsequently the fly chooses the aversive reaction, the belief will drift towards a stationary value (end point of red trajectory in Fig. 2B). We assume that the agent has implicit knowledge, e.g. gathered by experience or through genetic encoding, about the transition rates of the environmental state.
Based on belief values and costs of responding one may define different policies. A greedy policy selects the appetitive reaction if the agent believes that reward is more probable than punishment and costs of responding are equal for both actions, i.e. (Fig. 2C top, middle). If costs for one reaction are larger than for the other, the region in the belief space favoring this higher-cost reaction becomes smaller (Fig. 2C top, left and right). Immediately after conditioning, an agent has a strong belief that the environment is still in the same state as during conditioning. Thus, if the greedy policy determines action selection, an agent most likely chooses the conditioned response. As the belief drifts towards the stationary point, the stochastic costs of responding gain more influence on the decision making and thus an agent is more likely to have already forgotten the conditioning, i.e. the agent is more likely to choose the opposite of the conditioned response. We call this policy “greedy”, because it maximizes reward if only one choice is made but it is not necessarily optimal with respect to gaining future rewards. Technically, the greedy policy is equivalent to the optimal future discounted policy with discount factor , i.e. the policy that neglects future rewards.
In order to conveniently analyze the forgetting behavior under the greedy policy for different choices of the environmental parameters and (Fig. 1A), we use a re-parametrization with the “probability of the neutral state” and the “average reward” , where denotes the stationary state probability of state (see Methods for the relationship between and ). Changing the probability of the neutral state has almost no effect on the forgetting curve (Fig. 3A, solid vs. dashed line). Increasing the average reward has the consequence that in the stationary state more agents select the appetitive reaction than the aversive reaction (Fig. 3A, solid vs. dotted line). The speed of forgetting a conditioned state (p, n or r) is determined by the rate of transitioning away from this state. Fig. 3A (solid vs. dash-dotted line) shows the effect of changing the rate , whose inverse is equal to the average number of timesteps the environment spends in the rewarding state: forgetting is faster for a larger rate . The variance of the costs of responding determines the impact of the costs of responding on decision making. For large variance the forgetting curve is closer to 0.5 than for small variance, since for large variance it is more likely that the costs of responding have a strong impact on decision making (Fig. 3B).
While the difference in forgetting speed after appetitive and aversive forgetting could be a consequence of different transition rates and , such a difference also arises if these rates are equal but the agent uses a provident policy, i.e. a policy that also takes into account future rewards. In the long run the provident policy is superior to the greedy policy (Fig. 4B). We therefore determined numerically a policy which approximately maximizes the reward rate, i.e. the total reward accumulated over a long period divided by the length of this period (see Methods). The resulting policy is such that there are beliefs for which the appetitive reaction is chosen, even when the probability of punishment is larger than the probability of reward, i.e. , and the costs of responding are equal for both actions (Fig. 2C bottom, middle). The reason for this becomes clearer when we look at what economists call the opportunity cost, i.e. the additional gain that has not been harvested because of missing to choose the (often by hindsight) better option [14]. For the appetitive reaction, the agent's opportunity cost is given by the potentially lower cost for the aversive reaction. But for the aversive reaction, the agent's opportunity cost is not only the potentially lower cost for the appetitive reaction but also the lack of further information about the actual environmental state. This information is required for best exploitation in future trials. Assume, for instance, that at some point in time the agent believes that punishment is slightly more probable than reward and therefore sticks to the aversive reaction. Now, if the actual environmental state would be rewarding, the agent would not only miss the current reward but also misses subsequent rewards that could potentially be harvested while the state is still rewarding. When taking this opportunity cost into account, the agent will choose the appetitive reaction despite the belief state slightly favoring the aversive reaction. For an external observer this optimal choice behavior appears as a faster forgetting of the aversive memory. In short, the asymmetry in forgetting after aversive and appetitive conditioning (Fig. 4) arises because choosing the appetitive reaction is always informative about the current environmental state whereas choosing the aversive reaction is not.
The probabilistic calculations needed to derive the optimal provident behavior can be quite involved. We do not suggest that there is a neuronal circuitry in Drosophila which actually does these calculations. Yet it is interesting to note that a much simpler mechanistic decision making model already results in close to optimal behavior (Fig. 4B). This simple model allows an interpretation of the variables as synaptic strengths from odor sensitive neurons to decision neurons (Fig. 4C). In the absence of odor and behavioral feedback the synaptic strengths decay with different time scales towards a stationary level: decay is faster for synapses targeting the “avoid” neurons than for the “approach” neurons. One could speculate that the speed of this decay is governed by e.g. the concentration of Rac [8] or dopamine [2].
So far we have assumed that the transition rates between the environmental states are fixed. This is not an assumption Drosophila seems to make and in fact, would be an unrealistic model of the environment. The experiments by Tully et al. [9] show that forgetting depends not only on the number of conditioning trials but also on their frequency. In particular, forgetting is slower when the same number of learning trials is spaced out over a longer period of time. Spaced training is more informative about the environment being in a slowly changing mode than the temporally compressed massed training. Furthermore, reversal training during which in fast succession an odor is aversively, neutral and again aversively conditioned [8] results in faster forgetting and is informative about a fast changing environment. So the observed behavior provides rather direct evidence that adaptation in Drosophila does indeed take non-stationarity into account.
To include adaptation as a response to changing transition rates, we extended our model by a slowly varying meta variable which can either be in state “fast change” or “slow change” (Fig. 5A). The dynamics of the meta variable is governed by a Markov process with small transition rates. In state “fast change”, the environmental reward state changes more rapidly than in state “slow change”. In this setting, an optimal agent maintains a belief about both the environmental reward state and the “hidden” state that sets the time scale of the changes in . Spaced training increases the belief that the environment is in a slowly changing mode, whereas reversal learning leads to a strong belief about the environment being in the fast changing mode. The resulting greedy-optimal behavior is in qualitative agreement with the known behavior after spaced, massed and reversal learning (Fig. 5B) as observed for flies [8], [9], honey bees [15], pigeons [13], and humans [16].
We demonstrated that forgetting appears when an agent, subject to costs of responding, acts optimally in an environment with non-stationary stimulus-reinforcement associations. Based on reward maximization in a non-stationary environment, which is a reasonable objective not only for the fruit fly but for other species as well, our normative theory of forgetting includes an asymmetry in forgetting speed after aversive and appetitive conditioning and an adaptation of forgetting speed after spaced, massed and reversal learning. The asymmetry is the result of an economically optimal provident policy, which forages not only for immediate reward but also for information required for future exploitation. The adaptation of forgetting rate after spaced, massed and reversal learning is a consequence of the agents estimation of the current rate of environmental changes.
That costs of responding influence the action selection is an assumption which is in agreement with test-retest experiments [9], [11], [17]. In these classical conditioning experiment the flies are grouped according to whether they choose the conditioned response or not. Both groups are immediately retested to examine whether the flies stick to their decision. The outcome is: they do not. An equal fraction of flies chooses the conditioned response in both retest groups and this fraction is the same as in the first test containing all flies. This suggests that all flies maintain traces of the conditioning but that also other factors influence the choice in a stochastic way. Similarly, in our model the belief is a sufficient statistic of the past experiences that involve the conditioned stimulus and the stochastic costs of responding account for other factors that influence the choice.
A key assumption in our normative explanation of the differential forgetting in Drosophila is that the relationship between conditioned stimulus and reinforcement is non-stationary. Now, if this relationship were completely stationary, it would not need to be learned by the phenotype because it would already have been learned by the genotype, i.e. in this case the stimulus would be an unconditioned stimulus. Hence, from an evolutionary perspective, our assumption is close to being a truism. Nevertheless, many biological models of reinforcement learning have, for the sake of simplicity, assumed a stationary stimulus-reinforcement relationship [18], [19].
Experiments and models with non-stationary stimulus-reinforcement associations have suggested, similar to our findings, that in a more volatile environment the learning should be faster [20]–[24]. However, faster learning does not unconditionally imply faster forgetting. The asymmetry in forgetting speed after appetitive and aversive conditioning additionally requires an evaluation of the behavioral relevance of a specific memory content. Since the aversive reaction is not informative about the current state of association, aversive conditioning should be forgotten faster than appetitive conditioning.
Finding the optimal policy in an environment with a non-stationary stimulus-reinforcement relationship, as considered here, is computationally involving. As we have shown, however, approximately optimal decision making is still possible with a simplified neuronal model using experience induced synaptic updates. This model incorporates forgetting in the decay time constant of the synaptic strengths. As the parameters describing the changing environment are assumed to be constant across generations, the neuronal architecture and the forgetting rates can be considered to be genetically encoded.
Since the work of Ebbinghaus [25] on the forgetting rate of non-sense syllables and the observation of Jenkins and Dallenbach [26] that sleep between learning and recalling reduces forgetting, cognitive psychologists debate about the role of natural decay and interference in explaining forgetting [5]. While interference based explanations are favored by many [5], [12], Hardt et al. [6] recently advocated active processes behind decay-driven forgetting. They suggested a memory system that engages in promiscuous encoding and uses a flexible mechanism to remove irrelevant information later, mostly during sleep phases. In their view, different forgetting rates are a sign of such a flexible removal mechanism. But why do biological organisms need to actively remove irrelevant memories at all? Popular answers so far implicitly assumed that forgetting is ultimately the result of some limitation of the memory system, for instance, limited storage capacity, a limit on the acceptable read-out time for the memory or a decay of the biological substrate similar to unused muscles atrophy [6], [27]. In our model, however, forgetting does not result from a memory limitation, but emerges as an adaptive feature of the memory system to optimally cope with a changing environment while accounting for the relevance of different memory contents.
In time bin an odor can be associated with one of three environmental states: (reward), (neutral), (punishment). The time-discrete dynamics of the environmental state is given by a Markov Chain with state space and transition probabilities and , where for . For simplicity we did not include direct transitions between the rewarding and punishing state, i.e. . Including them would also allow for a behavior where the preference switches from the conditioned response to the opposite of the conditioned response before reaching the stationary state. The stationary distribution of this Markov chain, satisfying the self-consistency equation , is given by and , where .
In each time bin the agent has two behavioral options: approach the odor () or avoid the odor (). If the agent avoids, a neutral reinforcement signal is always returned. If the agent approaches, the external reinforcement signal depends on the environmental state: there will always be a positive signal if , always a negative signal if and if the odor is associated with the neutral state (), the agent will stochastically get a neutral signal with probability 0.99, while with probability 0.005 the agent will get a positive or a negative reinforcement signal. Positive and negative reinforcement signals during the neutral state are included to model situations, where reward or punishment depends on odor unrelated factors. For further use we summarize the information in this paragraph in the probabilities , with non-zero entries , , and , , .
The agent maintains a belief over the current environmental state given past reinforcement and actions . The belief state is updated by Bayesian filtering(1)with normalization . We use the abbreviation to denote the update of the belief given action and reinforcement signal .
We modeled costs of responding with exponentially distributed and uncorrelated random variables and with parameter , i.e. the probability density function of is given by if and otherwise. This distribution has mean and standard deviation . We assumed, that the agent receives an effective reward, which is a sum of the external reinforcement signal and the momentary cost of responding for the action chosen. During decision making, the agent knows the costs of responding for both actions but only has an expectation of the external reinforcement signal.
If the goal is to maximize immediate reward, the agent's policy depends on the expected return in the next step , which for action ap can be simplified to and for action av is always zero, i.e. . Including costs of responding, the policy that maximizes immediate reward selects the action for which is maximal.
A canonical choice of the objective to be maximized by a provident policy is the reward rate, i.e.with expected reward in time bin when acting according to policy . We approximately determined the policy which maximizes the reward rate by two methods: dynamic programming and linear programming on a quantized space.
Dynamic Programming allows to find a policy that maximizes the future discounted valueswith discount factor and expected reward in time bin after starting in belief state and acting according to policy . For finite state spaces and sufficiently close to 1 a policy that maximizes future discounted reward also maximizes the reward rate [28]. Without costs of responding one could directly apply the Incremental Pruning algorithm [29] to find a policy that maximizes the future discounted values. Here we derive dynamical programming in the presence of costs of responding.
Dynamic programming proceeds by iteratively constructing optimal finite-horizon values for some operator . Assume that we have the horizon- policy that maximizes the future discounted values of an episode of length . The horizon- policy consists of instructions for each step in the episode , where tells which action to take at the -th step before the end of the episode, given belief and costs of responding and . To construct the horizon- policy we need to extend the horizon- policy by the instruction for the first step, i.e. the -th step before the end of the episode. Without considering the costs of responding in the first step, the expected future discounted values for choosing action are given by(2)where and the value function is given by (we will use the indicator function , given by if is true and otherwise):
With the change of variables , the resulting probability density function (Laplace probability density), and the abbreviations , and , we get(3)
In the same manner we find the value function , which depends through on (see Eq. 2)(4)where now .
Due to the discount factor this recursion will eventually converge. In practice we will stop after iterations and define the policy , which approximates the future discounted policy. Note that in contrast to the finite horizon policies the policy is stationary: in a sequential setting there is no end of an episode on which the policy may depend.
The number of terms in a naive implementation of grows exponentially with . Without costs of responding the exponential growth can sometimes be prohibited by Incremental Pruning [29]. With costs of responding we are not aware of a way to prevent exponential growth. In Fig. 4 we approximated the stationary policy by taking the policy after 5 iteration with discount factor , i.e. . Since it is not clear whether for this choice of discount factor and number of iterations the resulting policy is a good approximation of the reward rate maximizing policy, we compared the result of dynamic programming with the policy obtained by linear programming on a quantized belief space.
For finite state and action space Markov Decision Processes linear programming can be used to find a policy that maximizes the reward rate [30], [31]. In our case, the policies act on the continuous space of belief states and cost of responding differences . Analogous to the finite state space problem, the optimization problem could be formulated as: find functions that implicitly define the policy [30] and satisfywhere denotes the expected reward for action , belief state and costs of responding differences and denotes the probability density to transition from and to and given action . A straightforward approach is to quantize the belief space and space of cost of responding differences, replace the integrals by sums and find through linear programming an approximation to the reward rate maximizing policy. We quantized the two dimensional belief simplex on a square lattice with different lattice spacings. Values that did not fall on lattice points where stochastically assigned to neighboring lattice points. The space of real valued cost of responding differences was quantized by segmenting the real line into adjacent intervals with equal mass of the probability density function. For each interval the average costs of responding for each action where computed. Using increasingly finer quantization we estimated the total reward to be between 600 and 655 for trials of time bins, which is in agreement with the estimate obtained with dynamic programming (Fig. 4B provident).
In Fig. 4 we demonstrate that also an agent with two uncoupled low-pass filters can show near to optimal behavior. The agent's decision to approach () or avoid () the odor depends on whether , where () are variables interpretable as synaptic strengths and where represents stochastic input due to costs of responding. The values of decay with different time-constants, in the case of no feedback, because the agent either stays away or no odor is present. If the agent approaches the odor and experiences reward, is set to a maximal value, while is set to zero; for odor plus punishment is set to a maximal value, while is set to zero. Formally, with the subscript standing for either ap or av, we get(5)The parameter controls the speed of forgetting, sets a baseline value and sets a maximum value. In Fig. 4 the parameter values where fit to the curves in sub-figure A (approx provident: , , , , ) and to the curves in sub-figure B (approx greedy: , , ).
To study the behavior of an agent that additionally has to estimate the rate of change we extended the basic model of the environment with a meta variable that controls the rate of change of the environmental state. In time bin the meta variable can be in one of two states: (fast) or (slow). The dynamics of the meta variable is described by a Markov Chain with transition probabilities and . If the meta variable is in the slow (fast) state the transition parameters of the environmental state are , and , . In the extended model the state space is given by the product space and the transition parameters are given by . The agent maintains a belief about both the environmental state and the state of transition speed.
In spaced training, the agent was aversively conditioned six times with intermittent waiting periods of 9 time bins. In massed training, the agent was aversively conditioned in 6 subsequent time bins. In reversal learning, the agent was exposed to the punishing, neutral and punishing environmental state in subsequent time bins. Forgetting curves are shown for the computationally less involving greedy policy. In order to compare massed with spaced training we choose a finer time discretization in the extended model, i.e. 10 time bins in the extended model correspond to 1 time bin in the basic model. In figure 5B the result is plotted in units of the basic model.
|
10.1371/journal.pcbi.1004796 | BIITE: A Tool to Determine HLA Class II Epitopes from T Cell ELISpot Data | Activation of CD4+ T cells requires the recognition of peptides that are presented by HLA class II molecules and can be assessed experimentally using the ELISpot assay. However, even given an individual’s HLA class II genotype, identifying which class II molecule is responsible for a positive ELISpot response to a given peptide is not trivial. The two main difficulties are the number of HLA class II molecules that can potentially be formed in a single individual (3–14) and the lack of clear peptide binding motifs for class II molecules. Here, we present a Bayesian framework to interpret ELISpot data (BIITE: Bayesian Immunogenicity Inference Tool for ELISpot); specifically BIITE identifies which HLA-II:peptide combination(s) are immunogenic based on cohort ELISpot data. We apply BIITE to two ELISpot datasets and explore the expected performance using simulations. We show this method can reach high accuracies, depending on the cohort size and the success rate of the ELISpot assay within the cohort.
| When studying the host immune response, a central question is: “which peptides elicit CD4+ T cell responses?” ELISpot assays are used to assess if subjects have responded to a given peptide. However, to determine which of the HLA-II molecules coded by the host HLA genotype is responsible for the reaction requires additional analysis. We present a Bayesian approach to solve this problem and have implemented it for use with the statistical language R under the BIITE moniker. Importantly, the aim of BIITE is to interpret experimental data, not to make in silico predictions. The method considers the immunogenicity of all HLA (in a cohort of patients) with respect to a given peptide simultaneously, in order to deal with linkage disequilibrium between genes of the HLA locus. Furthermore, users can enter additional information they might have (from literature or other experiments) in the form of prior information. The method is not exclusive to the HLA genes and can be used to attribute positive binary outcomes to any multi-allelic set of genes.
| Adaptive immunity relies on the recognition of non-self peptides by T cell receptors to mount an effective response against an infection. Both self and non-self peptides are presented on the cell surface by human leucocyte antigen complex (HLA) molecules using two pathways: the HLA class I pathway presents cytosolic peptides on HLA class I molecules to CD8+ T cells, while the HLA class II pathway presents peptides of proteins that have been internalized by antigen presenting cells (APCs) to CD4+ T cells.
Identifying which peptide:HLA (pHLA) complexes give rise to an immune response in any infectious, autoimmune, allergic or oncogenic disease is paramount to understanding host defence and pathogenesis and informing epitope vaccine design. Furthermore, the ability to define the specific pHLA complexes dominating a given response is central to both the theoretical understanding of HLA-disease associations and to the practicalities of designing experiments that use pHLA tetramers to enumerate and characterise specific T cells. Only a small proportion of HLA-binding peptides are able to elicit a T cell response in the context of a given HLA class I [1] or class II molecule; identifying which pHLA complexes are immunogenic is not trivial.
The most widely used experimental method for investigating immunogenicity in humans is the ELISpot assay [2]. Interpreting ELISpot responses of CD8+ T cells is relatively straightforward. Given an individual’s HLA class I genotype there are a limited number of HLA-I molecules that could be present (a maximum of 6); this together with the highly specific nature of the HLA class I peptide binding register means that identifying which HLA molecule is responsible for a positive response in a given individual to a certain peptide is usually unambiguous.
Here we focus on the more demanding problem of defining immunogenic pHLA-II complexes. That is, having established that an individual of known HLA class II genotype gives a positive CD4+ T cell ELISpot response to a given peptide, how could one infer which of the class II molecules was responsible for eliciting this response? Multiple factors complicate this analysis. Firstly, HLA-II molecules are heterodimers whose alpha and beta chains are encoded at separate genetic loci. Given their high polymorphism (the exception being the locus encoding the DR alpha chain) and the fact that their products might be paired either in cis or in trans, the number of expressed DP and DQ molecules in heterozygotes is theoretically doubled [3–6]. Furthermore, the DR alpha chain can be paired with the DQ beta chains [7]. Secondly, the DR beta chain can be sourced from between 2 to 4 loci depending on the subject. All humans possess two copies of the DRB1 locus. These can be complemented by a maximum of two of DRB3, DRB4 or DRB5 (one per chromosome). Consequently a maximally heterozygous individual may have 14 distinct HLA class II molecules. Thirdly, expression levels seem to differ [8] between different chains, leading to differential presentation of HLA-II molecules on the cell surface. Fourthly, (as for the class I genes), the genes of the HLA-II locus are in strong linkage disequilibrium, complicating the attribution of T cell responses to specific HLA-II loci. Lastly, the class II peptide binding grove is open at both ends and so it can accommodate peptides of variable length. This means that several amino acids in a given peptide could be anchor residues, complicating the in silico scanning of peptides for binding motifs. Together these factors mean that identifying which of an individual’s 3–14 possible HLA class II molecules is responsible for eliciting a positive CD4+ T cell response is problematic.
Historically, this problem has been addressed by cloning T cells and dissecting responses functionally, for example with HLA transfectant APC panels. However, this is intractable for high-throughput epitope mapping studies. While methods exist for predicting binding of peptides to HLA class II molecules, for example NETMHCIIpan [9], our aim is different in two important respects. Firstly, we are seeking a method to interpret experimental data rather than to make in silico predictions; secondly, we aim to infer immunogenicity rather than peptide binding. Paul et al. have recently described the RATE method [10] which addresses the same question. Their method calculates the relative frequency (RF) of positive CD4+ T cell ELISpot outcomes from multiple individuals in the HLA+ and HLA- groups in order to discover immunogenic pHLA combinations. In contrast, we propose a Bayesian framework to determine the immunogenicity of peptide:HLA-II complexes for a given peptide, which allows us to consider all HLAs simultaneously. We have implemented this in the R package BIITE (Bayesian Immunogenicity Inference Tool for ELISpot).
We will use the abbreviation HLA to denote HLA-II, but the same approach could be used to determine HLA class I peptides from CD8+ T cell ELISpot data. Assume we have ELISpot data D for a single peptide in a cohort of N individuals, in which a total of m HLA molecules are present. We wish to obtain the peptide:HLA immunogenicity, E, for each of the m HLAs as a number between 0 and 1; this is approximately the probability that a pHLA combination results in a positive ELISpot in a randomly chosen individual (with the relevant HLA allele) and would be exact if each subject presented exactly one HLA. Hence, the hypothesis space we will explore is [0, 1]m. Bayes’ theorem (see S1 Text for more information) states that the posterior likelihood P(H|D) for a hypothesis H = (E1,E2,…,Em) ∈ [0,1]m is proportional to the product of the prior P(H) and the likelihood P(D|H):
P(H|D)∝P(D|H)P(H).
We define the likelihood P(D|H) multiplicatively:
P(D|H)=∏i=1NP(Di|H),
where Di denotes the data for one individual and P(Di|H) is defined as
P(Di|H)= {∏j=1m(1−Ej)nijif subject i has a negative ELISpot,1−∏j=1m(1−Ej)nijif subject i has a positive ELISpot.
(1)
Here, nij is the copy number of HLA allele j in subject i. The logic behind this formula is explained in more detail in S1 Text.
The algorithm allows for the inclusion of pre-existing knowledge or beliefs. These could be results from previous experimental assays or from prediction algorithms. When no prior knowledge is given, the prior is the uniform distribution on [0,1]m. When prior knowledge is available, we assume it is on an HLA molecule-by-molecule basis:
P(H=(E1,E2,…,Em))= ∏j=1mP(Ej).
In our analysis, we used prior data from NetMHCIIPan [9]. NetMHCIIPan was used to predict the binding of peptide:HLA combinations with a cut-off of IC = 500nM. The combinations below this threshold were assigned a Beta prior with mode 0.35 and SD 0.2; combinations above this threshold were assigned a Beta prior with mode 0.001 and SD 0.15 (see S1 Fig).
In order to find the posterior density P(H|D), we implemented the Metropolis-Hastings algorithm [11], which is explained in more detail in S1 Text, to construct Markov chains of length 100,000, as these were found to return the same marginal distributions as chains of length 10^6. The model was implemented in R version 3.1 [12]. The graphical output was generated with the R package ggplot2 [13]. The code is available on http://github.com/liesb/BIITE as an R package. It can be installed using the R library devtools. In R, use the command install.packages("devtools") to install devtools. Then, to install BIITE, load devtools using the command library(devtools); next, use devtools::install_github("liesb/BIITE") to retrieve BIITE from its github repository. To load the package, use library(BIITE).
An example input file can be found in S4 Table. Each row in the file represents a subject. There should be a column for each HLA in the population, which contains the subjects’ copy number of that gene. For each peptide that is tested, there is a column, containing a logical value: TRUE if the subject delivered a positive ELISpot for the peptide, otherwise FALSE. The order of the columns does not matter, although preferably the ELISpots columns and the HLA columns form one block each. The HLA data can be entered in two-digit or four-digit format or a mixture of the two (see the example in S1 Table): most HLAs are defined on the two-digit level, but DRB1*15 has been split into DRB1*15:01 and DRB1*15:02). Since ‘*’ is a special character in the R language, it should not be used in the column names and can be replaced by an underscore.
In order to estimate the immunogenicity of each peptide:HLA combination, we consider the marginal distributions P(…,Ej,…|D) for each of the HLA molecules in the population. If the data contain little information about the immunogenicity of a pHLA combination (e.g. the HLA molecule is present only at very low frequencies in the population), then this marginal posterior distribution will not differ very much from the prior it was assigned.
If a pHLA combination has a relatively flat posterior distribution then its immunogenicity cannot be reliably determined; this will occur when limited information about the pHLA is contained in the data. We eliminate these unreliable combinations by only considering pHLA combinations which differ from the uniform distribution. We quantify the difference between the posterior and the uniform using the Kullback-Leibler divergence DKL(P(Ej|D)||Unif(0,1)). If this difference is smaller than DKL(Beta(2,1)||Unif(0,1)) we say there is insufficient information to determine the immunogenicity. We have chosen the threshold to be determined by the Beta(2,1) distribution, which naturally arises as the posterior distribution of the parameter of the binomial distribution after one (positive or negative) observation, when assuming a uniform prior. In other words, if we want to know what the probability p is of a coin landing heads in a toss, and we are only allowed one experiment with no prior information, Beta(2,1) (or Beta(2,1)) is the best description of p. Hence, requiring that the Kullback-Leibler divergence passes the aforementioned threshold, is asking that the posterior contains more information that we would get from a single coin toss. This threshold is arbitrary and can be adjusted by the user.
To obtain a single estimate for Ej, we propose using the mode of the marginal posterior distribution, but other summary statistics could be considered. In our analyses, using either the mean or median of the marginal posterior distribution yielded very similar results. These three summary statistics are all provided in the output of the algorithm (see S8 Table for an example where we have retained only the mode). Our interest lies in determining, for a given peptide, which HLA molecules were responsible for the observed positive ELISpot assays in the dataset, i.e., which HLA has the highest explanatory power. This can be done as follows. Based on the output file, we can rank the HLAs from highest posterior mode to lowest posterior mode; the HLAs ranked highest are the ones identified by BIITE to be the most immunogenic. Next, we turn to the subject data and determine which of their HLAs has the highest rank. For a subject with a positive ELISpot, this one ‘highest ranked HLA’ is considered to be the HLA responsible for the positive ELISpot result. To check whether a single HLA has high explanatory power, we determine how many subjects with a positive ELISpot carry that HLA, and in how many of these subjects the HLA is the highest ranked HLA.
To illustrate the application of the method we have created a worked example. The input file is provided in S4 Table, the corresponding output file (for pep_1) in S8 Table. The HLAs are ranked from high to low (‘Posterior mode’ column). For each HLA, we count how many subjects carry the HLA and have produced a positive ELISpot (‘# Carriers with positive ELISpot’). For example, of the 73 carriers of DQB*03, 67 had a positive ELISpot. Since DQB*03 is also the highest ranked HLA overall, all of these 67 positive ELISpots are ‘explained’ by DQB*03. For the second-highest ranked HLA, DRB1*14, 25 of 28 subjects had a positive ELISpot. Fourteen of those 25 are explained by DRB1*14 (the other 11 DRB1*14 carriers with a positive ELISpot also carry the higher ranked DQB*03, which ‘explains’ their positive ELISpot). In contrast, the lowest-ranked HLA, DQB*02, has 54 carriers that produced a positive ELISpot, but none of these positive results are explained by DQB*02.
The bacterium Burkholderia Pseudomallei is the causative agent of melioidosis. PBMCs from a small cohort (N = 38) of exposed sero-positive blood donors were used to test 17 overlapping 20-mers of the alkyl hydroperoxide reductase protein (AhpC, BPSL2096) by IFNγ ELISpot assay [14]. Results that were 2 standard deviations (SD) above the mean of medium only control were considered positive.
Each subject was HLA genotyped at the HLA-DRB1 and HLA-DQB1 loci. The observed alleles and their frequencies in the cohort are listed in S1 Table.
We applied the algorithm to each of the 17 peptides separately, both with and without the inclusion of prior data.
We used two datasets for validation. In the first dataset, immunogenicity of pHLA complexes for 6 HLAs (DRB1*01, DRB1*04, DRB1*15:01, DRB1*15:02, DQB1*06, DQB1*08) was determined using transgenic mouse models that express a single (human) HLA-II molecule (and no mouse MHC class II) and were challenged with the 17 different peptides in turn; IFNγ ELISpot was then used to determine whether peptides were immunogenic (number of SFCs larger than mean of SFCs numbers in medium-only + 2SD) [14]. The second dataset consisted of relative peptide binding affinities for all 17 overlapping 20-mers of the protein with respect to 10 HLA-II molecules; this was evaluated with competitive ELISA using an automated workstation. A peptide was considered to be binding to the HLA molecule when the relative binding affinity was less than 100 [14].
The immunogenicity of each pHLA combination was calculated as the mode of the marginal posterior distribution. We constructed ROC curves to compare these values to the pHLA binding data and transgenic immunogenicity data.
The bacterium Pseudomonas Aeruginosa (PA) is an environmental organism that can cause infection in damaged lung or the immune compromised host. PBMCs from a small cohort (N = 58) of patients with a chronic lung disease associated with recurrent lung infection and progressive lung damage, called bronchiectasis, were used to test 34 overlapping 20-mers of the OprF protein using IFNγ ELISpot assays [15]. We used a cut-off of 2SD above the mean of medium only control to decide whether a particular ELISpot result for an individual was positive or negative.
Analogously to the Burkholderia dataset, we had access to a immunogenicity data derived from transgenic mouse models and binding data from peptide-binding assays [15]. Transgenic mouse models were available for DRB1*01, DRB1*04 and DRB1*15; binding data was available for 7 HLA-DRB1 alleles.
Each subject was HLA-typed at the HLA-DRB1 and HLA-DQB1 loci. The observed alleles and their frequencies in the cohort are listed in S2 Table. Each subject is classified into one of two groups, based on the results of sputum microscopy and culture by standard microbiological techniques. Individuals in group 1 were never sputum culture positive for PA, while individuals in group 2 were sometimes or frequently sputum culture positive for PA over a period of 6 months.
We first applied the algorithm to each of the 34 peptides separately, both with and without the inclusion of prior data, including all subjects. Next, we applied the algorithm to each of the two groups (as described above) separately.
In order to assess the validity of the method and the effect of sample size on the outcomes of the algorithm, we resampled (with replacement) HLA class II haplotypes from the Burkholderia dataset (n = 10, 30, 50, 70, 100, 150, 200). We randomly assigned each peptide:HLA molecule an immunogenicity value Ej∈[0,0.55]; whether a peptide:HLA combination was considered immunogenic was decided by a cut-off on Ej; we considered a range of different cut-offs (0.1, 0.2, 0.3, 0.4, 0.5). We then simulated ELISpot data in the following way: for each subject, the probability of having a positive ELISpot was calculated using eq (1) above, whether the individual had a positive or negative ELISpot was then decided by drawing a random number from the unit interval and comparing it to this probability; if the random number was above the computed probability then the individual had a negative ELISpot, if it was below they had a positive ELISpot. We constructed ROC curves and calculated the AUC for the different cut-offs listed above.
Recently, Paul et al [10] have proposed a novel algorithm “RATE” to determine immunogenic peptide:HLA complexes from ELISpot data. We applied the RATE tool to both the Burkholderia and the Pseudomonas dataset. RATE returns two reports: a concise one, which contains the pHLA combinations who have a positive ELISpot relative frequency (RF) greater than two; and a complete report which ranks all peptide:HLA combinations in the population. We calculated the AUC for each dataset based on the RF of all the pHLA complexes whose posterior distributions passed the Beta(1,2) cut-off criterion described above.
While BIITE was designed to address the issue of elucidating peptide:HLA immunogenicity for class II HLA, we hypothesized that it could solve a larger class of problems (see Discussion). To explore the performance of BIITE on different datasets, we downloaded patient data and CD8+ T cell ELISpot outcomes for a range of HIV-1 peptides from [16]. We used 2-digit level typing for all HLAs, except for HLA-B*15 (where we used B*15:03 and B*15:10 if this subtyping was provided, and B*15 else), HLA-B*58 (B*58:01 and B*58:02) and HLA-A*30 (A*30:01 and A*30:02). For other HLA alleles, we either lacked sufficient HLA-typing on the 4-digit level, or there was one dominant 4-digit allele; see S5 Table for an overview. We applied the algorithm to the 32 peptides used for validation in [10] to sample the posterior probability distribution on the hypothesis space (sample size 250,000). The output was interpreted as described above and peptide:HLA combinations that were predicted to be immunogenic were matched against entries in the LANL A-list of HIV-1 epitopes and the general LANL database.
We designed and implemented a Bayesian framework, BIITE, to determine peptide:HLA immunogenicity from ELISpot data; Fig 1 outlines the algorithm. The algorithm determines the probability density function of the immunogenicity. The immunogenicity of a pHLA combination is the approximate probability that the peptide results in a positive ELISpot in a randomly chosen individual with the relevant HLA allele (this value would be exact if each subject expressed only one HLA). Strictly necessary inputs are the ELISpot experiment outcomes for each subject, together with their genotype for the loci of interest. Optionally, the user can also provide prior information to the algorithm; this is done on an HLA molecule-per-molecule basis. The output of the algorithm is a sample of the posterior distribution P(H|D), which we visualize by plotting the marginal posterior density for each peptide:HLA combination. This can be interpreted as the probability density function of the immunogenicity.
In order to test the BIITE algorithm, we used CD4+ T cell ELISpot data from a small cohort of exposed sero-positive individuals (N = 38). All overlapping 20-mer peptides from the BPSL2096 protein were tested (see Methods).
In Fig 2A, we present the marginal posterior distributions of the immunogenicity (solid lines) for all HLA molecules in combination with a single, representative peptide (BPSL2096, 1–20), calculated using a uniform prior. Especially for rare alleles (see S1 Table), the distributions are rather flat and uninformative. We also show the marginal posterior distributions, now assuming a non-uniform prior as described above (Fig 2A, dashed lines). The posterior distributions are now markedly more pronounced. This is of course driven by the prior assumptions, especially for the rare alleles. Nevertheless, there are divergences between the posterior and the prior. For example, for the representative peptide shown in Fig 2A both DRB1*01 (2 carriers) and DRB1*08 (4 carriers, one of which is homozygous), were assigned a positive prior (i.e. both were predicted to bind the peptide), yet they have different posteriors, reflecting the fact that both DRB1*01 carriers had positive ELISpot results, whilst there were two DRB1*08 carriers with a negative ELISpot.
We compared the posterior marginal distributions to the transgenic mouse model data (6 data points for each of the 17 peptides). To do so, one needs to collapse the posterior distribution to a decision: is the pHLA combination immunogenic or not, according to the model? For this analysis, we used the mode of the marginal posterior distribution as a summary statistic, and the decision rule was based on the mode being above a certain threshold. We then constructed a ROC curve by varying this threshold (Fig 2B). To interpret the strength of an algorithm, one compares it to the diagonal; this is the theoretical ROC curve for a random decision rule. The observed AUC is 71.3% when we use a uniform prior, and 65.4% when using a prior based on the predictions of NetMHCIIPan.
We performed an additional analysis on a Pseudomonas dataset, comprised of N = 58 subjects. The AUCs we found were 68.8% (without predictive prior), 79.3% (with predictive prior) (see Fig 3).
The subjects in the Pseudomonas study were divided into two groups, based on the number of positive sputum tests they had delivered. In this study, individuals with a high number of sputum cultures positive for PA had a lower immune responsiveness (lower number of peptide that elicited a T cell response) compared to individuals that were never sputum culture positive for PA, indicating that there were two different clinical groups with different peptide immunogenicities. Hence, we applied our algorithm to each of the groups separately, and predicted that the algorithm would perform poorly on the data from group 2 (who had very low response rates in their ELISpot assays). Indeed, for group 2 (subjects with at least one positive sputum sample), we found that BIITE was unsuccessful at resolving the HLA:epitope combinations found to be immunogenic in transgenic mice (AUC: 32% without prior; 69% with prior); this apparently low performance is not unexpected as these combinations were not immunogenic in humans. For group 1, BIITE reached AUCs of 69% (without prior) and 73% (with prior).
As described in the methods, we also analysed the HLA-I dataset that was analysed in [10]. For each peptide, HLAs were ranked by descending posterior mode provided by BIITE. To validate the results, we compared the outcome to the LANL A-list [18] of best defined epitopes in HIV-1. This is not ideal, since a peptide:HLA combination can be immunogenic without being featured on this A-list (i.e. we will tend to overestimate false positives); nevertheless, we used this list to determine lower bounds on the AUC. Both BIITE (AUC = 81%) and RATE (AUC = 77%) performed well and were able to identify A-list peptide:HLA combinations (S6 Table). However, RATE’s ‘concise output file’ (S7 Table) included a number of peptide:HLA combinations which are most likely not immunogenic but appear to be so due to linkage in the HLA locus (highlighted in red in S7 Table. Strong linkage: p-value chi-square test < 2.2*10E-16; linkage: p-value chi-square test < 0.01); this was not a problem with the BIITE algorithm. This shows that BIITE is successful in handling possible confounding due to linkage disequilibrium between genetic loci, a recurring problem that is not dealt with by RATE.
Determining the immunogenicity of peptide:HLA (pHLA) combinations is an important task when studying correlates of immunity in infection, which in turn may inform vaccine design. Different experimental and computational approaches have been used to establish this. The experimental approaches include assessing the binding strength of the pHLA complex, in vitro antigen presentation studies using CD4+ T cell clones, and HLA class II- transgenic mouse models to mimic pathogen peptide expression in a human host. Each of these methods has shortcomings. While binding (whether measured in binding assays or predicted) of a peptide and immunogenicity of that peptide are correlated, binding is insufficient to guarantee immunogenicity which entails other processes both in the antigen presentation pathways (such as cleavage of protein into smaller peptides and the loading of peptides onto HLA-II molecules) but also in the selection and expansion of the T cell clone(s) bearing the relevant TCR(s) in vivo. More direct assays of immunogenicity such as CD4+ T cell cloning from bulk cultures and assessment of peptide presentation using transfectants expressing individual HLA-II heterodimers are cumbersome and intractable for large-scale studies [19,20]. Transgenic mice are good in vivo models encompassing peptide processing and presenting pathways, but are labour-intensive to develop and maintain. Furthermore, the overlap between which peptides are immunogenic in humans and in mouse models is not perfect and seems to be highly dependent on the choice of the effector cell [21]. As the mouse model incorporates a single human HLA, it is also unsuitable to appreciate the effects of immunodominance, where different HLAs may present different peptides simultaneously to the host’s immune system [22].
ELISpot assays use ex vivo PBMCs from exposed hosts to test peptidic immunogenicity. In theory, this approach includes effects of peptide presentation pathway idiosyncrasies, binding affinities between peptide and HLA, and immunodominance. As a trade-off however, we lose the clarity of the experimental approaches where each HLA is studied independently, and it is difficult to assess which peptide:HLA combinations cause the observed ELISpot outcomes.
Here, we have presented a Bayesian framework to perform this analysis. We have shown its accuracy on both simulated and experimental data, and have determined the effect of sample size. While higher sample sizes (100–200 subjects) would have yielded better results, the analysis on a very limited dataset of 38 Burkholderia-sero-positive exposed subjects performed as expected (AUC = 71%) from simulated data. A caveat should be given regarding the composition of the data set: analysis on data sets with high rates of positive ELISpots suffer from poor identifiability. Nevertheless it is remarkable that with such a small dataset relatively high specificity and sensitivity could be achieved; while most algorithms for predicting immunogenicity or binding require training datasets with hundreds of data points per HLA allele, we typically have less than 10 data points per allele. Furthermore, unlike prediction algorithms, BIITE explicitly uses cohort data, which makes it suitable to interrogate differences in immunogenicity between patient groups responding differently to the same pathogen; this would be impossible for a binding prediction algorithm.
This tool is available online as the R package BIITE and is, in essence, not ELISpot or HLA class II-specific. In general, it can be used to attribute Boolean outcomes (such as therapeutic success or disease progression) to allelic variation on multiple genes. As an example, we have applied BIITE to an HLA class I ELISpot dataset (HIV-1). In general, to apply the algorithm to a new problem it is necessary to consider carefully the source of the priors if these are to be used; establish the length of the MH chain needed to obtain a representative sample of the posterior distribution; and to consider issues of sample size and collinearity in the dataset.
It also allows for the addition of prior information ad libitum: we can envisage a scenario where not only binding predictions (from NetMHCIIPan or other binding prediction algorithms) but also binding data and transgenic mouse data are included in the prior. Of course, conflicting binding information would result in different priors, and hence influence the outcome of the algorithm. However, the effect of the prior is dependent on sample size; in a suitably large dataset, the influence of the prior is outweighed by the likelihood factor. Alternatively, confidence in the prior can be expressed in the prior itself: the more ‘flat’ a prior, the less it will influence the posterior distribution.
In summary, we have developed a general approach which can be used for analysing complex biological data. Whether or not a peptide:HLA combination is immunogenic involves a number of interconnected immunological processes. To sidestep this complexity, reductionist experimental approaches are often taken, e.g. in vitro stimulation with APCs expressing a single HLA molecule; these approaches do not recapitulate a number of, potentially important, biological details. We have developed an approach that allows us to use the most pertinent but also most complex experimental data (ELISpot) and interrogate it to obtain accurate results.
|
10.1371/journal.ppat.1002580 | Calcium Influx Rescues Adenylate Cyclase-Hemolysin from Rapid Cell Membrane Removal and Enables Phagocyte Permeabilization by Toxin Pores | Bordetella adenylate cyclase toxin-hemolysin (CyaA) penetrates the cytoplasmic membrane of phagocytes and employs two distinct conformers to exert its multiple activities. One conformer forms cation-selective pores that permeabilize phagocyte membrane for efflux of cytosolic potassium. The other conformer conducts extracellular calcium ions across cytoplasmic membrane of cells, relocates into lipid rafts, translocates the adenylate cyclase enzyme (AC) domain into cells and converts cytosolic ATP to cAMP. We show that the calcium-conducting activity of CyaA controls the path and kinetics of endocytic removal of toxin pores from phagocyte membrane. The enzymatically inactive but calcium-conducting CyaA-AC− toxoid was endocytosed via a clathrin-dependent pathway. In contrast, a doubly mutated (E570K+E581P) toxoid, unable to conduct Ca2+ into cells, was rapidly internalized by membrane macropinocytosis, unless rescued by Ca2+ influx promoted in trans by ionomycin or intact toxoid. Moreover, a fully pore-forming CyaA-ΔAC hemolysin failed to permeabilize phagocytes, unless endocytic removal of its pores from cell membrane was decelerated through Ca2+ influx promoted by molecules locked in a Ca2+-conducting conformation by the 3D1 antibody. Inhibition of endocytosis also enabled the native B. pertussis-produced CyaA to induce lysis of J774A.1 macrophages at concentrations starting from 100 ng/ml. Hence, by mediating calcium influx into cells, the translocating conformer of CyaA controls the removal of bystander toxin pores from phagocyte membrane. This triggers a positive feedback loop of exacerbated cell permeabilization, where the efflux of cellular potassium yields further decreased toxin pore removal from cell membrane and this further enhances cell permeabilization and potassium efflux.
| The adenylate cyclase toxin (CyaA) of pathogenic Bordetellae eliminates the first line of host innate immune defense by inhibiting the oxidative burst and complement-mediated opsonophagocytic killing of bacteria. The toxin penetrates myeloid phagocytes, such as neutrophil, macrophage or dendritic cells, and subverts their signaling by catalyzing a rapid and massive conversion of intracellular ATP to the key signaling molecule cAMP. In parallel, the toxin forms cation-selective pores and permeabilizes the cytoplasmic membrane of phagocytes. This so-called ‘hemolysin’ activity synergizes with the enzymatic AC activity of CyaA in promoting apoptotic or necrotic cell death, depending on the toxin dose. Moreover, the pore-forming activity promotes activation of NALP3 inflammasome and release of interleukin IL-1β. We show here that the capacity of CyaA to permeabilize phagocytes depends on its ability to mediate influx of extracellular calcium ions into cells. This enables bystander CyaA pores to escape rapid macropinocytic removal from cell membrane and exacerbate the permeabilization of cells. These observations set a new paradigm for the mechanism of action of pore-forming RTX leukotoxins on phagocytes.
| By instantaneously disrupting bactericidal functions of host phagocytes, the adenylate cyclase toxin-hemolysin (CyaA, ACT, or AC-Hly) plays a major role in virulence of pathogenic Bordetellae [1]. The toxin rapidly paralyzes phagocytes [1], [2] by translocating across their cytoplasmic membrane an N-terminal adenylate cyclase enzyme (AC) domain (∼400 residues) that binds cytosolic calmodulin and converts ATP to a key signaling molecule, cAMP [1]. In parallel, the multidomain ∼1300 residues-long RTX (Repeat in ToXin) cytolysin moiety of CyaA acts independently as a pore-forming leukotoxin and hemolysin [1]. This employs a hydrophobic pore-forming domain (residues 500 to 700), a domain with covalently palmitoylated lysine residues 860 and 983, and a typical calcium-binding RTX repeat domain within the last 700 residues of CyaA that accounts for receptor binding [1]. CyaA can oligomerize into small cation-selective pores that mediate efflux of cytosolic potassium ions from cells [3]–[6], eventually provoking colloid-osmotic cell lysis [7]–[9]. This activity synergizes with cytotoxic signaling of the translocated AC enzyme in bringing about the final cytotoxic action of CyaA [8], [9].
The structure of CyaA in target membrane remains unknown. Accumulated indirect evidence strongly suggests that the cell-invasive AC and the pore-forming activities are associated with distinct subpopulations of CyaA conformers within target cell membrane. Indeed, translocation of the AC domain across cellular membrane and the pore-forming activity of CyaA can be dissociated by temperature, calcium concentrations, or altered acylation [5], [10], [11]. Moreover, the balance between the two activities can be largely shifted in either direction by specific residue substitutions. Several CyaA variants with strongly enhanced pore-forming activity and reduced or nil capacity to deliver the AC domain into cells could be generated. Recently, a CyaA defective in formation of toxin pores but intact in AC delivery into cells could also be constructed [3], [6], [12]. Summarizing this evidence, a model was proposed that predicts the co-existence of two distinct toxin conformers inside target cell membrane. These would operate in parallel and would employ the same segments of the hydrophobic domain in an alternative manner. One conformer would insert into cell membrane as toxin translocation precursor that accomplishes delivery of the AC domain into cells (cell-invasive activity). The second conformer would account for formation of oligomeric CyaA pores [3], [6], [12].
The primary targets of CyaA appear to be host myeloid phagocytes, to which the toxin binds through their αMβ2 integrin, known as CD11b/CD18, CR3 or Mac-1 [13]. Recently, we could show that AC translocation into cells occurs in two steps. First, the membrane–inserted translocation precursors of CyaA generate a calcium-conducting path across cell membrane and mediate influx of extracellular calcium ions into cells [14]. The incoming Ca2+ than activates calpain-mediated processing of the talin tether of CD11b/CD18, which triggers relocation of the toxin-receptor complex into cholesterol-rich membrane lipid rafts. There the translocation of the AC domain into cytosol of cells across the cytoplasmic membrane of phagocytes is completed [15]. Unlike for most other enzymatic toxins, the capacity of CyaA to deliver its AC enzyme component into target cells does not depend on receptor-mediated endocytosis. Instead, the AC domain translocates into cytosol directly across the cytoplasmic membrane of phagocytes. Indeed, inhibitors of endocytic pathways interfere neither with translocation of the AC domain into cells and elevation of cytosolic cAMP, nor with AC domain-mediated delivery of inserted CD8+ T cell epitopes into the cytosol of antigen presenting cells [16]–[18]. Clathrin-dependent endocytic uptake of CyaA together with its receptor CD11b/CD18 has, however, previously been observed [19], [20] and it was found to account for the capacity of CyaA-AC− toxoids to deliver cargo antigens into dendritic cells for endosomal processing and MHC II-restricted presentation to CD4+ T cells [18].
Therefore, we analyzed here the mechanism of endocytic uptake of CyaA from cell membrane and its relevance for toxin action. We show that by conducting calcium ions across the target membrane into cytosol of cells, the translocating CyaA molecules control the rate of toxin pore removal from cellular membrane and thereby support the permeabilization of phagocytes.
Recently we found that despite binding to CD11b/CD18, the CyaA constructs lacking the hydrophobic domain failed to deliver a fused MalE antigen for presentation to CD4+ T lymphocytes by dendritic cells (Figure S1). To address the role of functional integrity of CyaA in its endocytic trafficking, we employed live microscopy imaging.
To avoid the interference of massive phagocyte ruffling and death from ATP depletion and cAMP signaling provoked by enzymatically active CyaA at concentrations required for imaging [8], [21], we used fluorescently labeled and enzymatically inactive CyaA-AC− toxoids. To enable assessment of their capacity to deliver antigens for presentation on MHC class I and II molecules, the toxoids were further tagged by insertion of a CD8+ T-cell epitope (SIINFEKL) from ovalbumin at position 336 and of a MalE CD4+ T-cell epitope (NGKLIAYPIAVEALS) at position 108 of the AC domain, respectively. Trafficking of such tagged dCyaA was then compared to that of a doubly mutated CyaA-E570K+E581P-AC− toxoid (dCyaA-KP) that carried a combination of debilitating substitutions of glutamate residues at positions 570 and 581 within the pore-forming domain. That construct was previously shown to retain a full capacity to bind the toxin receptor CD11b/CD18, without being able to conduct calcium ions into cells, to associate with lipid rafts and translocate the AC domain across cell membrane, or to form oligomeric CyaA pores within cellular membrane, respectively [3], [15].
As shown in Figure 1, insertion of the epitope tags and fluorescent labeling did not alter the expected properties of the dCyaA toxoid, which elevated [Ca2+]i in J774A.1 cells (Figure 1A) and relocated into lipid rafts in J774A.1 membrane (Figure 1B). As also expected, the dCyaA-KP toxoid lacked all these activities. When examined by live cell imaging, the two fluorescently labeled toxoids exhibited intriguingly different kinetics and patterns of endocytic uptake. As documented by representative images in Figure 1C and quantified in Figure 1D, dCyaA was detected predominantly within the plasma membrane, or in fluorescent vesicles located beneath the membrane of J774A.1 cells throughout the 20 minutes of incubation at 37°C (Video S1). In contrast, dCyaA-KP was taken-up much faster and accumulated massively within endocytic vesicles dispersed through the cytosol of cells already within 5 minutes from toxoid addition. The dCyaA-KP toxoid was then almost completely removed from cell membrane within 10 minutes (Figure 1C, Video S2), as quantified by counting of the intracellularly located fluorescent vesicles for a set of inspected cells (Figure 1D, see Figure S2 for the method). Identical results were observed upon swapping of the fluorescent labels (not shown), thus excluding the impact of the used dyes on toxoid trafficking. While the two toxoids entered into quite different endocytic pathways, the uptake of both was receptor-mediated and depended on toxoid binding to CD11b/CD18. Blocking of the receptor by the M1/70 antibody abrogated, indeed, cell binding and endocytosis of both toxoids, as documented in Figure 1C and Figure 1E.
To characterize the differing uptake pathways, we next examined the co-localization of the toxoids with transferrin, a marker of clathrin-dependent endocytosis [22]. As shown by representative images in Figure 2A and quantified by Pearson's correlation coefficients (P.c.c.) for compared channels in Figure 2B, the co-localization of dCyaA with transferrin (Dyomics 547) increased in time and was near-complete after 60 minutes of co-incubation with cells. In contrast, a weak and progressively decreasing co-localization of dCyaA-KP with transferrin was observed over the same time interval. Hence, while dCyaA was transiting through the early and/or recycling compartment together with transferrin, the dCyaA-KP toxoid took a different pathway.
To corroborate this observation, the uptake of the two toxoids was examined in RAW 264.7 macrophages expressing a dominant negative variant of the EPS-15 protein (DN EPS-15-GFP, DIII), which selectively interferes with clathrin-dependent endocytosis [23]. As documented by a representative image in Figure 2C (left), the accumulation of transferrin and dCyaA at intracellular sites was abrogated in DN EPS-15-GFP-transfected cells (blue) and no co-localization of cell-associated dCyaA with transferrin was observed upon removal of surface-associated transferrin with a low pH buffer wash. As further documented by the z-axis projections, few if any intracellular vesicles containing dCyaA were observed inside DN EPS-15-GFP-transfected cells and the membrane-associated dCyaA was located exclusively inside fluorescent patches on cell surface. In striking contrast, the endocytic uptake of dCyaA-KP was unaffected in DN EPS-15-GFP-transfected cells and intracellular vesicles containing dCyaA-KP were clearly observed in the z-axis projections (Figure 2C, right panel). Neither of the two toxoids exhibited any co-localization with caveolin-1 (Figure S3). By difference to dCyaA, however, the dCyaA-KP exhibited a strong co-localization with soluble fluorescent BSA serving as a fluid phase uptake marker. As also shown in Figure 3C and quantified in Figure 3D, the kinetics of dCyaA-KP endocytic uptake was strongly decelerated upon pretreatment of cells with wortmannin (1 µM), a PI3 kinase inhibitor blocking macropinocytosis but not micropinocytosis [24]. In contrast, no inhibition of dCyaA-KP uptake was observed upon treatment with dynasore or chlorpromazine, the inhibitors of clathrin-dependent endocytosis [25], [26]. It can, hence, be concluded that dCyaA was endocytosed through a decelerated clathrin-dependent pathway, while dCyaA-KP was internalized with the cytoplasmic membrane by a macropinocytic mechanism.
Clathrin-dependent endocytosis was previously shown to enable dCyaA-mediated delivery of model antigens into CD11b+ dendritic cells for endosomal processing and MHC class II-restricted presentation to CD4+ T cells [18]. Therefore, we examined whether altered endocytic trafficking affected the antigen delivery capacity of the dCyaA-KP toxoid.
As documented in Figure 4A and quantified in Figure 4B, following one hour of incubation with RAW 264.7 macrophages expressing Rab-7-EGFP, fluorescently labeled material derived from either of the toxoids accumulated within organelles positive for the lysosomal/late endosomal marker Rab7. dCyaA and dCyaA-KP toxoid-derived fluorescence also co-localized to the same extent with the MHC II–EGFP molecules in a subset of intracellular vesicles of bone marrow derived dendritic cells from a MHC Class II–EGFP knock-in mice, within 120 minutes of incubation (Figure 4C, 4D). Despite entering cells by different endocytic pathways, hence, the two toxoids or their fragments reached similar acidic endosomes.
To test if faster macropinocytic uptake lead to alteration of dCyaA-KP processing, lyzates from toxoid-pulsed J774A.1 cells were probed in Western blots with the 9D4 MAb that recognizes C-terminal RTX repeats of CyaA. As shown in Figure 5A, comparable amounts of the ∼200 kDa forms of both toxoids and of their fragments were detected in lyzates of cells pretreated with inhibitors of endocytosis and protease inhibitors, like 0.01% sodium azide plus 10 mM 2-deoxy glucose (2DG), or a protease inhibitor cocktail plus 1 mM chloroquine. As compared to dCyaA, however, a notably faster degradation dCyaA-KP occurred in uninhibited cells (Figure 5A), as judged from the detected amounts of smear corresponding to partially digested molecules. dCyaA-KP toxoid, hence, appeared to be degraded faster and more completely, albeit the final protease-resistant fragments of both toxoids appeared to be of similar size.
To examine if faster uptake and degradation affected the capacity of dCyaA-KP to deliver cargo epitopes for endosomal processing and MHC II-restricted presentation, the relative capacity of the two toxoids to deliver CD4+ T cell epitopes was assessed. As shown in Figure 5B by comparison to dCyaA, the dCyaA-KP toxoid exhibited an about ten-fold reduced capacity to deliver the MalE epitope for presentation by BMDCs to CRMC3 CD4+ T hybridoma cells [27]. To corroborate this observation, the MalE epitope was replaced by the OVA258–276 epitope recognized by MF2.2D9 CD4+ T hybridoma and new dCyaA and dCyaA-KP toxoids were produced. As shown in Figure 5C, however, a similarly reduced capacity to deliver the OVA258–276 epitopes for MHC II-restricted presentation was again observed with dCyaA-KP, as compared to dCyaA. This suggested that faster proteolytic destruction of the cargo epitope during trafficking through the macropinocytic-like pathway may have accounted for the reduced efficacy of dCyaA-KP in antigen delivery.
To test whether the loss of capacity to mediate influx of Ca2+ ions into cells committed dCyaA-KP for the rapid macropinocytic uptake and subsequent degradation, the J774A.1 cells were incubated with dCyaA-KP in the presence of ionomycin, a calcium ionophore that permeabilizes cells for extracellular Ca2+. As documented in Figure 6A, Video S3, and as quantified in Figure 6B, respectively, internalization of dCyaA-KP into fluorescent intracellular vesicles was strongly delayed upon treatment of J774A.1 cells with 5 or 10 µM ionomycin in the presence of 1.9 mM Ca2+. As also shown in Figure 6A (Video S4) and quantified in Figure 6B and 6C, dCyaA-KP was redirected into a slower uptake pathway even more efficiently upon co-incubation at a 1∶1 ratio with intact dCyaA. This goes well with our previous observation that permeabilization of cells for Ca2+ in trans by intact dCyaA rescued in part the defect of a CyaA-KP construct and mobilized it into lipid rafts [15]. Upon co-incubation with dCyaA, a fraction of biotinylated dCyaA-KP protein was indeed found to float in sucrose gradients with detergent-resistant membrane (Figure 6D). Furthermore, following co-incubation with dCyaA, the dCyaA-KP appeared to be endocytosed with the same kinetics and through the same pathway as the intact toxoid (Videos S1 and S4). The two proteins co-localized within the same endocytic vesicles at 30 and 60 minutes of incubation with cells (Figure 6E, 6F). Hence, permeabilization of cells for Ca2+ by ionomycin or intact toxoid rescued dCyaA-KP from the rapid macropinocytic membrane uptake pathway and redirected it for decelerated and clathrin-dependent endocytosis.
We have previously observed that the CyaA-ΔAC hemolysin construct lacking the AC domain of CyaA (Δ1–373) was unable to promote Ca2+ influx into monocytes and was essentially unable to provoke lysis of J774A.1 cells [14]. However, on planar lipid bilayers, or on sheep erythrocytes devoid of endocytic mechanisms, the CyaA-ΔAC hemolysin exhibited the same specific pore-forming and hemolytic activities as the full-length dCyaA (CyaA-AC−) [4]. We thus hypothesized that its inability to lyze J774A.1 cells might be due to rapid removal of the CyaA-ΔAC pores from the cytoplasmic membrane of J774A.1 cells. As shown in Figure 7A, the CyaA-ΔAC hemolysin elicited much slower efflux of cytosolic K+ from J774A.1 cells then the full-length toxoid. In cells exposed to 3 µg/ml of enzymatically inactive CyaA-AC− a complete drop of cytosolic [K+]i concentration down to 10 mM was reproducibly observed within 20 minutes, while in cells exposed to equal amounts of CyaA-ΔAC the [K+]i only decreased to 80 mM (Figure 7A).
To determine if this was due to rapid removal of CyaA-ΔAC pores from cell membrane, we assessed the capacity of CyaA-ΔAC to elicit K+ efflux on cells with membrane trafficking inhibited upon preincubation in media containing 0.01% (w/w) sodium azide and 10 mM 2-deoxy-glucose (2DG). This treatment potentiated the cell-permeabilizing activity of full-length CyaA-AC− toxoid (Figure 7A, 7B), but did not promote any significant K+ efflux from cells on its own despite causing an about five-fold drop of cellular ATP levels (Figure 7A, 7B and 7C). As further documented in Figure 7E and 7F (Videos S5 and S6), the strong inhibition of the CyaA-ΔAC uptake was accompanied by a steep increase of the specific capacity of CyaA-ΔAC to promote K+ efflux (Figure 7A and 7B) from cells and lyze J774A.1 monocytes (Figure 7D). It can thus be concluded that rapid macropinocytic internalization with cell membrane was strongly restricting the cell permeabilizing and cytolytic capacities of the CyaA-ΔAC hemolysin pores.
As further shown in Figure 7G, despite the prolonged persistence in the cytoplasmic membrane of cells upon inhibition of endocytosis, the CyaA-ΔAC hemolysin failed to associate with lipid rafts (compare CyaA-ΔAC panel in Figure 7G and upper dCyaA panel of Figure 1B for comparable toxoid loading). This goes well with our previous observation that mobilization into lipid rafts with CD11b/CD18 depends on the capacity of CyaA to conduct calcium ions into cells to activate talin cleavage by calpain [15]. More importantly, this result shows that CyaA pores can form outside of lipid rafts within the bulk phase of cell membrane.
We next tested if Ca2+ influx induced in trans would delay removal of CyaA-ΔAC from cytoplasmic membrane and extend thereby its capacity to permeabilize J774A.1 cells. Mobilization of Ca2+ into cells with ionomycin, however, caused on its own a high unspecific leakage of K+ from J774A.1 cells (data not shown). Therefore, an alternative approach was used, exploiting the capacity of the 3D1 MAb to lock CyaA molecules in the conformation of membrane-inserted ‘translocation precursors’ that conduct calcium ions across the cytoplasmic membrane of cells [15]. CyaA-ΔAC was preincubated with 3D1, or with an isotype control IgG1 MAb (TU-01, 20 µg/ml) and the capacity of the CyaA-ΔAC to elicit K+ efflux from J774A.1 cells was assessed. As documented in Figure 8A, while the 3D1 MAb alone did not cause any elevation of [Ca2+]i, the binding of 3D1 conferred on CyaA-ΔAC the capacity to promote rapid influx of Ca2+ into monocytes. As shown in Figure 8B, this allowed association of detectable amounts of CyaA-ΔAC molecules with lipid rafts. More importantly, the capacity of CyaA-ΔAC to permeabilize J774A.1 cells for K+ efflux was strongly enhanced in the presence of 3D1 MAb and equaled the specific cell-permeabilizing activity of the intact and enzymatically active CyaA, as shown in Figure 8C and 8D. Furthermore, this enhancement of cell-permeabilizing activity was accompanied by a strong deceleration of endocytic uptake of CyaA-ΔAC (Figure 8E, 8F).
To test the hypothesis that toxoid-induced K+ efflux itself contributed further deceleration of endocytosis, the clathrin-mediated uptake of dCyaA was examined in media supplemented with 50 mM potassium ions. As documented in Figure 9A and quantified in Figure 9B, the endocytosis of dCyaA was clearly accelerated when efflux of K+ from cells was counteracted by addition of potassium ions into media. Collectively, hence, these results show that CyaA-mediated Ca2+ influx rescues hemolysin pores from rapid macropinocytic uptake from cell membrane and thus extends their cell-permeabilizing and cytolytic action.
The pore-forming activity was previously shown to synergize with the ATP-depleting and cAMP-signaling activities of the cell-invasive AC domain of CyaA [8], [9]. Therefore, we examined to which extent does endocytic uptake modulate the cytotoxicity of fully active (AC+) toxin.
We assessed first the impact of cAMP accumulation on the uptake of CyaA, using a fluorescently labeled CyaA-KP construct (AC+) that exhibits only a residual capacity to deliver the AC enzyme into cells. Unlike active toxin, hence, CyaA-KP does not provoke cell death at the high concentrations (1 to 5 µg/ml) employed in live cell imaging. As shown in Figure 10A, at such concentrations the dCyaA-KP produced easily detectable cAMP amounts in cells, while the extent of its endocytic uptake was the same as that of the enzymatically inactive (AC−) dCyaA-KP toxoid (Figure 10B, 10C). Since both proteins were unable to mediate detectable Ca2+ influx into J774A.1 cells (ref. [15] and Figure 1B), it can be concluded that cAMP levels alone do not noticeably influence the rate of endocytic uptake of CyaA.
It was next important to characterize the rate of endocytic uptake of CyaA at as low toxin concentrations as that presumably encountered by host phagocytes during Bordetella infections. Therefore, the kinetics of CyaA endocytosis by J774A.1 cells was assessed by a flow cytometry assay measuring the amount of biotinylated CyaA that remains accessible to binding by streptavidin on cell surface. As shown in Figure 11, pulsing of J774A.1 cells for 5 minutes at 37°C with 200 ng/ml of CyaA-biotin yielded about 1 ng of toxin stably bound to 3×105 washed cells, when kept on ice. In contrast, the amounts of surface-exposed CyaA-biotin decreased progressively upon transfer of cells to 37°C, with ∼80% of CyaA being removed from cell surface within 15 minutes. As also shown in Figure 11, the rate of CyaA-biotin uptake was reduced by about a factor of two in the presence of a 5-fold (non-saturating) excess of unlabeled CyaA-AC− toxoid (1 µg/ml) that enhanced Ca2+ influx and K+ efflux across the membrane of J774A.1 cells [14].
To address the relative contribution of cell surface retention of CyaA pores to the overall cytotoxic potency of low amounts of intact (AC+) CyaA, the natively fatty-acylated Bp-CyaA purified from B. pertussis 18323/pHSP9 was used [28]. Bp-CyaA exhibits about fourfold higher specific pore-forming (hemolytic) activity than the recombinant rEc-CyaA produced in E. coli ([11] and Figure S4). Therefore the use of Bp-CyaA allowed maximizing the observable amplitude of changes in cell permeabilizing and cytolytic capacity that would result from alterations of the rate of toxin pore removal from cell membrane. As indeed shown in Figure 12, as little as 100 ng/ml of Bp-CyaA promoted a detectable LDH release form J774A.1 cells already in 2 hours of incubation. Inhibition of endocytic uptake of Bp-CyaA with 2DG and sodium azide treatment then increased the cytolytic potency of Bp-CyaA by a factor of 2 to 3 (Figure 12A) and strongly increased the capacity of Bp-CyaA to permeabilize cells for K+ efflux, which become well observable already at 300 ng/ml of the toxin (Figure 12B).
We show here that by conducting Ca2+ ions across target cell membrane, CyaA decelerates its endocytic uptake and escapes from rapid macropinocytic removal from cell membrane and destruction in endosomes. By redirecting the toxin into a decelerated clathrin-dependent uptake pathway, the calcium-conducting activity of toxin translocation intermediates protracts toxin pore persistence within cytoplasmic membrane, thus extending phagocyte permeabilization and maximizing cytotoxic action of CyaA, as summarized in the model proposed in Figure 13.
This mechanism could be directly demonstrated here for the CyaA-AC− toxoid and CyaA-ΔAC hemolysin variants that lack the AC enzyme activity and could be used at sufficiently high concentrations for live cell imaging. Imposing on CyaA-ΔAC a conformation that enabled it to conduct Ca2+ into cells, indeed, rescued the hemolysin from the macropinocytic pathway, and by decelerating its removal form cell membrane, it particularly enhanced the otherwise very low cell-permeabilizing and cytolytic capacity of CyaA-ΔAC.
Using intact toxin (AC+) purified form B. pertussis cells at close to physiologically relevant concentrations (200 ng/ml), we found that inhibition of endocytic uptake of Bp-CyaA enhanced its capacity to lyze cells. It was, however, not possible to design an experiment that would directly prove the role of calcium-induced deceleration of CyaA uptake in its cytotoxic action. Such demonstration would only be possible if inhibition of Ca2+ influx would leave the other cytotoxic activities of CyaA unaffected. This would then allow to test whether cytolytic activity of CyaA is reduced by acceleration of the endocytic uptake of the intact toxin. However, Ca2+ influx into cells is a prerequisite for CyaA relocation into lipid rafts and subsequent AC domain translocation into cell cytosol [15]. Therefore, blocking of Ca2+ influx by whatever means inevitably ablates also the major component of the cytotoxic activity of CyaA, namely the capacity of its AC enzyme to reach cell cytosol and catalyze unregulated dissipation of cytosolic ATP into cAMP to impair cellular signaling.
We have previously shown that upon initial binding of the CD11b/CD18 integrin, the CyaA toxin inserts into the membrane lipid bilayer as a translocation precursor, in which the segments of the AC domain cooperate with segments of the pore-forming domain in forming a novel calcium-conducting path across the phagocyte membrane that mediates influx of extracellular Ca2+ ions into cells [14], [15]. Activation of calpain by incoming Ca2+ then yields cleavage of the talin tether of CD11b/CD18 and liberates the toxin-receptor complex for recruitment into membrane lipid rafts [15]. This process appears to be entirely independent of the cAMP-generating capacity of CyaA in CD11b+ J774A.1 macrophage cells, as a quite comparable amplitude and even faster initial kinetics of calcium mobilization into cells is observed with the enzymatically inactive dCyaA variant (CyaA-AC−) than by intact CyaA [14], [15]. The results presented in this study then show that it is rather the permeabilization of cells for influx of Ca2+, than the toxin relocation into lipid rafts itself, which allows the toxoid to escape the rapid macropinocytic removal from cell surface with the cytoplasmic membrane. Moreover, lipid rafts are tiny microdomains in the membrane that would accommodate only limited amounts of the hemolysin, while a quantitative escape from rapid macropinocytosis of dCyaA-KP toxoid and its redirection for decelerated endocytosis was observed upon co-incubation with dCyaA (Figure 6A). This indicates that by permeabilizing cells for calcium influx, the AC-translocating CyaA conformers provoke also deceleration of endocytosis of the bystander CyaA pores that can form outside the rafts, in the bulk phase of cell membrane.
Such conclusion would also be in line with the observed opposing phenotypes of the CyaA-E509K+516K (CyaA-KK) and CyaA-E570Q+K860R (CyaA-QR) variants of CyaA. While the CyaA-KK exhibits a strongly enhanced specific pore-forming activity, its capacity to promote Ca2+ influx into cells is very low [14], [29]. On the opposite, the CyaA-QR toxin exhibits a very low capacity to form pores, to permeabilize cells and elicit K+ efflux, while it mediates normal levels of calcium influx into cells [6], [15]. This suggests that K+ efflux and Ca2+ influx are two parallel and independent processes that are associated with distinct conformational and/or oligomeric states of the two co-existing conformers of membrane-inserted CyaA.
In this respect, it is puzzling that binding of the 3D1 antibody to CyaA-ΔAC exacerbated at the same time its capacity to conduct calcium ions into cells, as well as its cell permeabilizing capacity. This would, perhaps, be best explained by the bystander effect mentioned above. Due to association-dissociation equilibrium, the antibody would lock only a fraction of membrane-inserted CyaA-ΔAC molecules into a calcium-conducting conformation. The other fraction of hemolysin molecules, not bound with 3D1, would hence be free to form the cell-permeabilizing oligomeric pores promoting K+ efflux, benefiting from the calcium-induced deceleration of the endocytic uptake of the hemolysin pores. It remains, nevertheless, to be conclusively shown that bridging of CyaA-ΔAC dimers by the bivalent 3D1 antibody does not also contribute to the enhanced cell permeabilization capacity and K+ efflux mediated by the CyaA-ΔAC hemolysin complexes with the antibody. 3D1 might, indeed, potentially stabilize the formed oligomeric pores against dissociation within the membrane. At present it can neither be formally excluded that the complex of membrane-inserted CyaA-ΔAC with 3D1 might be conducting at the same time the Ca2+ ions into cells and the K+ ions out of the cell. For intact CyaA the accumulated evidence strongly argues against such possibility (see above). Due to absence of the translocated AC domain, however, the 3D1-locked conformers of CyaA-ΔAC may be capable of oligomerization into pores enabling K+ efflux, or the calcium-conducting path formed by these conformers may be accessible for efflux of cytosolic K+ ions in the absence of the AC domain.
A particularly intriguing observation is that the Ca2+ influx and K+ efflux-promoting activities of the toxin synergize in manipulating the pathway and kinetics of CyaA endocytosis. Indeed, the decrease of intracellular potassium level was repeatedly shown to cause inhibition of clathrin-mediated endocytosis [30], [31]. Hence, the more the cell becomes permeabilized for efflux of K+, upon inhibition of macropinocytic uptake of toxin pores by incoming Ca2+, the more the potassium leakage from cells through toxin pores further decelerates the clathrin-dependent removal of CyaA pores from cell membrane. Such cooperation of Ca2+ influx and K+ efflux-mediating activities of CyaA would thus generate a positive feedback loop, exacerbating potassium depletion due to steadily increasing extent of cell membrane permeabilization by persisting and/or newly forming CyaA pores.
The above outlined positive feedback loop of potassium efflux was apparently operating under the used experimental conditions, since vesicles containing dCyaA were observed to accumulate as tightly attached to, or located just beneath the cell membrane, for over 20 minutes from toxoid addition (Figure 1 and Video S1). This shows that excision of clathrin-coated vesicles from cell membrane was inhibited and protraction of cell membrane permeabilization by the persisting pores then fed back into deceleration of endocytic uptake of dCyaA.
The enzymatically active CyaA could not be used in this study for analysis of endocytic trafficking of CyaA by live cell microscopy imaging, as at the high concentrations of labeled proteins (>1 µg/ml), required for this type of analyzes, the enzymatically active CyaA uncontrollably impairs intracellular trafficking by rapid depletion of ATP and induction of cell death [8], [9]. Indeed, the cytotoxic action of enzymatically active CyaA on CD11b+ phagocytes was documented repeatedly at doses lower than 10 ng/ml, where less than 1 ng/ml of active CyaA toxin quantitatively inhibits oxidative burst in neutrophils [32]. As low CyaA doses as 5 to 10 ng/ml elicit monocyte ruffling and a near-instant inhibition of CR3-mediated opsonophagocytosis or arrest of the fluid-phase uptake, respectively. This appears to be due to cAMP signaling-mediated inhibition of the small GTPase RhoA and possibly of the PI3 kinase [21]. The results obtained here with the toxoids appear, nevertheless, to be relevant also to the understanding of endocytosis and of cytotoxic action of the enzymatically active CyaA. By using a newly developed FACS assay for cell surface accessibility of bound CyaA, we could approach here the kinetics of endocytic removal of intact CyaA from cell surface at close to physiologically low toxin concentrations (200 ng/ml). Under such conditions, the receptor-bound CyaA was found to be progressively removed from cell surface over 15 minutes of incubation with about one quarter of toxin molecules escaping the uptake into endosomes and remaining exposed on the surface of cell membrane. This endocytic uptake of intact CyaA from cell membrane was noticeably slowed down in the presence of a five-fold excess of enzymatically inactive CyaA-AC− molecules that promoted Ca2+ influx into cells and K+ efflux in trans. Moreover, inhibition of the endocytic uptake through inhibition of ATP re-synthesis strongly enhanced the capacity of native Bp-CyaA toxin to permeabilize and lyze cells already at as low toxin concentrations as 200 ng/ml. This points towards a more important contribution of the pore-forming activity to the overall cytotoxic action of CyaA than previously recognized [8], [9].
Enzymatically inactive but pore-forming CyaA-AC− toxoids have been extensively used over the past 18 years for delivery of AC-inserted CD8+ and CD4+ T cell epitopes from viruses, mycobacteria or tumors into the MHC class I and II-restricted antigen presentation pathways of CD11b-expressing dendritic cells [33]–[36]. As dCyaA-based vaccines for cancer immunotherapy are currently in phase I of clinical studies (www.genticel.com), the here reported insight into dCyaA endocytosis and trafficking paves the way towards deciphering of the efficacy of T cell vaccine delivery by CyaA toxoids. Moreover, an endotoxin-free CyaA-AC− (dCyaA) toxoid was recently observed to alter the expression levels of a dozen of genes involved in innate immune response signaling in bone marrow-derived macrophages [37]. This is likely due to the capacity of the toxoid to promote Ca2+ influx into cells and permeabilize cells for K+ efflux. A long-sought evidence for a role of the pore-forming capacity of CyaA in Bordetella pertussis infection has, indeed, been recently reported by Dunne and coworkers [29]. This study showed that by eliciting K+ efflux from dendritic cells, and perhaps some other CD11b-expressing phagocytes, the pore-forming activity of CyaA contributes to activation of the NALP3 inflammasome and thereby to induction of innate IL-1β response, which supports the clearance of Bordetella bacteria at later stages of infection. The results reported herein show that the capacity of CyaA to permeabilize cells for K+ efflux depends on the capacity of the toxin to promote Ca2+ influx into cells and escape the rapid macropinocytic removal from target cell membrane. This reveals a further layer of sophistication of CyaA action on host cells, thus underpinning the key role of this toxin with multiple ‘talents’ in the virulence of Bordetellae in mammals.
Mouse monoclonal antibody (MAb) anti-CD71 (clone MEM-189) was from Exbio (Vestec, Czech Republic), anti-NTAL MAb (clone NAP-08) was a generous gift from Pavla Angelisova (Institute of Molecular Genetics, Prague, Czech Republic) and anti-CyaA MAbs clone 9D4 and 3D1 were kindly provided by Erik L. Hewlett (University of Virginia School of Medicine, Charlottesville, USA). Fura-2/AM, PBFI/AM, Alexa Fluor 488, LysoTracker-red DND-99, DAPI (4′-6-diamidino-2-phenylindole) and Mowiol were purchased from Molecular Probes (Eugene, OR). Transferrin coupled to Dyomics 547, BSA-Dyomics 547 and mouse monoclonal antibody against α-tubulin (TU-01, IgG1 isotype control) were from Exbio (Vestec, Czech Republic). Ionomycin, valinomycin, nigericin, wortmannin, chlorpromazine, dynasore, 2-deoxy-D-glucose, sodium azide, chloroquine and Pluronic F-127 were purchased from Sigma (St. Louis, MO). Complete Mini protease inhibitors cocktail was purchased from Roche (Basel, Switzerland). NHS-Sulfo-LC-Biotin was purchased from Pierce (Rockford, IL, USA). The SIINFEKL peptide corresponding to the CD8+ T-cell epitope encompassing the chicken ovalbumin (OVA) residues 257–264 and to the CD4+ T-cell epitope NGKLIAYPIAVEALS peptide corresponding to the Escherichia coli MalE protein (residues 100–114 [18]), respectively, were purchased from Neosystem (Strasbourg, France).
Construction of dCyaA was described earlier [18]. dCyaA-OVA258–276, harboring the VQLTGLEQLESIINFEKLTEWTSS NVMEERKIKVYLPRIVH peptide from hen egg ovalbumin protein (OVA, residues 249–284) and carrying the MHC Class II sequence IINFEKLTEWTSSNVMEER (OVA258–276) was constructed according to the standard protocols by insertion of a corresponding double-stranded synthetic oligonucleotide between codons 107 and 108 of the cyaA gene (Holubova et al., Infect. Immun. doi:10.1128/IAI.05711-11, published online ahead of print on 3 January 2012). The corresponding CyaA protein variants, harboring the E570K and E581P substitutions, dCyaA-KP and dCyaA-OVA258–276-KP were constructed by recombination with previously described plasmid constructs [18], [38]. Toxoid constructs were genetically detoxified by insertion of the dipeptide sequence GlySer between residues 188 and 189, thus disrupting the ATP binding site of the enzyme [39]. CyaA derivatives were produced using E. coli strain XL1-Blue (Stratagene, La Jolla, CA) in the presence of CyaC acyltransferase produced from the same plasmid and were purified by chromatography on DEAE-Sepharose and Phenyl-Sepharose as described earlier [40]. Enzymatically active CyaA (rEc-CyaA or CyaA) produced in E. coli XL1-Blue and Bp-CyaA produced in B. pertussis strain 18323/ pHSP9 [28] were purified from urea extracts by combination of chromatography on DEAE-Sepharose and calmodulin agarose. All experiments were repeated with proteins from at least two independent preparations.
J774A.1 macrophages (3×105) were incubated in D-MEM with 200 ng/ml of CyaA-biotin for 5 min at 37°C, prior to removal of unbound toxin by three washes in cold D-MEM medium. Cells were lyzed with 0.1% Triton X-100 for determination of cell-bound AC activity. Toxin-induced lysis of J774A.1 cells was determined using the CytoTox 96 kit assay (Promega, Madison, USA) as the amount of lactate dehydrogenase released from 2×105 cells in 2 hours of incubation with CyaA at 37°C in D-MEM. 105 J774A.1 cells were incubated with different concentrations of the CyaA derived constructs for 30 minutes in D-MEM medium without FCS (fetal calf serum, Life Technologies, Gaithersburg, USA). The reaction was stopped by addition of 0.2% Tween-20 in 50 mM HCl and samples were boiled for 15 min to denature cellular proteins. The samples were neutralized by addition of 150 mM unbuffered imidazol and concentration of cAMP was determined by a competition imunoassay performed as previously described [8].
Toxoids were labeled with the amine-reactive Alexa Fluor 488 or Dyomics 647 dyes upon binding to Phenyl-Sepharose during the last purification step in 0.1 M NaHCO3 pH 8.3 at room temperature for 1 hour. Unreacted dye was washed-out from the resin with 50 mM Tris-HCl buffer (pH 8.0) and labeled toxoids were eluted from Phenyl-Sepharose in TUE buffer (Tris 50 mM, 8M Urea, 2 mM EDTA, pH 8.0). The molar ratio of protein∶dye was approximately 1∶4 for all toxoid preparations. On-column biotinylation of rEc-CyaA toxin was performed after the DEAE-Sepharose purification step using NHS-Sulfo-LC-Biotin (Pierce, Rockford, IL, USA) to reach a biotin∶toxin molar ratio of 20∶1. Biotin coupling was at room temperature and was stopped after 40 min by washing of the resin with 15 ml of 50 mM Tris-HCl, pH 8.0, and then extensively with PBS (phosphate-buffered saline). Purified biotinylated toxin was then eluted with 50 mM HEPES, 8 M urea, and 2 mM EDTA.
J774A.1 cells (ATCC TIB 67) were maintained in RPMI 1640 medium containing 10% FCS and antibiotic/antimycotic solution (0.1 mg/ml streptomycin, 1000 U/ml penicillin and 0.25 µg/ml amphotericin (Sigma, St. Louis, MO)). For all experiments the RPMI 1640 medium used for cell cultivation was replaced by Dulbecco's modified Eagle's medium (D-MEM) containing 1.9 mM Ca2+ without FCS, to avoid uncontrollable chelation of calcium ions by the phosphate ions contained in RPMI 1640 medium, as calcium is required for CyaA activity. Bone marrow derived dendritic cells (BMDCs) from MHC Class II/EGFP knock-in mouse [41] were flushed from the femur bone marrow cavity with PBS/2.5% FCS, plated at approx. 106 cells/well in 2 ml of DMEM/25 mM HEPES/10% FCS without phenol red but containing 5 ng/ml GM-CSF (Sigma, St. Louis, MO). Cells were cultured on 25-mm circular cover slips, with media replacement every second day. Non-adherent cells were removed by gentle washing on days 2 and 4. BMDCs from conventional 6- to 8-week-old female C57BL/6 mice were obtained from femoral and tibial bones and cells were cultured in RPMI 1640 medium supplemented with 10% FCS, 20 ng/ml GM-CSF and antibiotic/antimycotic solution for 7 days, as previously described by [42].
RAW 264.7 (ATCC TIB 71) cells (5×104) grown on coverslips were maintained in RPMI medium supplemented with 10% FCS and transfected with pEGFP constructs bearing cDNA encoding Eps-15 (DIII) or Rab-7 (kind gift of J. Forstova, Charles University, Prague) using a FuGENE-6 transfection reagent (Roche).
Detergent resistant membranes (coalesced lipid rafts) were separated by flotation in discontinuous sucrose density gradients. J774A.1 cells (2×107) were washed with prewarmed DMEM and incubated with 500 ng of CyaA-derived proteins at 37°C for 10 min. Cells were washed with ice-cold PBS, scraped from the Petri dish and extracted at 4°C in 200 µl of TBS buffer (20 mM Tris-HCl, pH 7.5, 150 mM NaCl) containing 1% Triton X-100, 1 mM EDTA, 10 mM NaF and a Complete Mini protease inhibitor cocktail (Roche) for 60 min. The lyzate was clarified by centrifugation at 250× g for 5 min and the post-nuclear supernatant was mixed with equal volume of 90% sucrose in TBS. The suspension was placed at the bottom of a centrifuge tube and overlaid with 2.5 ml of 30% sucrose and 1.5 ml of 5% sucrose in TBS. Buoyant density centrifugation was performed at 150,000× g in a Beckman SW60Ti rotor for 16 h at 4°C. Fractions of 0.5 ml were removed from the top of the gradient. For immunodetection, the separated proteins were transferred onto Immobilon-P membrane, blocked with 5% non-fat milk powder in TBST buffer (20 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.05% Tween-20) and probed with the indicated mouse monoclonal antibody. CyaA toxoids were recognized in Western blots by the 9D4 antibody binding to the C-terminal RTX repeats of CyaA. The signal was developed using a secondary peroxidase-conjugated sheep anti-mouse IgG and chemiluminescent detection system (SuperSignal West Femto Maximum Sensitivity Substrate chemiluminescence reagent kit, Pierce, Rockford, IL).
J774A.1 mouse monocytes were grown on glass coverslips to subconfluence in the absence of the pH indicator (to avoid cellular autofluorescence). Cells were incubated with Alexa Fluor 488-labeled toxoid variants in serum free D-MEM medium at 37°C in the presence of 10 µg/ml transferrin-Dyomics 547 or in the presence of 50 µg/ml BSA-Dyomics 547 (Exbio, Czech Republic). For in vivo imaging, J774A.1 cells grown on glass bottom microwell dishes (MatTek, USA) were incubated in the presence of labeled toxoids at 37°C in HBSS medium alone, or in the presence of inhibitors. BMDCs from day 5 were incubated for 2 hours with 1 µg/ml of Dyomics-647-labeled toxoid variants at 37°C in DMEM without serum. For simultaneous visualization of early and recycling endocytic compartments, DCs were co-incubated with 25 µg/ml transferrin-Dyomics-547 for the last 30 minutes. Cells were fixed (4% PFA in PBS) and mounted in Mowiol. Images were captured using a Leica confocal microscope TCS SP2 (Wetzlar, Germany) or a CellR Imaging Station (Olympus, Hamburg, Germany) based on Olympus IX 81 fluorescence microscope.
For colocalization analysis the 3D stack of desired colour channels was acquired comprising most of the cell volume (usually 10–15 planes in the z-axis). Analysis was performed using a macro in WCIF ImageJ software. Special care was taken of the pixel shift between individual colour channels (calibration with fluorescent beads). The threshold levels for each image and channel was found using “Huang dark” method. Individual cells were selected (as ROI) and colocalization analysis was performed using an ImageJ plug-in (http://www.uhnresearch.ca /facilities/wcif/imagej/colour_analysis.htm). Pearson's correlation coefficients for each image (individual cell recorded in all z-axis planes) were calculated for the given channels. Usually about 60 cells were analyzed and average Pearson's correlation coefficient (P.c.c.) of all z-axis planes ± standard deviation was calculated. A value of 1 represents perfect correlation, zero represents random localization.
For this purpose a script based on WCIF ImageJ software was used (see Figure S2 for detailed description).
Calcium influx into cells was measured as previously described [14]. Briefly, the J774A.1 cells were loaded with 3 µM Fura-2/AM (Molecular Probes) at 25°C for 30 min and time course of calcium entry into cells after addition of CyaA-derived proteins was determined as ratio of fluorescence excited at 340/380 nm. Fluorescence measurement of cytosolic K+ was performed as described before [6]. Briefly, J774A.1 cells were loaded with 9.5 µM PBFI/AM (Molecular Probes) for 30 min at 25°C in the presence of 0,05% (w/w) Pluronic F-127 (Sigma-Aldrich) in the dark. Fluorescence intensity of PBFI (excitation wavelength 340, emission wavelengths 450 and 510 nm) was recorded, ratio of these intensities are shown in the graphs. Calibration experiments were performed in solutions containing 50 mM HEPES (pH 7.2), with varying concentrations of potassium acetate (10, 30, 60, or 140 mM) and sodium acetate (5, 85, 115, or 135 mM). Cellular plasma membrane was permeabilized for potassium ions and protons by valinomycin and nigericin (3 µM; Sigma-Aldrich) for 30 min.
The H-2b-restricted T cell hybridoma CRMC3, recognizing the NGKLIAYPIAVEALS epitope of the MalE protein from E. coli (MalE100–114, abbreviated MalE), and the I-Ab-restricted T cell hybridoma MF2.2D9, recognizing the IINFEKLTEWTSSNVMEER epitope of ovalbumin (OVA258–276), respectively, were used. BMDC (105 cells/well) were pulsed with proteins or peptides for 4–5 hours, followed by medium disposal and cultivation with CRMC3 hybridoma (105 cells/well) for 18 hours or the MF2.2D9 hybridoma (5×104) for 14 hours, respectively. Prior to CRMC3 hybridoma addition the BMDCs were washed with PBS. After cell co-incubation, cultures were frozen for at least 2 hours at −80°C. T cell stimulation was monitored by determination of IL-2 amounts released into culture supernatants using two alternative methods. CRMC3 culture supernatants (100 µl) were added to cultures of the IL-2-dependent CTL-L cell line (104 cells/well, final volume 200 µl) for 48 hours, followed by addition of [3H]thymidine (50 µCi/ml; Perkin Elmer, Courtabeuf, France). Cells were harvested 6 hours later using an automated cell harvester (Molecular Devices, Lier, Norway) and the incorporated thymidine was determined by scintillation counting. The concentration of IL-2 released into MF2.2D9 cell culture supernatants was determined using a sandwich ELISA with noncompeting pairs of anti–IL-2 mAbs (JES6-1A12 and biotinylated JES6-5H4, both from BD Pharmingen, San Diego, CA, USA.
|
10.1371/journal.pntd.0006011 | Clinico-pathological features of erythema nodosum leprosum: A case-control study at ALERT hospital, Ethiopia | Leprosy reactions are a significant cause of morbidity in leprosy population. Erythema nodosum leprosum (ENL) is an immunological complication affecting approximately 50% of patients with lepromatous leprosy (LL) and 10% of borderline lepromatous (BL) leprosy. ENL is associated with clinical features such as skin lesions, neuritis, arthritis, dactylitis, eye inflammation, osteitis, orchitis, lymphadenitis and nephritis. ENL is treated mainly with corticosteroids and corticosteroids are often required for extended periods of time which may lead to serious adverse effects. High mortality rate and increased morbidity associated with corticosteroid treatment of ENL has been reported. For improved and evidence-based treatment of ENL, documenting the systems affected by ENL is important. We report here the clinical features of ENL in a cohort of patients with acute ENL who were recruited for a clinico-pathological study before and after prednisolone treatment.
A case–control study was performed at ALERT hospital, Ethiopia. Forty-six LL patients with ENL and 31 non-reactional LL matched controls were enrolled to the study and followed for 28 weeks. Clinical features were systematically documented at three visits (before, during and after predinsolone treatment of ENL cases) using a specifically designed form. Skin biopsy samples were obtained from each patient before and after treatment and used for histopathological investigations to supplement the clinical data.
Pain was the most common symptom reported (98%) by patients with ENL. Eighty percent of them had reported skin pain and more than 70% had nerve and joint pain at enrolment. About 40% of the patients developed chronic ENL. Most individuals 95.7% had nodular skin lesions. Over half of patients with ENL had old nerve function impairment (NFI) while 13% had new NFI at enrolment. Facial and limb oedema were present in 60% patients. Regarding pathological findings before treatment, dermal neutrophilic infiltration was noted in 58.8% of patients with ENL compared to 14.3% in LL controls. Only 14.7% patients with ENL had evidence of vasculitis at enrolment.
In our study, painful nodular skin lesions were present in all ENL patients. Only 58% patients had dermal polymorphonuclear cell infiltration showing that not all clinically confirmed ENL cases have neutrophilic infiltration in lesions. Very few patients had histological evidence of vasculitis. Many patients developed chronic ENL and these patients require inpatient corticosteroid treatment for extended periods which challenges the health service facility in resource poor settings, as well as the patient’s quality of life.
| Leprosy reactions (Type 1 and 2) are important causes of nerve damage and illness. Erythema Nodosum Leprosum (ENL) also called type 2 reactions is a severe multisystem immune-mediated complication of borderline and lepromatous leprosy. ENL causes high morbidity and mortality and usually requires urgent medical attention. ENL can occur before, during, or after completion of MDT. The diagnosis and treatment of ENL is largely based on clinical symptoms. However, the clinical symptoms are heterogeneous and may vary from patient to patient. Although thalidomide is an effective drug for ENL treatment, it is not available in many leprosy endemic countries including Ethiopia. In spite of its adverse effects, in many endemic countries corticosteroid is the only available drug for ENL treatment, usually being used for prolonged periods. Therefore, alternative and effective drugs are required to reduce the burden of ENL. To establish which drugs will be effective in the treatment of ENL it is necessary to have a clear picture of the clinical and histological features of the disease. We systematically documented these features of ENL and compared them with matched non-reactional LL controls. Thus, the findings will help to develop better ENL diagnosis and treatment options.
| Leprosy is a disease caused by Mycobacterium leprae, an intracellular acid-fast bacillus[1]. It mainly infects the skin and peripheral nerves[2]. The disease manifests with a spectrum of clinical pictures ranging from the localized tuberculoid leprosy (TT) to the generalized lepromatous leprosy (LL) types forming the two poles of the five point spectrum [3].
Erythema nodosum leprosum (ENL) is an immune-mediated inflammatory complication affecting about 50% of patients with lepromatous leprosy (LL) and 10% of borderline lepromatous (BL) patients [4–6]. ENL can occur before, during or after successful completion multi-drug therapy (MDT). The onset of ENL is acute, but it may pass into a chronic phase and can be recurrent [7].
ENL affects multiple organs and causes systemic illness [8].It is clinically characterized by the occurrence of crops of tender skin lesions [9]. Histologically, neutrophils are considered the hall mark of ENL[10]. The histology of ENL lesions shows an intense perivascular infiltrate of neutrophils throughout the dermis and subcutis [10]. However, not all clinically confirmed ENL cases have neutrophilic infiltration in lesions[11].
The underlying immunologic mechanisms of ENL have not been fully understood. The hypothesis of ENL as an immune-complex mediated disease proposed in the 1960s has yet to be supported by definitive evidence. Granular deposits of immunoglobulin and complements in the dermis of ENL lesion has been found by using direct immunofluorescence techniques which were absent in non-reactional LL lesions [12–14]. However, some investigators have reported the presence of immunoglobulin and complement deposits in ENL lesions as well as in LL lesions [15–17].
The contribution of cell-mediated immunity in the pathogenesis of the disease has been suggested but not supported by definitive evidence[18]. Several studies [19–23] have reported increased percentage of CD4+ T-cells and reduced CD8+ T-cells with an increased CD4+/CD8+ ratio in patients with ENL compared to patients with non-reactional lepromatous leprosy. Other studies have however, also reported a reduced CD4+/CD8+ ratio and increased percentage of CD8+ T-cells in patients with ENL compared to patients with LL [24].
The inflammatory condition of ENL may cause significant morbidity and mortality if it is not treated on time.[25]. In Ethiopia, patients with ENL are treated with corticosteroids for several months or years. Many patients require high doses of prednisone to control inflammation which could lead to complications. A significant proportion of deaths associated with long-term use of these drugs has been reported [25].
Having awareness of the diverse clinical features of ENL is useful for the accurate diagnosis and successful management of the disease. However, there are only few prospective studies describing the clinical features and there relative frequencies in ENL. A cross-sectional international multicentre study of the clinical features of ENL including 292 patients in 7 countries has reported that a significant number of patients had extra-cutaneous pathology such as peripheral oedema, large joint arthritis, lymphadenitis, and orchitis [9].
We set up a case control follow up study to investigate the clinico-pathological features of ENL. We compared the clinical and histological features in patients with ENL reactions to matched uncomplicated non-reactional LL patient controls before and after prednisolone treatment of ENL cases. ENL patients have diverse clinical manifestations. Therefore, prospective documentation of the clinical manifestations of patients with ENL is useful for accurate diagnosis of ENL. Unlike previous cross-sectional studies, in the present study we obtained clinical data and clinical sample (skin biopsy) from cases (ENL) and controls (LL) before, during and after treatment. The controls were matched with cases with respect to age, sex and duration of leprosy diagnosis. Hence, the present findings are more informative and show the dynamics of clinical features of ENL before and after treatment.
Informed written consent for blood and skin biopsies were obtained from patients following approval of the study by the Institutional Ethical Committee of London School of Hygiene and Tropical Medicine, UK, (#6391), AHRI/ALERT Ethics review committee, Ethiopia (P032/12) and the National Research Ethics Review Committee, Ethiopia (#310/450/06).
A case control study was conducted between December, 2013 and October, 2015 at All Africa Leprosy and, Tuberculosis Rehabilitation and Training Centre (ALERT) Hospital, Ethiopia. This is the main leprosy specialized hospital in Ethiopia. Hence, it is an ideal hospital to obtain referred leprosy patients from all regions in the country.
Children below 18 years old, adults above 65 years old, pregnant and lactating mothers, patients with other clinical forms of leprosy (TT, BT, BB, BL and T1R) were excluded from the study. Forty-six untreated patients with ENL and 31 LL controls were enrolled into the study and followed for 28 weeks. The controls were age and sex matched with cases (ENL).
ENL was clinically diagnosed when a patient with LL leprosy had painful crops of tender cutaneous erythematous skin lesions [5]. Lepromatous leprosy was clinically diagnosed when a patient had widely disseminated nodular lesions with ill-defined borders and BI above 2 [7].
New ENL was defined as the occurrence of ENL for the first time in a patient with LL. The nature of ENL was defined as acute for a single episode lasting less than 24 weeks while on corticosteroids treatment, recurrent if a patient experienced a second or subsequent episode of ENL occurring 28 days or more after stopping treatment for ENL and chronic if occurring for 24 weeks or more during which a patient required ENL treatment either continuously or where any treatment free period had been 27 days or less [7].
Clinical data were collected using a standard form that had been developed by the Erythema Nodosum Leprosum International STudy (ENLIST) group. Demographic, clinical and laboratory data were recorded including evidence of any nerve function impairment (NFI) using voluntary muscle and Semmes-Weinstein monofilament sensory testing. Nerve function impairment (NFI) was defined as clinically detectable impairment of sensory or motor nerve function. New NFI was defined as NFI present for less than six months[26]. The bacterial Index (BI) at leprosy diagnosis was obtained for all recruited patients. BI at ENL reaction was also obtained at enrolment.
Six millimetre skin biopsies were obtained from each ENL case before and on 24th week after prednisolone treatment of ENL cases. Similarly, 6mm biopsy was obtained during enrolment and on the 24th week of recruitment from matched non-reactional LL controls. Biopsies were taken from the active erythematous new skin lesions in all patients with ENL and from nodular LL lesions. Biopsies were obtained from the same area for cases and control. Biopsies were stored in 10% formalin until processed. Sections were stained with Haematoxylin and Eosin stain and examined by two histopathologists independently. The pathologists were not aware of the clinical diagnosis. Bacterial index (BI) was obtained for each patient as a routine investigation.
When a polymorphonuclear neutrophilic infiltrate on the background of a macrophage granuloma accompanied by oedema and often with evidence of vasculitis and/or panniculitis was seen, the sample was classified as ENL. The presence of macrophage and foam cell collections with numerous bacilli interspersed with sparse number of lymphocytes in histological sections was defined as LL [27].
The anonymised clinical and Histopathology data were entered into an Excel database and analysed using Stata 14 version 2 and SPSS 23 version 1 Statistical Software. Depending on the nature of the variable and the normality of the data, either parametric or non-parametric analysis was used. Categorical variables were analysed by non-parametric methods and normally distributed numerical variables with parametric methods. Whenever mean is used for comparison, data presentation has followed the form of mean ± standard error of the mean (SE). The level for statistical significance was set at α = 5% with 95% confidence interval.
Clinical data were obtained on 77 patients (46 LL patients with ENL reactions and 31 non-reactional LL patients) at recruitment (Table 1). The male to female ratio was 2:1 with a median age of 27.5 [range: 18–56] years in patients with ENL and nearly 3:1 with a median age of 25.0 [range: 18–60] years in patients with non-reactional LL controls. The age range of females in both groups was relatively narrow (18–35 years) compared to males (18–60). More than half of the patients with ENL had previously been treated with MDT. Half of the patients with ENL had acute ENL at the time of enrolment with mean BI 3.9 ±0.205 SE (standard error). Recurrent ENL cases had the highest mean BI (4.9 ±0.409 SE) at leprosy diagnosis whereas acute and chronic cases had comparable mean bacterial index (BI) (Table 1).
Pain was the most common symptom reported by patients with ENL. Ninety-eight percent of the patients with ENL had pain at enrolment. About 80% of the patients with ENL had reported skin pain and more than 70% had nerve and joint pain during enrolment. Other pain sites reported include bone, digits, eyes, muscles, lymph nodes and testes (Fig 1).
Fever was reported by 31 (71.7%) patients with ENL. Sixteen (34.8%) patients with ENL reported depression and 47.8% nasal stuffiness. Other reported symptoms included peripheral oedema, insomnia, anorexia, weight loss, joint swelling and malaise (Fig 2).
About 96% individuals had nodular cutaneous lesions, about two-third had subcutaneous nodules and a quarter of patients had scar. While one-third of the patients had ulcerated lesions, only 4% had necrotic lesions. Eight patients (17.3%) had vesicles, bullae or pustular lesions (Fig 3).
In most patients with ENL (73.9%), the number of skin lesions recorded at the time of enrolment was between 11 to 50. Few patients had five or less skin lesions. Almost all patients (97.8%) had skin lesions on the upper limbs. Many patients also had skin lesions on the lower limbs (95.7%) or on the head and neck (63.0%). Half of the patients reported reduced nerve sensation. Paraesthesia and hyperaesthesia were reported by 13% and 23.9% of patients respectively (Table 2).
More than half (52.2%) of patients with ENL had old nerve function impairment (NFI) while 13% had new NFI at the time of enrolment. Facial oedema was reported in 56.5% of the patients with ENL and nearly half (47.8%) of the patients had oedema on their lower limbs. Other organs involved in the patients with ENL were small joint arthritis (28.3%), large joint arthritis (15.2%), conjunctivitis (4.3%), lagophthalmos (2.2%), scleritis (8.7%), lymph node (15.2%) and dactylitis (2.2%) (Table 2).
Paraffin- embedded sections of skin biopsy samples from ENL and LL lesions were examined by a histopathologist (Fig 4). Neutrophils infiltration was noted more ENL lesions (58.9%) than LL lesions (14.3%) before treatment (P = 0.004). Lymphocytes infiltration was recorded in all ENL and LL lesions. Foamy histiocytes were more frequently seen in LL lesions (95.3%) than in ENL lesions (85.3%) although the difference was not statistically significant at enrolment. After 24 weeks treatment of ENL, the percentage of foamy histiocytes was significantly decreased in ENL cases (42.2%) compared to LL cases (85.7%) (p = 0.001). Panniculitis was diagnosed in 62.5% of lesions from patients with ENL reactions. After 24 weeks of ENL treatment, neutrophils infiltration was noted in 5 biopsies from patients with ENL reactions, lymphocytes infiltration was seen in 20 biopsies of patients with ENL (Table 3).
The number of male patients with ENL recruited to the study was twice the number of female patients and similar to a five-year retrospective data (2008–2013) which showed the number of male to female ratio to be 1.7:1 [7]. In our study, the median age for male and female patients with ENL was 28.0 and 26.7 years respectively. Both male and female patients with ENL were relatively older than the LL patient controls (median age: male = 26 years, female = 21 years). The slight difference in median age between the two groups could be explained by natural the course of the disease. Patients usually develop ENL reaction after having either LL or BL clinical forms for some time. Interestingly, the age range of females in both groups was relatively narrow (18–35 years) compared to males (18–60 years) indicating that either younger females are more likely to have access to health institutions for various reasons than older females in low-income countries where health facilities are relatively inadequate[28] or ENL is relatively common among younger females of child bearing age due to various biological reasons [29–31].
Our data confirm that a significant proportion of cases had chronic ENL (39%). This implies that these patients require, in our setting, corticosteroid treatment for extended periods, often at high doses… But high doses of corticosteroids do not always control the inflammation and also pose life-threatening risks for patients [9, 32, 33]. Chronic ENL cases are a burden to referral hospitals in these resource poor settings. as well as to their communities. A study in rural India has shown that families with at least one ENL case incur loss of more than 40% of total household income compared to families without ENL case due to out of pocket expenditure for treatment-seeking (direct cost) and loss of income resulting from reduced productivity (earning potential) of household members (indirect cost). This implies that households affected by ENL face significant economic burden and are at risk of being pushed further into poverty [34].
In this study, several cutaneous manifestations of ENL were documented highlighting the heterogeneous nature of ENL clinical manifestation. Pain was was a symptom reported by 98% of the patients. Most patients had skin pain (80.4%), nerve pain (73.9%), joint pain (71.7%) and bone pain (69.2%). The most frequent site of pain due to ENL in our study was the skin which is explained by the fact that 95% of patients with ENL had skin lesions. Our finding is in agreement with a previous report [7]. Bone pain was reported in two-third of our study patients which is higher than the previous report [7]. The difference between the two studies is likely due to the retrospective nature of the previous study which was not reliant on case note recording unlike the current study.
The nerve function impairment (NFI) was reported in 65% of our study patients, which was higher than the 51.3% NFI in six countries as reported by Walker et al [35]. Among the 65% of patients reporting NFI, 80% of them had old NFI. This highlights the prevalence of NFI in patients with ENL the high risk of developing permanent disability. A study by Santos Santos, de Mendonça Neto [36], in northern Brazil had identified NFI and leprosy reactions as the main risk factors associated with the development of disability in leprosy patients. The same authors reported that NFI was strongly associated with physical disability in children under 15 [37]. In our study, 50% of patients with ENL had WHO disability grade-1 (G1D) while 4.3% had Grade- 2 disability (G2D). The proportion of grade 2 disability was lower than the national figure (10.2%) in 2014 [38].
Histopathologically, neutrophil infiltration was noted in 58.8% of patients with ENL compared to 14.3% in LL controls before treatment. This confirms that a neutrophilic infiltration cannot be used as the sole histological marker for ENL The absence of neutrophil infiltration has been reported in 36% of ENL skin lesions in Pakistani patients who had classical signs and symptoms of ENL[11]. Similarly, a cross-sectional study on the histological features of leprosy reactions in Indian patients by Sarita, Muhammed [39] showed that 43% ENL skin lesions did not have histological evidence of neutrophil infiltration. Our findings agree with these two studies. Previous studies by others [40–42], reported finding neutrophil infiltration in all ENL lesions. The varying reports of neutrophil infiltration in ENL lesions could be attributed to several factors. If the definition of ENL includes the presence of neutrophils in the case definition then all cases will have it, as did Aldhe et al who investigated the presence of cellular neutrophil infiltration on histologically confirmed ENL cases [42]. Delay between the onset of reaction and the timing of obtaining the biopsy in those without neutrophilic infiltrate, as dermal oedema may be missed in older reactional lesions could cause these differences. Discordance between pathologists and standard operating procedures (SOPs) of slide preparations are also potential areas that should be further investigated to evaluate their impact on the findings of neutrophil infiltration in tissue sections. Previous reports suggested vasculitis as part of ENL reaction commonly seen in Indian patients [43], only 5(14.7%) of our patients had evidence of vasculitis. Similar observations had been made by Sarita et al and Adhe et al [39, 44].
Inclusion of a large number of patients with ENL and LL controls was one of the strengths of this study. The other strength of the study had been that clinical data were obtained from each patient three times unlike the previous cross-sectional studies. A weakness of the study is that there may have been biased recruitment because of the need to have good follow–up of patients.
In conclusion, we have shown that skin, nerve and joint pain are the most common clinical symptoms reported in patients with ENL. These clinical conditions are usually difficult to manage with corticosteroids at referral Hospitals. Most of our patients with ENL then developed chronic ENL and these patients require in patient corticosteroid treatment for extended periods which challenges the health service facility in resource poor settings. More than half of the patients with ENL had old NFI which indicates that these patients are at a higher risk of developing permanent disability. Hence, better attention to care and NFI needed in these patients.
|
10.1371/journal.pcbi.1000892 | Frequency-Dependent Selection Predicts Patterns of Radiations and Biodiversity | Most empirical studies support a decline in speciation rates through time, although evidence for constant speciation rates also exists. Declining rates have been explained by invoking pre-existing niches, whereas constant rates have been attributed to non-adaptive processes such as sexual selection and mutation. Trends in speciation rate and the processes underlying it remain unclear, representing a critical information gap in understanding patterns of global diversity. Here we show that the temporal trend in the speciation rate can also be explained by frequency-dependent selection. We construct a frequency-dependent and DNA sequence-based model of speciation. We compare our model to empirical diversity patterns observed for cichlid fish and Darwin's finches, two classic systems for which speciation rates and richness data exist. Negative frequency-dependent selection predicts well both the declining speciation rate found in cichlid fish and explains their species richness. For groups like the Darwin's finches, in which speciation rates are constant and diversity is lower, speciation rate is better explained by a model without frequency-dependent selection. Our analysis shows that differences in diversity may be driven by incipient species abundance with frequency-dependent selection. Our results demonstrate that genetic-distance-based speciation and frequency-dependent selection are sufficient to explain the high diversity observed in natural systems and, importantly, predict decay through time in speciation rate in the absence of pre-existing niches.
| Ecological opportunity, or filling a pre-existing unoccupied adaptive zone, is considered the dominant mechanism explaining the initial explosion of diversity. Although this type of niche filling can explain rates of diversification in some lineages, it is not sufficient for a radiation to occur. Instead of attributing the propensity to have an explosion of new species to external influences like niche availability, an alternative hypothesis can be based in frequency-dependent selection driven by the ecology in which organisms are embedded or endogenous sources mediated by gametes during fertilization. We show that genome diversification driven by higher reproductive probability of rare genotypes generates rapid initial speciation followed by a plateau with very low speciation rates, as shown by most empirical data. The absence of advantage of rare genotypes generates speciation events at constant rates. We predict decline over time and constant speciation rate in the cichlids and Darwin's finches, respectively, thus providing an alternative hypothesis for the origin of radiations and biodiversity in the absence of pre-existing niche filling. In addition to predicting observed temporal trends in diversification, our analysis also highlights new mechanistic models of evolutionary biodiversity dynamics that may become suitable to generate neutral models for testing observed patterns in speciation rates and species diversity.
| Speciation is one of the most complex phenomena in nature, yet the effects of its tempo and mode for biodiversity patterns are still controversial [1], [2]. Pre-existing niches is considered the dominant mechanism explaining the initial explosion of diversity observed in radiations [3]–[7]. In contrast, non-adaptive radiations [8], [9] driven by niche-independent mechanisms such as sexual selection, rapid range expansion across multiple barriers or the simultaneous formation of multiple geographical barriers, dispersal limitation or isolation by distance without physical barriers due to genetic incompatibilities do not predict such a temporal trend of declining speciation rates during a radiation [10]–[13].
Although ecological opportunity (the availability of an unoccupied adaptive zone) or rapid range expansion across multiple barriers can explain rates of diversification in some radiating lineages, this is not sufficient for a radiation to occur [14]–[17]. Instead of attributing the propensity to have a radiation with decaying through time speciation rates to external influences like niche availability or rapid range expansion an alternative hypothesis can be based in the genome properties evolved during the evolutionary history of organisms. We explore this hypothesis using two models, one with frequency-dependent selection and one without it. Both models involve DNA sequence-based evolution of populations via a process of sexual reproduction, assortative mating, mutation, and genetic-distance-based speciation.
The models we have analyzed in the present study are similar in spirit to previous speciation models [12], [13], [18], [19] but different in two key details: (1) no approximations of the tempo and mode of speciation incorporating sexual reproduction and frequency-dependent selection have previously been shown to explain observed patterns of decay through time of the speciation rate during a radiation without invoking pre-existing niches. Furthermore, we show that the decay through time of the speciation rate during a radiation without invoking pre-existing niches has dramatic consequences to species richness and diversity, and (2) we contrast the models with two small radiations for a broad range of parameter values: the Tilapia cichlid genus [11] and the Darwin's finches [20], two groups where assortative mating has been previously documented [21]–[24]. We note that larger radiations cannot be handled computationally. This represents a current limitation to explore broad patterns of speciation and diversity that requires further research.
We simulated the evolution of a population whose members, at the beginning, have identical genomes. The population evolves under the combined influences of sexual reproduction and mutation (Text S1). During reproduction, potential mates are identified from those whose genomes are sufficiently similar to that of the reproducing individual (). This parameter implicitly captures the effects of the accumulation of genetic incompatibilities by prezygotic or postzygotic reproductive isolation [18], [25]–[27]. A mate is chosen from this set at random. An offspring is then dispersed in the environment. This minimal form of mating called assortative mating [13], [28], [29] is sufficient for speciation at least when there is no genetic linkage [18], [19]. Genomic similarity between two individuals is defined as the proportion of identical nucleotides along the genome. The genomic similarity among individuals can be represented by an evolutionary graph in which nodes are individuals and edges connect reproductively compatible individuals [30] (Figs. 1 and 2). We identify a species as a group of organisms reproductively separated from all the others by genetic restriction on mating, but connected among themselves by the same condition. Thus, two individuals connected at least by one pathway through the evolutionary graph are considered conspecific, even if the two individuals themselves are reproductively incompatible.
We consider three main assumptions that allow us to approximate the tempo of speciation and also to identify the conditions for each of two alternative modes of speciation in the evolutionary graph: (1) Our density of individuals is one per site, and these numbers are kept constant by assuming zero-sum dynamics. Birth-death zero-sum stochastic models are equivalent to their non zero-sum counterparts at stationarity [31]; (2) Factors influencing speciation may differ between regions of the genome, and regions of the genome involved in reproductive isolation may differ between taxa and the temporal stages of the speciation process [32]. In our model, the genome of each individual is considered effectively infinite (i.e., a very large string of nucleotides, Text S1), and (3) The mate choice function explaining the viability of the offspring is given by a step-shaped function. This is the simplest representation of Dobzhansky-Muller reproductive incompatibility [33]–[35]. Functions with equal viability in a range [, ] (see Material and Methods), declining linearly and exponentially [12] give qualitatively the same results as the results presented here using the step-shaped function.
At the beginning of the simulation, all individuals are reproductively compatible, corresponding to a completely connected graph. Because of mutations that can eventually reduce genetic similarity below the threshold required for mating, the graph will lose connections as generations pass (Fig. 1). The rate at which connections are lost in the evolutionary graph, and thus the tempo of speciation, depends on the mechanisms driving genome diversification.
To explore the tempo of speciation and its implications for biodiversity patterns, we generated a second model with negative frequency-dependent selection. In this model there are not external factors creating pre-existing niches, which can be populated only by individuals of a specific genotype and can be filled up to a carrying capacity. In contrast, any rare genotype has higher fitness than common types. The reason may be natural selection driven by the ecology in which the organism is embedded (e.g., bacteria or pathogens attacking reproductive proteins of common types without altering the probability to die among individuals) [36], [37] or some form of sexual selection that lead to rare-type advantage (e.g., sexual conflict, molecular incompatibility or heterozygote advantage in sexually selected genes) [38]–[40] and have more success at mating, whereas common types are likely–but not guaranteed–to become rare. Despite potential costs of the rare types (i.e., Allee effects, mating costs, etc), experimental and theoretical studies have shown that the selective value of a given genotype is often a function of its frequency in the population [41]–[46]. In summary, frequency-dependent selection in this context is a type of sexual selection with niches not imposed from outside the system but created by rare types with greater mating success that can spread their alleles more quickly through the population. Apart from the asymmetry introduced by the different reproductive probabilities at the individual level, these two models are identical (Text S1).
With appropriate parameter values satisfying the mathematical condition required for speciation ( where is the genetic similarity matrix at equilibrium) both models can produce speciation events (i.e., sexual isolation of subpopulations in the genome space, Equation A-30 and Box 1 in Text S1).
We identified two distinct modes of speciation that can, under the right conditions, occur in the evolving graph: mutation-induced speciation and fission (Fig. 1). Mutation-induced speciation happens when a newly produced offspring is disconnected from its parents. This form of speciation requires the mutation rate to exceed some minimum value () necessary to satisfy the inequality , where is the offspring and are the parents of (Fig. 2). Because the minimum number of steps equals 1, the minimum mutation rate to have mutation-induced speciation is given by:(1)For example, if offspring become inviable once genetic divergence exceeds (i.e., ), then the minimum mutation rate needed to achieve mutation-induced speciation is . There is a second mode of speciation which is also a consequence of mutations in the evolutionary graph. We call this mode “fission” because it takes place when the death of an individual breaks a link in what was the sole genetic pathway connecting some members of a species; this gives rise to one or more new species. Because of the strict condition for mutation-induced speciation to happen, fission is the only mode of speciation in the biologically relevant portion of model parameter space (Section A3 in Text S1).
The speciation rate () in the genetic similarity matrix () has two different dynamics according to the initial minimum genetic similarity value to have fertile offspring ():(2)where is the expected mean genetic similarity in the matrix at equilibrium [18], with the population size. If , then is the rate of dropping links in the evolutionary graph that is proportional to the speciation rate for the model without frequency-dependent selection (Fig. 2 and Text S1). Fitting to the speciation rates obtained via simulation yielded least-squares regression coefficient estimates of and the slope ():(3)This approximation suggests that the long term rate of speciation is independent of population size (Section A3 and Figs. 1 and 3 in Text S1).
The models generate changes over time in the tempo of speciation, the distribution of incipient species abundance, and both the number and diversity of contemporary species. In Figs. 3 and 4, we summarize the following two key predictions for the species number through time and species richness consistent with Darwin's finches and cichlid fish.
First, we predict that whether the rate of speciation will remain constant or decline over time depends on the addition of frequency-dependent selection. Fig. 3a shows how the number of extinct and extant species varies over time. After a transient period, during which mutation introduces genetic variability into the initially identical population, the number of species increases rapidly. The two models then diverge dramatically. In the model without frequency-dependent selection, speciation rate remains constant. This pattern is consistent with the literature on whole-tree cladistic analysis [47], the record of marine invertebrate fossils from the Phanerozoic eon [48], and (over shorter time frames) observed genetic differences among North American songbirds [49]. The number of contemporary species (Fig. 3b), diversity (Inset Fig. 3b), and the abundance of the new species (Fig. 3c) are lower than in the frequency-dependent model. In the frequency-dependent case, rapid speciation is followed by a plateau with few speciation events, consistent with molecular data for several groups showing declining speciation rates through time [16], [50]–[53]. This model predicts a greater number of contemporary species, higher diversity, and a more symmetric abundance distribution of incipient species; these are all attributes of rapid radiations.
Second, frequency-dependent selection reproduces cichlid radiations in absence of pre-existing niches and the absence of frequency-dependent selection generates the Darwin's finches radiation. Fig. 4a and 4b show the best fit to the data for the number of species and speciation events through time. We predict decline over time and constant speciation rate in the cichlids and Darwin's finches with and without frequency-dependent selection, respectively (data not shown). The expected distributions of species abundance derived from those predictions depart dramatically. For the genus, the model predicts high diversity, with most species having similar abundances (inset Fig. 4a); for the Darwin's finches, the model predicts much lower species diversity, with most species being rare (insets Fig. 4b).
Most speciation studies have concluded that sympatric speciation only occurs if a stringent set of conditions is met [4], [7]. Likewise, for the models we have explored, sympatric speciation can be highly unlikely or even impossible in biologically relevant areas of parameter space (i.e., , where , Text S1). Note, however, that even though geographical barriers and dispersal limitation, and/or range expansion have played an important role in radiations, those factors do not generate decay through time in speciation rate in the absence of niche filling [10]–[13] (see also Fig. 5 in Text S1). Interestingly, the absence of frequency-dependent selection does not capture the exponential growth in number of species in the last stage of the Darwin's finches radiation. Time lag for extinctions [50], taxonomic splitting but also the increase in heterogeneity with time in the Galápagos archipelago (i.e., more islands, habitat diversity and food types) [20] are some of the factors that may hamper model predictions in this case. Nevertheless, the balance of results for both the cichlids and the Darwin's finches suggest that neutral and frequency-dependent selection mechanisms have played a role in radiating lineages.
Current biodiversity theory, from population genetics [13] to island biogeography and its extensions [54], explain species abundance patterns for many groups, but cannot predict different trends in the tempo of speciation nor their implications for radiations and diversity patterns. The models we have explored generate alternative tempo of speciation and these models can be compared with the patterns of diversity underlying classic radiations. In the context of these models, we have also determined the conditions necessary for the mutation-induced mode of speciation; if these are not met, then fission must be the only speciation mode. Finally, we have shown that frequency-dependent selection generates more symmetric and larger incipient species abundances, resulting in lower extinction rates. These results reinforce the notion that the incipient species abundance can have a dramatic impact on contemporary diversity patterns [54], and suggest that both the tempo and mode of speciation themselves have a large effect on current community dynamics.
Alternative models of speciation that incorporate additional molecular or ecological components exist (i.e., spatial heterogeneity and dispersal limitation [19], [55], recombination rate, insertions and deletions [56] and the explicit mechanisms that cause genetic incompatibilities [57], [58]); however, it is not yet possible to evaluate those models with speciation rates and diversity data. Fitting models with a large number of parameters remains a challenge for the future - we have shown that a quasi-likelihood method offer a powerful approach.
In summary, the particular mechanisms underlying the dynamics of the evolutionary graph affect the tempo of speciation and diversity, but we nevertheless find theoretical distributions in agreement with the observed patterns of radiations and biodiversity for diverse taxa. Underlying the result are two simple models of a sexually reproducing population with and without frequency-dependent selection and with mating restrictions that depend on genetic distance. By examining these models under different parameter combinations and confronting them with data, we conclude that the properties of genomes during lineage diversification may influence patterns of radiations and biodiversity and the pre-existing environmental niches are not necessary for radiations to occur.
Our simulation is a stochastic, individual-based, zero-sum birth and death model of a sexual population with overlapping generations and age-independent birth and death rates. For the simulations reported in the paper, we considered haploid and hermaphroditic individuals where only one individual can exist in each site. Genomes consist of an infinite string of nucleotides and the genomic similarity between two individuals is defined as the proportion of identical nucleotides along the genome. Reproduction starts with a randomly selected individual looking for a mate among all the sufficiently similar individuals. To qualify, an individual must have a genetic similarity greater than the minimum value required for fertile offspring. From all such potential mates, we select the second parent at random. In the frequency-dependent selection model, individuals with few connections, and therefore with more rare alleles, have more success at mating and their alleles spread quickly through the population.
Mating produces a haploid offspring that differs from both parents following free recombination and mutation (Text S1). Each nucleotide is inherited from one of the parents with the same probability. The results reported here are for asynchronous mating. Synchronous mating gave similar results, although speciation times were typically longer. According to tests of multiple model variants in the model without frequency-dependent selection, including parameter variation, self-incompatibility (i.e., by adding a to limit the reproduction of excessively similar individuals, Fig. 5a in Text S1), and mating and dispersal limited to adjacent patches (i.e., 8-patch Moore neighborhood) with and without a wrapped torus (Fig. 5b in Text S1), our results apply quite generally, with the key required properties to generate declining through time speciation rates being the limitations on genetic distance associated with mating and the frequency-dependent selection mechanism.
Results for Fig. 3 are obtained by time-averaging over replicates lasting generations each. Given individuals in the initial population, a generation is an update of time steps. Parameter variation does not affect the overall behavior.
Results for Fig. 4 are obtained after replicates for each parameter combination lasting generations each. We sampled the transients (each generation) and the steady state at the end of each replicate for the species through time and species abundance. We have explored parameter combinations in the range , , and community size, that satisfy the mathematical condition required for speciation (, equation A-30 and Box 1 in Text S1). Our results apply quite generally in a broad range of community size (Fig. 3 in Text S1) and speciation rates (Fig. 4 in Text S1).
The fit to the number of species and speciation events through time was done following these steps: 1) Normalize time for observed data and each simulation from the first speciation event to present time within the range [0, 1], 2) From each possible interval, starting with the size of the data until the size of the output in each simulation ( generations with increments of 1 generation at each time), we generated the sequence of speciation times that minimizes the difference with the observed speciation times, and 3) Identify the best fit as the one that minimizes the sum of the absolute values of the misfits:(4)where is defined as , i.e., in terms of the misfit between observed and simulated species richness, , and the misfit in the timing of speciation events . and are our model parameters. Our search is performed for a broad range of plausible empirical values for and constant and satisfying (Text S1).
If our errors per data point are a random variable following the exponential distribution, , and, assuming error independence, our measure of misfit is the model negative log-likelihood [59]. Confidence intervals have been calculated by taking the percentiles and from the distributions of values of different model replicates. Model replicates were generated with the best parameter estimates for and along with a family of pairs within 2 log-likelihood units away from the minimum [60] (Fig. 4 in Text S1).
|
10.1371/journal.pcbi.1000810 | Polarizable Water Model for the Coarse-Grained MARTINI Force Field | Coarse-grained (CG) simulations have become an essential tool to study a large variety of biomolecular processes, exploring temporal and spatial scales inaccessible to traditional models of atomistic resolution. One of the major simplifications of CG models is the representation of the solvent, which is either implicit or modeled explicitly as a van der Waals particle. The effect of polarization, and thus a proper screening of interactions depending on the local environment, is absent. Given the important role of water as a ubiquitous solvent in biological systems, its treatment is crucial to the properties derived from simulation studies. Here, we parameterize a polarizable coarse-grained water model to be used in combination with the CG MARTINI force field. Using a three-bead model to represent four water molecules, we show that the orientational polarizability of real water can be effectively accounted for. This has the consequence that the dielectric screening of bulk water is reproduced. At the same time, we parameterized our new water model such that bulk water density and oil/water partitioning data remain at the same level of accuracy as for the standard MARTINI force field. We apply the new model to two cases for which current CG force fields are inadequate. First, we address the transport of ions across a lipid membrane. The computed potential of mean force shows that the ions now naturally feel the change in dielectric medium when moving from the high dielectric aqueous phase toward the low dielectric membrane interior. In the second application we consider the electroporation process of both an oil slab and a lipid bilayer. The electrostatic field drives the formation of water filled pores in both cases, following a similar mechanism as seen with atomistically detailed models.
| Many biomolecular processes involve charged species moving between regions of high polarity, such as the water phase, and regions of lower polarity, such as the lipid membrane. Due to the change in electrostatic screening between these two environments, the strength of the interactions between the moving charge and the surrounding molecules also changes. This has important consequences for the way biological activity is controlled. To help understand the forces driving the movement of biomolecules, we developed a computational model which is capable of describing these processes at near-atomic detail. To do so efficiently, we use a coarse-grained description of the molecules, in which some of the atomistic detail is averaged out. To capture the inhomogeneous nature of the dielectric response, we re-introduce some detail in the water model; the new model effectively mimics the orientational polarizability of real water molecules, and screens electrostatic interactions realistically. This enables the study of a number of important biological processes that were hitherto considered challenging for coarse-grained models, such as the permeation of ions across a lipid membrane and the rupture of membranes due to an electrostatic field, at relatively low computational cost.
| Since the first introduction of physics-based coarse-grained (CG) models in computational biology [1], CG models have become increasingly popular in the simulation of complex biological systems [2]. They significantly reduce the computational complexity in comparison to all-atom (AA) models and allow sampling over much longer time scales and of larger system sizes. One of the most widely applied CG models is the MARTINI force field [3]. The MARTINI model was initially developed for lipid systems [4] and has recently been extended for proteins [5] and carbohydrates [6]. In general a four-to-one mapping is used in MARTINI, which means that on average four atoms and associated hydrogens are represented by one CG bead. The CG particles interact with the other CG particles in the system by means of Lennard-Jones (LJ) interactions; in addition charged groups (e.g. ions, lipid head groups, charged amino acid side chains) interact via a Coulombic energy function. Water is treated explicitly, at the same level of coarse-graining as all other molecules implying that four water molecules are combined into a single coarse-grained bead.
MARTINI water beads, just as many other CG water models, do not bear charges and, consequently, are blind to electrostatic fields and polarization effects. To compensate for the neglect of explicit polarization, screening of electrostatic interactions is done implicitly, assuming a uniform relative dielectric constant. While this is a reasonable approximation for bulk water, problems arise at the interfaces between water and other phases and in the vicinity of charged particles. Because of the implicit screening, the interaction strength of polar substances is underestimated in non-polarizable solvents. Correct modeling of the partitioning of polar and charged compounds into a low dielectric medium, e.g. a lipid bilayer, has proven a big challenge for CG models in general [7]. Applications involving the formation of polar/charged complexes in a non-polar environment are especially prone to be affected. A potential solution is to make the interaction potentials dependent on the local environment (see e.g. [8]), especially useful in solvent free approaches. With explicit solvent particles present, more flexibility is achieved with a polarizable water model.
Attempts to include the effect of polarization in simplified water models date already back to the early days of biomolecular modeling. Notably the development of the soft sphere dipole model is worth mentioning [9], [10]. In this model, water molecules are represented by point dipoles that can reorient in response to the electrostatic field of an embedded (macro)molecule. Recently, induced dipoles were also added to a CG solvent model, and made compatible with a CG protein force field [11]. The polarizability challenge also stands in all-atom (AA) force fields at a more fine-grained level. The AA force fields lack electronic polarizability, which has proven to be a significant drawback in simulations of ions and highly polarizable systems [12], [13]. Several approaches to develop polarizable AA force fields, such as the inducible point dipole model [14], the model with Drude oscillators [15], [16], the fluctuating charge model [17] and the multipole expansion model [18] exist. The general idea of all these methods is to introduce a fluctuating dipole to each polarizable particle, which responds to the local electric field in the vicinity of this particle.
In this work, we introduce orientational polarizability to the water beads of the MARTINI force field using an approach similar to that of the Drude oscillator [15], [16]. The resulting polarizable CG water model, in combination with the MARTINI force field, allows modeling the interaction of water with charged particles in a more realistic way. In the parameterization of the polarizable water model the following three criteria were used: i) The dielectric constant of bulk polarizable water should be sufficiently close to the value in real water; ii) The particle density of the polarizable water should be close to the particle density of the water in standard MARTINI; iii) The reproduction of partitioning free energies between water and organic solvents for a large variety of small compounds, one of the corner stones of the MARTINI model, should remain unaffected.
The rest of this paper is organized as follows. In the next section, we first describe the details of the model, and the way we set out to parameterize it. This is followed by the Results section in which we explore the parameter space and arrive at the optimal parameter set, based on reproduction of the density and dielectric constant of bulk water, and the water/oil partitioning behavior of the MARTINI building blocks. We then test a number of properties of the new model, including the dynamical behavior of bulk water, the surface tension of the water/vapor interface, and structural properties of ionic solutions and of a lipid bilayer. We also look at the effect of long-range electrostatic interactions. Finally, two applications are shown which would not have been feasible with the standard MARTINI model, nor with most other CG models. The applications are a realistic description of the free energy of ion transport across a lipid bilayer, and the electroporation process of both an octane slab and a lipid bilayer. A discussion section about the limitations and prospects of the model concludes this paper.
The polarizable CG water consists of three particles instead of one in the standard MARTINI force field (Fig. 1). The central particle W is neutral and interacts with other particles in the system by means of the Lennard-Jones interactions, just like the standard water particle. The additional particles WP and WM are bound to the central particle and carry a positive and negative charge of +q and −q, respectively. They interact with other particles via a Coulomb function only, and lack any LJ interactions. The bonds W-WP and W-WM are constrained to a distance l. The interactions between WP and WM particles inside the same CG water bead are excluded, thus these particles are “transparent” toward each other. As a result the charged particles can rotate around the W particle. The dipole momentum of the water bead depends on the position of the charged particles and can vary from zero (charged particles coincide) to 2lq (charged particles are at the maximal distance). A harmonic angle potential with equilibrium angle θ and force constant Kθ is furthermore added to control the rotation of WP and WM particles and thus to adjust the distribution of the dipole momentum. The average dipole momentum of the water bead will depend on the charge distribution and is expected to be on average zero in an apolar environment, such as the interior of the lipid bilayer. In contrast, some non-zero average dipole will be observed in bulk water or in some other polar environment. The masses of the charged particles as well as of the central particle are set to 24 amu, totaling 72 amu (the mass of four real water molecules).
There are five adjustable parameters in the polarizable water particle: The charge q, the distance l, the angle parameters θ and Kθ, and the atom type of the central particle W. The accessible range of the dipole momentum of the water bead is determined by both l and q; to restrict our parameter space we used q as the only adjustable parameter and fixed l at the value 0.14 nm. This distance is small enough to prevent the overlap of the charged particles of adjacent water beads (which could result in very large forces) and large enough to represent the instantaneous dipole of a cluster of four water molecules. Similarly, only Kθ was varied. The equilibrium angle was fixed at θ = 0 to ensure that the water bead in an apolar solvent has a vanishing dipole moment (recalling that one CG water bead effectively represents a cluster of four real water molecules).
It is clear that the polarizable water beads attract each other stronger than the standard CG water beads because of additional electrostatic interactions between their charged particles WP and WM. This additional attraction should be counter-balanced by a reduced LJ self-interaction of W particles. Thus we tested less attractive interaction levels II, III and IV (the standard MARTINI water has the atom type P4, which has a self-interaction strength level I. Note that, for each of these levels, the LJ parameter σLJ = 0.47 nm. The LJ well depth εLJ = 5.0, 4.5, 4.0, 3.5 kJ mol−1 for levels I–IV, respectively). Concerning the LJ interactions between the W particles and other particles in the MARTINI force field, our expectation was that these could stay unaffected, i.e. correspond to those for a P4 particle (for the full interaction matrix, see [3]). However, as we will show below, the cross-interaction strength has to be reduced slightly in order to reproduce the correct partitioning behavior.
Since the polarization of water is treated explicitly in our polarizable model, the global dielectric constant εr = 15 used in the standard MARTINI should be adjusted accordingly. This value of εr compromises between large ε in water and small ε in the hydrophobic regions like the core of the lipid membrane. In the polarizable model, the global dielectric constant is reduced to εr = 2.5 to ensure a realistic dielectric behavior in the hydrophobic regions. Other force field parameters are the same as in standard MARTINI [3].
All simulations were performed with the GROMACS suite of programs, versions 3.3.1 [19], 4.0.2 and 4.0.5 [20]. Standard simulation parameters associated with the MARTINI force field [3] were used unless stated otherwise. A time step of 20 fs was used in all simulations. We have repeated some of the simulations using 10 fs and 30 fs time steps; the results were virtually identical to the ones reported below. Temperature and pressure were kept constant by using weak coupling schemes [21], with time constants of τT = 0.3 ps and τp = 3.0 ps, respectively. The distance l between the central W particle and the charged WP/WM particles was constrained using the LINCS algorithm [22]. Visualization of the results was done with VMD [23]. Error estimates were obtained using a block-averaging procedure [24]. Details of the system composition and set-up are given alongside the presentation of the results. Times are reported as actual simulation time, except when explicitly stated as effective time in order to compare the kinetics to either all-atom simulations or experiment. The effective time accounts for the speed-up in coarse-grained dynamics (see [4]) and equals four times the actual simulation time. The parameter files are available in Dataset S1. They can also be downloaded from http://cgmartini.nl, together with some example applications.
The neglect of orientational polarizability in many water models associated with CG lipid force fields [54], [55], [56], [57], [58] is arguably one of the crudest approximations made. Water in those force fields is represented by spherically symmetric interaction sites either based on analytic potentials or effective potentials derived from atomistic simulations. None of these water models include electrostatic interactions, implying they are non-polarizable. The MARTINI model suffers from the same approximation. The inability to form a transmembrane water pore upon dragging a lipid across the membrane [3], or upon binding of antimicrobial peptides [41], [59], [60] are examples pointing at the shortcoming of the standard MARTINI water model.
To improve the behavior of the water model, inclusion of electrostatic interactions is needed; to account for the orientational polarizability, the minimum requirement is a point dipole, as in the models of e.g. Warshel and coworkers [9], [10], [13] and Orsi et al [61]. Our new water model is a three-bead model, consisting of a central particle with two charges-on-a-spring embedded, and was chosen as it combines simplicity with versatility. It is similar to the classical Drude model used in polarizable all-atom (AA) force fields to mimic electronic polarization [15], [16]. In contrast to the AA case, were the charged particles are massless and their position is solved in an expensive, self-consistent way, in our CG model the particles carry mass and follow the normal equation of motions. The model has only few adjustable parameters, yet enough of them to reproduce the dielectric properties of bulk water on the one hand and keeping at par with the standard MARTINI philosophy on the other. Despite the limited amount of free parameters in the model, a full exploration of parameter space is practically impossible; guided partly by intuition and partly through extensive testing we eventually settled on a combination of parameters which, overall, perform very well. Compared to the standard MARTINI water model, the polarizable model has improved properties, not only with respect to its dielectric behavior, but also for instance in the somewhat reduced freezing point. It can not be excluded that other combinations of parameters might perform even better, and we anticipate that further optimization of the model will take place in the future alongside with extending the range of applications of the model.
The main reason for having included polarizability into the model is the expectation that processes involving interactions between charged and polar groups in a low-dielectric medium are more realistically described. As an example we presented two applications for which standard CG models, including MARTINI, are less well suited, namely the translocation of ions across a lipid membrane and the electroporation of an octane slab and a lipid bilayer. Both processes involve the movement of charges from a high dielectric environment (water) to a low dielectric medium (membrane interior). A realistic description of such processes requires a model capable of performing local electrostatic screening. The two applications presented show that, despite being coarse-grained, our polarizable water model can do this at a level comparable to that of atomistic simulations. This opens the way to explore a number of important (bio)physical processes using the MARTINI model, including membrane poration by antimicrobial and cell penetrating peptides, DNA transfection, salt-induced membrane fusion, functioning of the voltage gated membrane channels, electroporation, and electrokinetic phenomena in general.
Finally, it is important to point out a few limitations of the polarizable water model: First, it is slightly more expensive from a computational point of view (for a pure water system the simulations are slowed down by a factor of approximately three). Second, the current parameterization of the model is not as thoroughly tested yet in comparison to the standard MARTINI model. For example lipid phase behavior, or the effect on proteins and peptides is largely unexplored. Third, despite an overall improved performance, some properties are still not at par with experimental measurements or data from atomistic simulations. These include the air/water surface tension, which is significantly too low, and also the sign of the membrane dipole potential which is opposite to that observed with more detailed force fields. Further improvement could be obtained by changing the analytical form of the non-bonded potential (i.e. moving away from the LJ 12-6 form), and by adding polarizability to other beads in the force field. The latter idea may also lead to a more realistic description of the protein backbone, allowing secondary structure formation to be described with MARTINI, an option we are currently exploring. We finally note that the polarizable MARTINI water model is not meant to replace the standard MARTINI water model, but should be viewed as an alternative with improved properties in some, but similar behavior at reduced efficiency in other applications.
|
10.1371/journal.pcbi.1005700 | Extracting replicable associations across multiple studies: Empirical Bayes algorithms for controlling the false discovery rate | In almost every field in genomics, large-scale biomedical datasets are used to report associations. Extracting associations that recur across multiple studies while controlling the false discovery rate is a fundamental challenge. Here, we propose a new method to allow joint analysis of multiple studies. Given a set of p-values obtained from each study, the goal is to identify associations that recur in at least k > 1 studies while controlling the false discovery rate. We propose several new algorithms that differ in how the study dependencies are modeled, and compare them and extant methods under various simulated scenarios. The top algorithm, SCREEN (Scalable Cluster-based REplicability ENhancement), is our new algorithm that works in three stages: (1) clustering an estimated correlation network of the studies, (2) learning replicability (e.g., of genes) within clusters, and (3) merging the results across the clusters. When we applied SCREEN to two real datasets it greatly outperformed the results obtained via standard meta-analysis. First, on a collection of 29 case-control gene expression cancer studies, we detected a large set of consistently up-regulated genes related to proliferation and cell cycle regulation. These genes are both consistently up-regulated across many cancer studies, and are well connected in known gene networks. Second, on a recent pan-cancer study that examined the expression profiles of patients with and without mutations in the HLA complex, we detected a large active module of up-regulated genes that are both related to immune responses and are well connected in known gene networks. This module covers thrice more genes as compared to the original study at a similar false discovery rate, demonstrating the high power of SCREEN. An implementation of SCREEN is available in the supplement.
| When analyzing results from multiple studies, extracting replicated associations is the first step towards making new discoveries. The standard approach for this task is to use meta-analysis methods, which usually make an underlying null hypothesis that a gene has no effect in all studies. On the other hand, in replicability analysis we explicitly require that the gene will manifest a recurring pattern of effects. In this study we develop new algorithms for replicability analysis that are both scalable (i.e., can handle many studies) and allow controlling the false discovery rate. We show that our main algorithm called SCREEN (Scalable Cluster-based REplicability ENhancement) outperforms the other methods in simulated scenarios. Moreover, when applied to real datasets, SCREEN greatly extended the results of the meta-analysis, and can even facilitate detection of new biological results.
| Confidence in reported findings is a prerequisite for advancing any scientific field. Such confidence is achieved by showing replication of discoveries in new studies [1]. In recent years studies have shown low reproducibility of results in several domains, including economics [2], psychology [3], medicine [4], and biology [5–7]. A new methodology called replicability analysis was recently suggested as a way to statistically pinpoint replicated discoveries across studies while controlling for the false discovery rate (FDR) [8]. This type of analysis is essential when trying to detect new hypotheses by integration of existing data from multiple high-throughput experiments.
The practical importance of replicability analysis is twofold. First, it quantifies the reliability of reported results. Second, collated information from multiple studies can identify scientific results that are beyond the reach of each single study. Indeed, in Genome Wide Association Studies (GWAS) replicability analysis allowed detection of new results that were not identified in meta-analysis, demonstrating that the two approaches are complementary [9].
Meta-analyses are widely applied and have been extensively studied in statistics [10] and in computational biology [11, 12]. However, in recent years the changes in the scale and scope of public high-throughput biomedical data have posed new methodological challenges. The first, and more obvious, is accounting for inflation in the number of false discoveries due to the multiplicity of outcomes, as hundreds of thousands and even millions of hypotheses are tested (see Zeggini et al. [13] for example). The second challenge is directly assessing consistency of results, which is not addressed by the classic null hypothesis of meta-analysis that the effect size is 0 in all the studies. Third, there is a need to distinguish between true effects that are specific to a single study and true effects that represent general discoveries and thus are replicable. For example, Kraft et al. [14] suggested that the effect of common genetic variants on the phenotype may correlate with population biases in a specific GWAS. While these are real discoveries in the sense that similar estimated effects are expected to be observed if the experiment could be replicated on the same cohort, their scientific importance is limited because they are specific to that cohort. For this reason, the authors argue that it is important to identify the association in additional studies conducted using a similar, but not identical, study base.
In recent years several frequentist approaches were suggested for the problem. Benjamini and Heller [15] introduced an inferential framework for replicability that is based on tests of partial conjunction null hypotheses. For meta-analysis of n studies of the same m outcomes and u = 1…n, the partial conjunction Hu/n(g) is that outcome g has a non-null effect in less than u studies. Thus H1/n(g) is the standard meta-analysis null hypothesis that outcome g has a null effect in all n studies. The authors introduced p-values for testing Hu/n(g) for each outcome. Benjamini, Heller and Yekutieli [8] applied the Benjamini-Hochberg FDR procedure [16] (BH) to the partial conjunction hypotheses p-values, and suggested setting u = 2 in order to assess replicability. Heller et al. [17] developed an approach for checking if a follow-up study corroborates the results reported in the original study. Song and Tseng [18] proposed a method to evaluate the proportion of non-null effects of a gene. However, they used the standard meta-analysis null hypothesis and their method cannot handle composite hypotheses, which are partial conjunctions Hu/n with u > 1.
Bayesian methods handle these shortcomings and offer a powerful framework for replicability analysis. For analyzing results from a single study, Efron introduced an empirical Bayes framework called the two-groups model [19]. It allows explicit analysis of the distribution of the statistic (e.g., p-values) of the underlying null and non-null groups. This clustering-based structure is then used to quantify the FDR of a rejection rule, and to compute a single point statistic, which is referred to as local Bayes FDR, or simply fdr. Heller and Yekutieli [9] introduced a method called repfdr, which extends the two-groups model for testing the partial conjunction hypotheses to the multi-study case. Formally, the problem is as follows: given an n × m matrix Z, where Zi,j is the p-value (or z-score) of object (e.g., gene) i in study j, our goal is to identify the objects that are k-replicable (i.e., significant in k or more studies) while controlling the fdr.
Repfdr estimates the posterior probabilities of the various configurations of outcome effect status (null or non-null) across studies, and computes the fdr for each partial conjunction null by summing the posterior probabilities for the relevant configurations. The authors showed that their approach controls the FDR and offers more power than the frequentist methods. However, repfdr is not scalable and can only handle a few datasets. In addition, it was particularly designed for GWAS datasets in which the number of tested objects (e.g., SNPs) is very large (i.e., > 100k).
In this study, we propose ways to overcome the limitations of repfdr. Our focus is on allowing efficient computation when m is large and n is limited, such as in gene expression datasets (i.e., n ∼ 20k). To reach this goal we make three simplifying assumptions: (1) we ignore the effect size, (2) we ignore the direction of the statistic, and (3) we assume that the studies originate from independent clusters. In addition, to handle larger values of m we compute an upper bound for the fdr, cutting the running time substantially. Our main algorithm is called SCREEN (Scalable Cluster-based REplicability ENhancement). It first detects the study clusters, then uses Expectation-Maximization (EM) to model each cluster, and finally merges the clusters using dynamic programming. Other algorithms that we propose here include two variants of SCREEN that differ in the way the studies are clustered: SCREEN–ind assumes independence and treats each study as a single cluster, and repfdr-UB puts all studies in one cluster. We compared SCREEN to other algorithms using various simulated scenarios and showed that only SCREEN had consistently low empirical false discovery proportions, and very high detection power.
We applied SCREEN to two cancer datasets, where each is a collection of case-control gene expression experiments. In both cases SCREEN greatly improved the results obtained by standard meta-analysis, and provided new biological insights. The first dataset is a collection of 29 case-control gene expression cancer studies from different tissues. Here, SCREEN detected a large set of genes that are consistently up-regulated, highly enriched for cell proliferation and cell cycle regulation functions, and are well connected in known gene networks, indicating their functional coherence. The second dataset is a recent pan-cancer study that examined the expression profiles of patients with and without mutations in the HLA complex across 11 cancer types [20]. SCREEN detected a large set of up-regulated genes that are related to immune responses. Importantly, SCREEN reported many more immune response genes than the original study thanks to our ability to quantify the fdr, and allowed detection of prominent genes and pathways that were not reported previously.
Outline: After introducing some background and notation, we present a dynamic programming algorithm for calculating the fdr under the assumption that the studies are independent. Second, we discuss EM-based algorithms for dealing with dependence. We then show that given the prior probability of only a subset of all configurations an upper-bound for the fdr can be computed. This leads us to a much faster algorithm that can handle many studies. Third, we extend the dynamic programming algorithm to handle independent clusters of studies. This will lead us to the complete SCREEN algorithm, which is based on EM-based dependency modeling within study clusters and merging the results using dynamic programming. Finally, we discuss experimental results on simulated and real datasets.
We start with a brief introduction to the single-study model. For a full description and background see [19]. Given a large set of N hypotheses tested in a large-scale study, the two-groups model provides a simple Bayesian framework for multiple testing: each of the N cases (e.g., genes in a gene expression study) are either null or non-null with prior probability π0 and π1 = 1 − π0, and with z-scores (or p-values) having density either f0(z) or f1(z). When the assumptions of the statistical test are valid, we know that the f0 distribution is a standard normal (or a uniform distribution for p-values), and we call it the theoretical null. The mixture density and probability distributions are:
f ( z ) = π 0 f 0 ( z ) + π 1 f 1 ( z ) F ( z ) = π 0 F 0 ( z ) + π 1 F 1 ( z )
For a rejection area Ƶ y = ( - ∞ , y ), using Bayes rule we get:
F d r ( Ƶ y ) ≡ P r { n u l l | z ∈ Ƶ y } = π 0 F 0 ( y ) / F ( y )
We call Fdr the (Bayes) false discovery rate for Ƶ: this is the probability we would make a false discovery if we report Ƶ as non-null. If Ƶ is a single point z0 we define the local (Bayes) false discovery rate as:
f
d
r
(
z
0
)
≡
P
r
{
n
u
l
l
|
z
=
z
0
}
=
π
0
f
0
(
z
0
)
/
f
(
z
0
)
Previous work have shown that: (1) the Bayes Fdr is tightly related to false discovery control in the frequentist sense, and (2) using a threshold on the local fdr for defining discoveries is equivalent to the optimal Bayes rule in terms of classification between nulls and non-nulls. Moreover, a threshold of 0.2 on the local fdr was suggested [19, 21]. Note that in our Bayesian setting computing the local fdr of a gene is an estimation problem. Within the context of our study we incorporated and tested two established methods for two-groups estimation studies: locfdr [22] and estimation based on mixture of Gaussians, which we call normix, see Materials and methods for details.
Consider an extension of the two group model to analysis of n genes over m > 1 studies. The data for gene i are a vector of m statistics Zi,⋅ = (Zi,1, ⋯, Zi,m) that are all either z-scores or p-values. For simplicity, from now on we assume that these data are z-scores. The unknown parameter for gene i = 1, ⋯, n is a binary configuration vector Hi,⋅ = (Hi,1, ⋯, Hi,m), with Hi,j ∈ {0, 1}. If Hi,j = 0 then gene i is a null realization in study j, and it is a non–null realization otherwise.
We assume that in each study j the parameters of the two-groups model θ j : ( π 0 j , f 0 j , f 1 j , f j ) are fixed and focus on replicability analysis. Generally, unless mentioned otherwise, we assume that the genes are independent. However, note that estimation of θj can account for gene dependence within study j [23, 24]. Finally, we also assume that the z-scores of a gene are independent given its configuration. That is,
P ( Z i , · | H i , · ) = ∏ j = 1 m P ( Z i , j | H i , j ) = ∏ j = 1 m ( f 0 j ( Z i , j ) ) ( 1 − H i , j ) ( f 1 j ( Z i , j ) ) H i , j
Next, we use h ∈ {0, 1}m to denote an arbitrary configuration vector, and π(h) to denote a probability assigned to the parameter space. We assume that the researcher has a set of configurations H 1 ⊆ { 0 , 1 } m that represents the desired rejected genes. Here we will assume that H 1 corresponds to genes that are non-null in at least k studies: H 1 = { h : | h | ≥ k }, where | h | = ∑ j = 1 m h j.
As a note, selection of k depends on the research question at hand. For example, Heller and Yekutieli used k = 2 to detect minimal replicability of SNPs in a GWAS [9]. Low k values can also be reasonable if the m studies represent different biological questions that are related, such as differential expression experiments from different cancer subtypes. On the other hand, if the m studies represent tightly related experiments such as biological replicates then larger k (e.g., m/2) seems more reasonable.
The local false discovery rate (fdr) of a gene i can be formulated as:
f d r ( Z i , · ) = P r ( H 1 ¯ | Z i , · ) = ∑ h : h ∉ H 1 P ( h | Z i , · ) = ∑ h : h ∉ H 1 P ( Z i , · | h ) P ( h ) P ( Z i , · )
For a given k and H 1 = { h : | h | ≥ k } we get:
f d r k ( Z i , · ) = ∑ h : | h | < k P ( Z i , · | h ) P ( h ) P ( Z i , · )
We first address the case where studies are independent.
Lemma. If the studies are independent (in the parameter space) then:
f d r k ( Z i , · ) = ∑ h : | h | < k ∏ j = 1 m P ( Z i , j | h j ) P ( h j ) f j ( Z i , j )
Proof. First, note that under the independence assumption P ( h ) = ∏ j = 1 m P ( h j ) = ∏ j = 1 m π 0 j ( 1 - h j ) ( 1 - π 0 j ) h j. Second, as the z-scores are independent given the configuration vector h we get that:
P ( Z i , · ) = ∑ h P ( Z i , · | h ) P ( h ) = ∑ h ∏ j = 1 m P ( Z i , j | h j ) π 0 j ( 1 - h j ) ( 1 - π 0 j ) h j = ∏ j = 1 m ( P ( Z i , j | h j = 0 ) π 0 j + P ( Z i , j | h j = 1 ) ( 1 - π 0 j ) )
Proposition: If the studies are independent then fdrk can be computed in O(mnk).
Proof. By the lemma, the fdr of a gene is based on the product of the two-group model densities in each study. Therefore:
f
d
r
k
i
n
d
e
p
(
Z
i
,
·
)
=
∑
h
:
|
h
|
<
k
∏
j
=
1
m
(
π
0
j
f
0
j
(
Z
i
,
j
)
)
1
-
h
j
(
(
1
-
π
0
j
)
f
1
j
(
Z
i
,
j
)
)
h
j
f
j
(
Z
i
,
j
)
We use dynamic programming to calculate fdrk(zi) as follows. Define:
U [ i , j , k * ] = ∑ h : | h | = ( k * - 1 ) ∏ j = 1 m ( π 0 j f 0 j ( Z i , j ) ) 1 - h j ( ( 1 - π 0 j ) f 1 j ( Z i , j ) ) h j f j ( Z i , j )
These values can be calculated (for each gene i) by updating a table of m × (k + 1) values. The base cases are:
U [ i , j , 1 ] = ∏ j = 1 m π 0 j f 0 j ( Z i , j ) f j ( Z i , j )
The recursive formulas are:
U [ i , j , k * ] = π 0 j f 0 j ( Z i , j ) f j ( Z i , j ) U [ i , j - 1 , k * ] + ( 1 - π 0 ) j f 1 j ( Z i , j ) f j ( Z i , j ) U [ i , j - 1 , k * - 1 ]
Finally, to obtain the fdr of a gene we sum over the values in the last column:
f d r k i n d e p ( Z i , · ) = ∑ k * = 1 k - 1 U [ i , m , k * ]
The running time for analyzing each gene is O(mk) and the total running time is O(nmk).
The empirical Bayes method of [9] estimates the prior distribution π(h) directly from the data. This approach has two drawbacks. First, the EM algorithm explicitly keeps a value for each possible configuration, which makes the algorithm intractable when m is large. Second, the estimation for rare configurations might be inaccurate, unless n >>2m. As an alternative, we develop an algorithm that keeps in memory only a small set of high probability configurations. We then use these estimates to obtain an upper bound for the fdr of a gene.
We first describe the EM in the full configuration space, and then modify it for the constrained case. That is, the EM is guaranteed to improve the solution and converge. The unrestricted EM formulation is based on repfdr [9, 25], as follows:
The E-step:
P ( H i , · = h | Z i , · , π ( t ) ( h ) ) = f ( Z i , · | h ) π ( t ) ( h ) ∑ h ′ f ( Z i , · | h ′ ) π ( t ) ( h ′ )
The M-step:
π ( t + 1 ) ( h ) = 1 n ∑ i P ( H i , · = h | Z i , · , π ( t ) ( h ) ) = 1 n ∑ i f ( Z i , · | h ) π ( t ) ( h ) ∑ h ′ f ( Z i , · | h ′ ) π ( t ) ( h ′ )
This process guarantees convergence to a local optimum. Our goal is to limit the search space.
Lemma. The EM algorithm above can be used to find a local optimum estimator under the constraint ∀ h ∉ H ′ π ( h ) = 0, for any non-empty configuration set H ′.
Proof. Note that during the EM iterations, if at some time point t π(t)(h) = 0 then ∀t* > t π(t*)(h) = 0. Therefore, setting the starting point of the EM such that ∀ h ∉ H ′ π ( 0 ) ( h ) = 0 satisfies the constraint and ensures convergence.
In this section we present a restricted version of the EM above, which we call repfdr-UB. To describe it, we need additional notation. Given a configuration vector h ∈ {0, 1}m, let h[l] be the vector containing the first l entries of h. Given a real valued vector v, let v(i) denote the i’th smallest element of v.
Our algorithm is based on the simple observation that if π(h[l]) ≤ ϵ then any extension of h[l] cannot exceed ϵ. That is, π(h[l + 1]) ≤ ϵ regardless of the new value in study l + 1. Our algorithm works as follows. The user specifies a limit to the number configurations kept in the memory—nH. For simplicity we assume that nH is a power of 2. We first run the unrestricted EM algorithm on the first log2(nH) − 1 studies. Each subsequent iteration adds a new study. In each iteration l we keep four parameters: (1) H ^ l—the set of the top nH probability configurations, (2) π ^ l—the vector of their assigned probabilities, (3) ξ ^ l—an estimation of ∑ h [ l ] ∈ H ^ l π l ( h [ l ] ), and (4) ϵ ^ l—an estimation of the maximal probability among the excluded configurations.
Initially, l = log2(nH) − 1 and H ^ l contains all possible configurations of the first l studies. In addition ξ ^ l = 1, and ϵ ^ l = 0. In iteration l + 1 we run the restricted EM algorithm on all possible extensions of H ^ l. That is, the input configuration set for the EM is a result of adding either 1 or 0 at the l + 1 position of each configuration in H ^ l. The EM run produces initial estimations for our parameters, on which the following ordered updates are applied:
π ^ l + 1 = ξ ^ l π ^ l + 1 (1) H ^ l + 1 = { h [ l + 1 ] ; π ^ l + 1 ( h [ l + 1 ] ) ≥ π ^ ( n H / 2 ) l + 1 } (2) ξ ^ l + 1 = ∑ h [ l + 1 ] ∈ H ^ l + 1 π l + 1 ( h [ l + 1 ] ) (3) ϵ ^ l + 1 = max ( ϵ ^ l , max h [ l + 1 ] ∉ H ^ l + 1 ( π ^ l + 1 ( h [ l + 1 ] ) ) ) (4)
Note that in step Eq (2) above we keep the top nH/2 configurations in H ^ l + 1. This set is then used as input to the EM run in the next iteration. We repeat the process above until l = m. The output of the algorithm is H ^ m , π ^ m , ξ ^ m , ϵ ^ m.
In this section we apply the ideas from the previous sections to obtain an algorithm for calculating the fdr under the assumption that the studies originate from independent clusters. We call this algorithm SCREEN (Scalable Cluster-based REplicability ENhancement, see S1 Text for an overview of the method). Briefly, our algorithm has three stages. First, we use the EM-process on each study pair to create a network of study correlations. We then cluster the network to obtain a set of study clusters that are likely to be independent, see Materials and methods for the full description of this step. Second, we run the EM on each cluster separately. Finally, we merge the results from the different clusters using dynamic programming. Note that this algorithm is a heuristic as it uses EM within each cluster.
We analyzed two real datasets. The first, which we call Cancer DEG, is a collection of gene expression studies that compared cancer to non-cancer tissues. The second, called HLA, is from [20], where Shukla et al. tested differential expression between cancer samples with and without somatic mutations in the HLA complex across 11 TCGA cancer subtypes.
We presented here several novel algorithms for detecting replicated associations using an empirical Bayes approach. Our main algorithm, called SCREEN, outperformed other approaches in many scenarios and had consistently low false discovery proportion and high true discovery rates in all simulations.
SCREEN works in three stages. First, it clusters the studies based on their pairwise correlations, which are learned via EM. Second, it performs replicability analysis within each cluster using our restricted EM approach. This method goes beyond previous studies by restricting the possible number of study configurations that are kept in memory. As a result, the method can analyze large study clusters, by computing an upper bound for the fdr instead of an exact estimation. Finally, the results of the replicability analyses of the clusters are merged using dynamic programming. For a given k, the output of SCREEN is the fdrk value for each gene, which can be used to detect genes that are non-null in at least k studies.
We have shown that SCREEN performs well on various simulated scenarios, as well as on real datasets. Specifically, we analyzed two collections of cancer-related gene expression studies. In both cases the discovered gene sets highlighted active gene modules with pertinent functions.Such modules are revealed by the projection of the discovered genes on an interaction network and focusing on well-connected subnetworks. Notably, standard meta-analysis does not reveal many of the genes and generates more fragmented and less coherent subnetworks. For example, in the cancer DEG datasets, some of these genes are central in the network and are known master regulators of cell cycle (e.g., CDK1). In summary, we demonstrated replicability analysis as a standard tool for analyzing a large collection of studies, and provided novel algorithms that are accurate and scalable.
While the current version of SCREEN does not model the direction of the statistic directly (i.e., up- or down-regulation), we addressed this point empirically in our examples and showed that most genes were consistent in their direction. For the sake of functional analysis we required a gene to have the same direction in at least 75% of the studies. This threshold reflects a reasonable selection between the number of reported genes and the consistency requirement (see S8 Fig). Of course, users can change this threshold to require a higher (or lower) consistency in direction in specific applications. Also, note that detection of “mixed sign” genes is an important feature of our analysis: the causality of such genes is questioned as they likely represent downstream effects.
The strategy of SCREEN can be extended to a more complex definition of replicability across study clusters. For example, a researcher may seek genes that are replicable across one or more study clusters, where a gene is replicable in a cluster only if it is non-null in at least some predefined percentage of the studies in that cluster. See S1 Text for a discussion on this topic.
A variety of methods for integration of different studies have been developed in GWAS but their goals are different from SCREEN’s. These methods typically assume a standard meta-analysis null hypothesis that the effect size is zero across all studies [31]. As such, they do not address the question of replication directly even if a clustering of the studies is considered [32]. Moreover, some of the Bayesian approaches that were developed for GWAS utilize a subjective prior and do not estimate it form the data [33]. Finally, as we have shown, SCREEN outperforms repfdr, which was originally developed for GWAS data [9].
Our study has some limitations that can be addressed in future research. First, we assumed that genes are independent. This assumption is usually made by state of the art methods, but is often incorrect. In our case, it was used to obtain tractable algorithms (both the dynamic programming and the EM). Second, while our algorithms report fdrk values of genes, we currently do not estimate their variance. Third, selection of k was done manually on real datasets based on the specific biological question and the number of reported genes (e.g., Fig 3A).
Fourth, our restricted EM approach to analyze study clusters is a heuristic that only guarantees convergence into a local optimum. Thus, while our algorithm has a deterministic starting point for the EM, setting starting points at random may change the obtained upper bounds for the fdr. Indeed, when we tried using random starting points, the set of reported genes in the cancer DEG dataset changed (by up to 50 genes, but still leaving over 80 genes from the original results), and no change was observed in the HLA dataset. Another important aspect of the heuristic is the number of allowed gene configurations. As an example, SCREEN with nH = 10000 configurations on the cancer DEG dataset finds more than 100 additional genes for k = 20, whereas the fdr of the genes reported in our original analysis (Fig 4) is kept low, illustrating that the results are consistent (see S9 Fig).
Fifth, while our simple Exp-count approach to estimate the expected number of non-null realizations of a gene performed reasonably well in some simulated scenarios, it is only partially justified theoretically (see S1 Text). Finally, our methods rely on fixed estimates of the two-groups model of each study. Future methods could go a step further and estimate all parameters (i.e., both the study parameters and the gene configuration probabilities) in a single flow.
We tried two implementations of two-groups estimation algorithms. The first locfdr [22], provides two options to learn the empirical null: maximum likelihood and central matching. By default, we used the maximum likelihood estimator. However, in practice this algorithm might converge to a solution in which π ^ 0 > 1. Whenever this occured, we tried the central matching approach instead. If the new estimator also had π ^ 0 > 1 we used the theoretical null.
The second approach was based on two previous methods: Znormix [34] and fdrtool [35]. Znormix uses EM to learn a mixture of Gaussians, whereas fdrtool assumes that the null distribution is a half normal distribution. Here, we applied an EM approach to the absolute values of the z-scores. We extended these methods by learning a mixture of a half normal with σ ≥ 1 for the null distribution, and a normal distribution with μ > 0 for the non-nulls. We call this approach normix.
In practice, we discovered that our EM algorithm is sensitive to high values in the estimation of f1. In addition, the methods above do not exploit additional information that could be obtained from the two-groups model: an estimation for the power of a study [19]. That is, this is a measure of how separated the two groups are. In our analyses, we took a very stringent approach: in each study we multiply f1(z) by the estimated power of that study. The effect is a shrinkage in the f1(z) values that is proportional to the estimated quality of the study.
SCREEN relies on a known partition of the studies into clusters. In this section we use an empirical Bayes approach to obtain the clusters. Our analysis has two main parts: learning a network, and clustering.
First, we create a correlation network among the studies. For each study i, let ai = P(hi = 1) be the marginal non-null probability in that study. For studies i, j let ai,j = P(hi = 1 ∧ hj = 1) be the shared non-null probability of the two studies. We estimate these parameters as follows: ai and aj are taken from the two-groups model of each study, and ai,j is estimated by running our EM approach on the data of these two studies. The correlation of the studies is then estimated by:
r i , j = a i , j - a i a j a i ( 1 - a i ) a j ( 1 - a j )
We obtain a robust estimation of ri,j by taking the mean of 100 bootstrap runs of the procedure above. That is, in each run we reestimate ai,j by running the EM on a bootstrap sample of the genes (n/2 genes out of n, sampled with replacement).
Next, we cluster the network using the infomap algorithm [36]. Here, communities are detected using random walks in the underlying graph. As the input for this algorithm is an unweighted network, we used a threshold of 0.1 for the absolute correlation of study pairs to determine edge presence. This threshold is relatively low for general clustering tasks as it does not guarantee high homogeneity within clusters. However, it guarantees that the clusters discovered by SCREEN will be well-separated. In practice, our clustering approach found the correct clustering of studies in all simulations performed.
In order to evaluate our fdr approaches we compared them to several extant methods for multi-study analysis. Here we outline them briefly.
Network analysis and visualization was done in Cytoscape [39]. The GeneMANIA Cytoscape app [40, 41] was used to create the gene networks of the selected gene sets. GO enrichment analysis was performed using Expander [42].
The datasets and an implementation of SCREEN are available in S1 Data.
|
10.1371/journal.pntd.0004680 | The Role of Serotype Interactions and Seasonality in Dengue Model Selection and Control: Insights from a Pattern Matching Approach | The epidemiology of dengue fever is characterized by highly seasonal, multi-annual fluctuations, and the irregular circulation of its four serotypes. It is believed that this behaviour arises from the interplay between environmental drivers and serotype interactions. The exact mechanism, however, is uncertain. Constraining mathematical models to patterns characteristic to dengue epidemiology offers a means for detecting such mechanisms. Here, we used a pattern-oriented modelling (POM) strategy to fit and assess a range of dengue models, driven by combinations of temporary cross protective-immunity, cross-enhancement, and seasonal forcing, on their ability to capture the main characteristics of dengue dynamics. We show that all proposed models reproduce the observed dengue patterns across some part of the parameter space. Which model best supports the dengue dynamics is determined by the level of seasonal forcing. Further, when tertiary and quaternary infections are allowed, the inclusion of temporary cross-immunity alone is strongly supported, but the addition of cross-enhancement markedly reduces the parameter range at which dengue dynamics are produced, irrespective of the strength of seasonal forcing. The implication of these structural uncertainties on predicted vulnerability to control is also discussed. With ever expanding spread of dengue, greater understanding of dengue dynamics and control efforts (e.g. a near-future vaccine introduction) has become critically important. This study highlights the capacity of multi-level pattern-matching modelling approaches to offer an analytic tool for deeper insights into dengue epidemiology and control.
| The fluctuations of multi-serotype infectious diseases are often highly irregular and hard to predict. Previous theoretical approaches have attempted to disentangle the drivers that may underlie this behaviour in dengue dynamics with variable success. Here, we examine the role of such drivers using a pattern-oriented modelling (POM) approach. In POM, multiple patterns observed at different scales are used to test a model’s proficiency in capturing real-world dynamics. We examined dengue models with combinations of cross-immunity, cross-enhancement, seasonal fluctuations in the transmission rate, and with sensitivity analyses of asymmetric transmission rates between serotypes as well as the possibility for four subsequent heterologous infections. We demonstrate the ability of POM to model dynamical drivers that have gone unnoticed in single pattern or synthetic likelihood approaches. Further, our results present a determining role of seasonality in the selection and operation of these processes in governing dengue dynamics, in particular when full, heterologous immunity is assumed to occur after a secondary infection. We show that this structural model uncertainty can have important practical significance, as demonstrated by the differences in control efforts required to disrupt transmission. These results highlight the importance of localised model selection and calibration using multiple data-matching, as well as taking explicit account of model uncertainty in predicting and planning control efforts for multi-serotype diseases.
| With a 30-fold increase in incidence over the last five decades, dengue poses an increasing threat to about two thirds of the world population [1]. Dengue, caused by a group of viruses belonging to the Flavivirus genera, circulates in four major serotypes (DENV 1–4) [2], and manifests in a wide spectrum of clinical forms, from subclinical to classic dengue fever to the more serious forms of the disease, namely, dengue haemorrhagic fever (DHF) and dengue shock syndrome (DSS). In the absence of treatment, dengue can be highly fatal in subjects with DHF or DSS, with a case-fatality rate of 15%, which may be reduced to 1% with adequate medical intervention [3]. Despite on-going efforts, no effective antiviral drugs are available against the disease and the potential impact of the recently licenced vaccine has yet to be determined. This limits control efforts primarily to vector control [4].
Dengue dynamics are characterized by highly seasonal, multi-annual fluctuations, with replacement of serotypes occurring at varying intervals. An example of these patterns arising in a newly emerging dengue setting is illustrated in (Fig 1) [5,6]. This is thought to result from a complex interplay between environmental factors, vector ecology and host-pathogen dynamics [7]. Various hypotheses have been proposed to uncover the main drivers of dengue dynamics and to reveal how such drivers interact among themselves to govern infection and disease patterns in the field. Emphasis has been on unravelling the roles that cross-immunity (CI), cross-enhancement between serotypes, and seasonal variation in the transmission rate, play in capturing the complex dynamics of dengue [8]. Cross-enhancement is believed to be caused by antibody-dependent enhancement (ADE), where heterotypic antibodies facilitate cell entry through the formation of virion-antibody complexes, ultimately leading to increased viral titers upon secondary infection [9,10]. This is thought to result in increased susceptibility to a secondary heterologous infection and, upon these secondary infections, in a more serious form of disease and increased infectiousness. Enhanced disease severity is however believed to have minor impact on the dynamics as the proportion of DHF and DSS cases is substantially small (1% of confirmed cases [11]). By contrast, including sufficiently high levels of enhanced infectiousness or susceptibility (60–130%) in simulation models has been found to induce asynchronous outbreaks of different serotypes [12,13], an outcome which has been indicated to underlie the manifestation of the 3–5 year epidemic cycles observed for dengue dynamics in Thailand [14,15]. Decomposing ADE into both enhanced infectiousness and susceptibility has further been shown to mimic this effect at lower, more realistic values of ADE, while also reducing the magnitude of oscillations to more plausible levels and decreasing the risk of stochastic extinction [15]. Similarly, relaxing the common assumption of complete immunity after two heterologous infections results in asynchronous, multi-annual outbreaks at lower levels of ADE and R0 [16]. While most modelling endeavours have assumed serotypes to have identical characteristics, allowing for a small amount of asymmetry in the transmission rate is found to increase serotype persistence in the presence of ADE [17]. Furthermore, the inclusion of short-lived cross-immunity in models was found to be sufficient to reproduce the observed out-of-phase, irregular oscillations and 3-year cycles [18–21]. An alternative hypothesis has been proposed by Lourenço et al., who demonstrated that spatial segregation between human hosts and its vectors can be sufficient to capture the semi-regular dengue patterns observed, even in the absence of immune interactions [22]. By contrast, to mimic the distinct seasonal signature of dengue dynamics, the incorporation of seasonal forcing into the vector population dynamics or transmission rate has been found to be essential [19,22,23].
The above results hint at the complexity of dengue transmission and suggest that multiple mechanisms could underlie disease dynamics in any particular site. A key question in understanding dengue dynamics and control, therefore, is how best to use observed data in order to identify the processes governing the transmission of the disease in a given location. Recently, there has been increasing recognition that for complex systems, such as dengue, model matching to single or a few patterns is not sufficient to narrow down the range of possible explanatory mechanisms [24], and that matching to multiple patterns observed at various scales and hierarchical levels is required for identifying the mechanisms that generate such patterns, and hence are likely to be key elements of the system’s structure. Tying ecological models to multiple system patterns concurrently may also aid in detecting the right level of complexity and improve the predictive ability of such models for replicating local dynamics [24]. Methods such as Pattern Oriented Modelling (POM) allow for such a multi-scope approach by facilitating the design, selection, and calibration of models of complex systems [25–30].
This study applied a POM approach to modelling global dengue infection data in order to determine whether the above proposed mechanisms related to serotype interactions and seasonal forcing of the transmission rate were able to explain all of the observed dynamical patterns in the field. We further used the modelling results to investigate the vulnerability of dengue to interruption in transmission as a result of vector control, and examined how such vulnerability was related to the identified processes governing disease transmission. We demonstrate that model selection is largely driven by the seasonality of the system, with CI being a preferred mechanism in the case of low, and ADE in the case of highly seasonal transmission regimes. At similar levels of transmission rate, resistance to control efforts was found to increase in dengue systems with CI. The results highlight the utility of the POM approach for detecting and fitting of appropriately structured disease transmission models based on observed data. In addition, they also reveal challenges in structural and parameter identifiability that would remain unnoticed when guided by individuals patterns used in isolation.
Five characteristic dengue patterns were used to filter out unrealistic model structures and reduce parameter uncertainty. The patterns were selected to reflect the breadth of characteristics used in single pattern matching approaches [12,15,16,18,22], include strong and weak patterns that are common across endemic regions and those which are relatively stable over time and encompass different levels of organization [24]. The patterns (i.e. mean duration between peaks, multi-annual fluctuations, frequent replacement of one circulating serotype by another, serotype co-dominance and asynchronous serotype cycling) were derived from literature describing dengue case data and serotype epidemiology from different endemic regions across the world [5,6,31–42]. The observed patterns are described in Table 1.
We used a deterministic Susceptible-Infected-Recovered (SIR) modelling framework to describe the circulation of four different dengue serotypes (DENV1-4) in a population [13]. The full system of ordinary differential equations is shown in (Fig 2). The model consists of 26 compartments, each of which represents a fraction of the population. The population size is modelled to be stationary; hence births and deaths occur at an equal rate (μ). New-borns are assumed to be immunologically naïve to all serotypes and are born into the class of susceptibles (S). Although the presence of maternal antibodies is shown to affect the risk of infection, the impact on the overall dynamics is believed to be minimal and thus not taken into consideration [43]. Susceptibles become primarily infected by serotype i (Ii) at rate βSIi and αTRANSβSIji proportional to the number of primarily and secondarily infectious individuals respectively. The parameter αTRANS>1 indicates enhanced transmissibility of secondarily infected individuals. A seasonal change in the transmission rate (β(t)) is incorporated through a sinusoidal function with a forcing period of one year: β(t) = β0(1−β1 cos(2πt)) where β0 indicates the mean transmission rate and β1 the strength of seasonal fluctuation and t time in years. The transmission rate (β(t)) is assumed to be equal across serotypes. Individuals remain infectious for a period of 1/γ. After recovery from a primary infection, individuals become immune to all serotypes (Ci) for a period 1/ρ after which they move to the partially immune stage (Pi). The P-class individuals are assumed to experience full immunity against the serotype i and enhanced susceptibility (αSUS>1) to all other serotypes. They acquire secondary infection (Iij) at rates αSUSβPiIj and αTRANSαSUSβPjIkj proportional to the number of cases respectively primarily and secondarily infectious to a different serotype (with k≠j and j≠i). The duration of the infectious period is assumed to be equal upon secondary and primary infection. To account for imported cases and prevent the ODE-models to simulate unrealistically low levels of infections, individuals (susceptible or partially immune) can also acquire infection through an infectious contact with an individual from an external population at rate βδ, where δ signifies the import rate [23]. As tertiary and quaternary infections are rarely observed [44], we assume that after recovery from a secondary infection, individuals become life-long immune to all serotypes. An adaptive time step fourth and fifth -order Runge-Kutta solver was used with initial conditions for I1-4 1x10-7, 2x10-7, 3x10-7 and 4x10-7 and S=1−∑1−4iIi. All other state variables were initialized at zero. The implementation of the model, as well as the analysis of its simulation results were carried out in the Matlab, version 2014b (www.mathworks.com).
In this analysis we assume the following hypotheses (see Table 2). H1: The most parsimonious hypothesis is represented by the base-model with neither ADE (αSUS = 1 and αTRANS = 1) nor CI (individuals upon recovery from primary infection go straight to the P-class). H2: The base-model with CI. H3: The base-model with enhanced susceptibility, further referred to as ADE (αSUS>1 and αTRANS = 1). H4: H3 with CI. H5: The base-model with both enhanced susceptibility and transmissibility (i.e. ADEx2 with αSUS>1 and αTRANS>1) but no CI. H6: H5 with CI. In all models, an annual seasonal forcing in the transmission rate is assumed.
The variables that we estimated from the simulated data to contrast the dynamics of each model against the characteristics of dengue dynamics are: 1) Mean inter-peak period; 2) Presence of a multi-annual signal; 3) Duration of serotype replacement; 4) Intensity of single-serotype emergence; and 5) Serotype phase-locking.
The mean inter-peak period (MIPP) is defined as:MIPP=YN, where Y is the number of years analysed and N the number of peaks occurring during that period. To ensure comparability of the simulated estimates with reported observations on the inter-epidemic period, peaks were defined to have a minimum proportion of infectious people of 1/4000. To assess the presence of significant multi-annual signals in addition to the near yearly MIPP, a spectral density approach was used. To reduce the confounding effect of very low amplitude fluctuations, the time series were smoothed using a moving average filter. The power spectral density of the smoothed time series was assessed with the Welch’s overlapped segment averaging estimator [45]. To evaluate the significance of the periodic signals, the signals were compared to the null-continuum. The null-continuum is a greatly smoothed version of the raw periodogram, encapsulating the underlying shape of the distribution of variance over frequency [46]. A signal was assessed to be significant if the lower bound of the 90% confidence interval of the raw periodogram exceeded the null continuum [46]. The duration of serotype replacement is defined as the mean number of years before a dominant serotype during a peak is replaced by another serotype in a subsequent peak. The intensity of single serotype emergence (ε) was defined as by Recker et al. [47]:ε=1N∑iNγmaxi−γsubiγmaxi, where N defines the number of peaks occurring during the analysed number of years,γmaxi the prevalence of the dominant serotype and γsubi the prevalence of the serotype with the second-highest peak. Model runs with either complete co-dominance (ε<0.01) (i.e. there are multiple serotypes present at any point in time) or complete single serotype dominance (ε>0.99) were omitted. Lastly, serotype phase-locking here is defined as the perfect synchronization of serotypes and is detected by comparing the MIPP of serotype i to the aggregated MIPP. Simulations in which MIPP = MIPPi are discarded based on the presence of perfect phase-locking.
To determine which of the hypotheses or models capture the observed dengue dynamics and at which parameter values, we used a pattern oriented modelling approach (Fig 3) [25–28]. Model performance was assessed based on the extent to which a model captured all the 5 characteristics of dengue simultaneously, as defined above (Table 1). Models were assessed using the following steps. First, Latin hypercube sampling [48] was employed to select a sample of Ω (= 5,000) parameter vectors from a conjoint parameter distribution, encompassing the transmission rate (β0), the level of seasonal forcing or seasonality (β1) and, depending on the model, a combination of enhanced susceptibility (αSUS), enhanced transmissibility (αTRANS) and the rate of loss of CI (ρ) (Table 3). Uncertainty in the values of these parameters was addressed by assigning uniform distributions from their ranges deemed realistic according to literature (Table 3). The resulting ensemble of models (Model 1–6 with Ω parameter vectors) was run for 1400 years. The model outputs for the last 400 years were considered to determine whether the model mimicked all five dengue characteristics (a model is assumed to match a characteristic if the simulated response falls within the range of that characteristic pattern given in Table 1). The resulting set of passing (good) parameters G (where G ⊂ Ω) was retained as a multivariate distribution for further analysis.
To assess the impact of simplifying model assumptions on pattern-matching, we repeated the POM exercise for two distinct scenarios. One, we allowed for transmission rates to be uneven between serotypes (the asymmetric model). More specifically, serotype-specific transmission rates were drawn from a normal distribution with standard deviation 0.15 [17]. Two, we used a model variant that allows for four heterologous infections prior to acquiring complete immunity (the 4-infection model, equations are provided in S1 Text [52]).
We used logistic regression to assess the sensitivity of pattern-matching (binary response variable) to the parameters (independent variables). We normalised the independent variables on a 0 to 1 scale to obtain comparable regression coefficients: coefficients larger than|3| indicate strong sensitivity while parameters with small coefficients (<<|1|) have little impact on the model matching the patterns [53]. Two-way interactions were included in the construction of the logistic regression models: logit(p) = b0+b1β0+b2β1+b3αSUS+b4αTRANS+b5ρ+interactions, with p being the probability of a pattern-match, b0 the intercept and b1-n the regression coefficients.
Additionally, the identifiability of each of the parameters was examined using a principal component analysis (PCA) [54,55]. The identifiability of a parameter is a function of dependence, prior uncertainty and the model’s sensitivity to the parameter and defines how well one can estimate a parameter. We assessed the parameter identifiability for the full model (ADEx2+CI), using its passing distribution (G). First, the variance-covariance matrix(Σ)was constructed from the log-transformed G. Next, the principal components (PCs) were derived from Σ. The PCs of Σ define the 5-dimensional ellipsoid that approximates the population of passing parameter values. The eigenvalues (λi) denote the respective radii and the eigenvectors representing how much each parameter contributes to the direction of each radius. As such, λi gives an indication of the variance explained by the ith PC. The overall variance of all PCs was defined as ∑i=15λi=trace(Σ), thus the proportion of the total variation in G that was explained by the ith PC is was estimated by:λtrace(∑). We interpret these results as follows: A smaller λi indicates that the model is more sensitive to changes in the direction described by the ith component, whereas a larger λi signifies that the model is less sensitive to changes in the direction of the component. Parameters contributing most to a large λi are responsible for a big portion of the variation in the parameter space and are thus considered less identifiable.
We examined the vulnerability of the models to sudden reductions in the transmission rate that may be brought about by vector control. The models were run for all parameter sets in G for a burn-in period of 1000 years after which the system was perturbed by a reduction in the transmission rate (i.e. β0 is reduced by 90%) for a control period of w weeks per year. We varied w from 1 week to 52 consecutive weeks, starting at the valley of the sinusoidal function, which mimics the onset of the rainy season. After the control period of w weeks, β0 returns to its original value. These control runs were performed for 30 years after the burn-in period. The intervention of w weeks was assumed to be successful if no more than one peak occurred over the time-course of the model simulation. We assessed the probability of control for model i, where i represents 1 to 6, by calculating the proportion (Pwi) of Gi presenting successful control as a function of the number of weeks the transmission was disrupted. Here, Pwi=NwiGi with Nwi being the number of parameter vectors out of Gi that showed successful control for model i given w weeks of interruption in transmission. A composite average (Pw) for each control period w was derived by weighing the individual probability values of the models by the sizes of their passing parameter distributions (Gi), such that:Pw=∑i=16NwiGi.
Lastly, we estimated the values of the basic reproduction rate (R0) for each of the parameter vectors in G to assess the relation between transmission potential and the models’ vulnerability. The R0 of the model was derived using the next generation method [56–58] (Proof provided in S2 Text) and is defined as: R0=β0γ+μ, where β0 defines the transmission rate, 1/γ the duration of the infectious period and 1/μ the average life expectancy of the human host [59].
Fig 4 demonstrates the accepted parameter distributions (G) for the 2-infection models. While some parameters demonstrate broad distributions indicating limited uniqueness and abundant parameter interactions, others show clear preferential values and ranges that are sensitive to the structural components of the model. Overall it appears, as can be expected, that the more complex models fit the patterns at a wider parameter range.
Fig 4A shows that models with CI selected for relatively higher transmission levels relative to models with ADE only. For low transmission levels, the full model outcompeted all the other models, indicating that more complex models may be necessary to fit dengue dynamics at lower values of R0. These results are insensitive to the assumption of low levels of asymmetry in transmission rates (S2aA Fig). In contrast to this, the 4-infection models display similar fits at lower transmission levels (S2bA Fig).
Seasonality appeared to be the most prominent driver of model fit and selection in the 2-infection model (Fig 4B). Models with CI showed a marked shift towards lower seasonal forcing relative to the base-model. In fact, at low seasonality (β1<0.06) there is a strong preference for the inclusion of CI, as is especially notable from the elevated density levels of the ADE+CI and ADEx2+CI models. At high seasonality (β1>0.17) only the more complex models provided an adequate fit. At intermediate levels of seasonality (β1: 0.1–0.15) multiple models were equally proficient at replicating the dynamics, indicating a region of large model uncertainty. The model’s structural sensitivity to seasonality persisted when asymmetry in transmission rates was assumed (S2aB Fig). However, when we allowed for tertiary and quaternary infections, the medians and shapes of the passing parameter distributions for β1 were similar across the models (S2bB Fig).
The addition of CI to models with ADE results in higher levels of αSUS (Fig 4C), yet had minor impact on the median levels of αTRANS (Fig 4D). While previous publications suggested reduced estimates of αSUS and αTRANS upon the inclusion of decomposed ADE, analysis of the 2-infection model does not support this observation [15]. We did, however, observe this pattern in the 4-infection and asymmetric 2-infection model (S2aD and S2bD Fig).
The inclusion of ADE to the models with CI profoundly affects the estimated duration of cross-immunity by allowing for the selection of a much wider range of ρ (Fig 4E). Whereas the CI-model by itself only captures the characteristics at durations of cross-immunity shorter than half a year, the inclusion of ADE allows for cross-immune periods of up to 2 years, which is in line with the previous estimates [21]. Interestingly, in the case of 4-infection, the CI-only model performed well for a wider range of durations of cross-immunity, including estimates from Reich et al. [21].
Exploring the behaviour of the models in terms of MIPP and duration of serotype replacement (Table 4) reveals as to why there are differences in model fits across the range of seasonal forcing (S1aAA–S1aAF and S1aCA–S1aCF Fig). Increased levels of seasonal forcing are associated with longer MIPP. Temporary CI introduces a lag before a secondary infection can be acquired and thus generates a necessary build-up time period during which susceptible individuals accumulate in sufficient number to fuel the next outbreak. Thus, while an increase in seasonal forcing is characterized by longer inter-epidemic periods, at similar levels of seasonal forcing, the models with CI demonstrate a longer MIPP than the models without CI (S1aAA–S1aAF Fig). This allows the CI-only models capture the characteristic MIPP at lower seasonal levels than the models with just ADE. At higher levels of seasonal forcing, CI contributes to MIPPs that are longer than are characteristic to dengue. This effect is less pronounced in the 4-infection models. The overall immune population is smaller in the 4-infection models and therefore of less influence on the frequency of outbreaks. The same can be observed for the duration of serotype replacement (S1aCA–S1aCF Fig). In contrast to CI, the inclusion of ADE to the model results in shorter cycles, thus successful fits are observed at higher levels of seasonal forcing (S1aAA–S1aAF Fig).
Lastly, we observe a prominent impact of seasonal forcing on the occurrence of phase-locking. S1aEA–S1aEF Fig demonstrate a threshold-like value of β1 above which the system is forced into synchronized serotype dynamics. This threshold is relatively stable across the simple model structures (see also Fig 4B) and unaffected by the value of R0. Only the addition of decomposed ADE disrupts this behaviour, thereby being a possible driver of irregular serotype behaviour at higher seasonal regions. These phase-locking thresholds are stable to some level of asymmetry in transmission rates (S1bEA–S1bEF Fig), however they completely vanish in the case of 4-infection models (S1cEA–S1cEF Fig).
The logistic regression coefficients for the full-model given in Table 5 illustrate the differential roles each of the parameters play in explaining the dengue characteristics. β0 is found to be an important driver of the multi-annual signal. And in conjunction with β1 and αTRANS, it is the dominant factor for the absence of phase-locking. As can be expected, β1 is the main driver for reproducing a seasonal signature. The parameter for CI (ρ) interacts with β1 in reproducing this pattern and is thus also an important determining factor in fitting the MIPP. The R2-values for each of the regression models illustrate that the separate parameter values provide reasonable information about whether a characteristic is met or not. However, when assessing the simultaneous fit, the predictive power of the parameters is negotiated by interactions between the parameters and the separate characteristics. In particular the interactions between β1 and ρ govern simultaneous fitting (S3a Fig). These interactions are conserved when fitting the asymmetric 2-infection and symmetric 4-infection model (S3b and S3c Fig).
Strong, multi-level parameter interactions typically result in limited parameter identifiability. Indeed, the PCA reveals that, in particular the estimates for β1 and ρ are found to be little constrained by the characteristic patterns (Fig 5). The parameters β1 and ρ dominate the first two components, which explain the largest portion of the total variance in the passing parameter space (Gfull) (55%). While this observed lack of uniqueness may result from the limited influence the parameters have on replicating the dynamics and the substantial width of the criteria, complex interactions between patterns and parameters can also underlie this phenomenon. Indeed, as observed earlier, β1 and ρ are correlated with each other as well with other model parameters, which substantially impedes parameterization efforts (S3a Fig). Parameters β0, αSUS and αTRANS contribute equally to the smallest component, indicating that these are more constrained by the examined characteristics and the level of uncertainty and are less affected by dependence to other parameters (Fig 5). Allowing for asymmetry in transmission or tertiary and quaternary infections reduces the contribution of seasonality to the first component, leaving the duration of cross-immunity as the most important factor in explaining the variance in the passing parameter distributions (S5a and S5b Fig).
Fig 6 depicts the probability of achieving successful control (≤ 1outbreak in 30 years) as a function of w weeks of reduced transmission (e.g. due to implementation of vector control). The duration of control required to reach a desired probability of successful control can be used to quantify the level of resistance or vulnerability of a dynamical transmission system.
The inclusion of ADE or ADEx2 reduces the resistance of the model to perturbations (dark blue and pink lines), provided no CI is assumed (Fig 6). Including CI to the model offsets this effect and demonstrates a resistance profile similar to the base-model at longer control efforts, yet shows larger vulnerability at shorter durations of control. The exception is the full-model, which converges with the ADE-model at longer control durations.
The large resistance to control in the base-model is a consequence of the high values of R0 required for this model to meet the criteria (R0>2.2) (Fig 7A). At those levels of R0 the ADE-model demonstrates higher vulnerability to control as a result of decreased persistence (Fig 7C). The enhanced vulnerability of the ADE-model relative to the base-model as seen in Fig 6 is a consequence of low transmission rates. The inclusion of CI to either model enhances the resistance of the model especially at lower values of R0 (Fig 7D). Longer durations of cross-immunity are associated with greater resistance (S7DE Fig), while increased enhancement results in decreased resistance (S7CC and S7DC Fig).
This differential vulnerability is in part due to low infection persistence levels, a typical property of models with ADE only [12,15,23]. The addition of CI counters this effect with and without ADE (Fig 7C, 7D and 7F). This difference in infection persistence between CI and ADE systems, however, diminishes at high levels of seasonal forcing and R0. At these high transmission levels, both the models with CI (ADEx2+CI) and without CI (ADEx2) represent extreme fluctuations and long periods of non-persistent dynamics (S4aF and S4aG Fig). Thus, the differential model preference affects predicted control efforts more substantially in lower than higher seasonal scenarios.
We used a pattern-oriented modelling approach to test a range of multi-serotype models and parameter values for their ability to simultaneously replicate multiple dengue fever patterns derived from literature (Table 1) and case data from Trinidad and Tobago (Fig 1). Despite using such a multiple-pattern data fitting approach, we show that all the investigated model structures were effective at fitting each of the characteristic dengue patterns across some part of the model parameter space, suggesting the occurrence of equifinality, i.e. that observed infection patterns can be reproduced by more than one mechanism or combinations of mechanisms [60]. This implies that there could be multiple acceptable models for describing globally observed dengue dynamics, none of which can easily be rejected and therefore should all be considered in assessing the mechanisms determining disease transmission [61–63]. Three major efforts that would help disentangle the dominant drivers of dengue are: 1) better estimates of model parameters, in particular the duration of cross-immunity and the strength of seasonal forcing; 2) improved understanding on the contribution of post-secondary infections to dengue transmission dynamics; and 3) additional, more detailed patterns, such as (i) time series of serotype-specific dengue cases and (ii) levels of sero-prevalence in populations. Some of these patterns may well differ across geographic regions.
Based on the sizes of the passing parameter distributions, a preference for the most complex 2-infection model was apparent (Table 4). Remarkably, the model that performs best across all models is the 4-infection model with CI only. This indicates that, in some instances, the use of multiple patterns for model selection can help filter out overly specialized models and fetch simple, more generalized models that perform better across different scales. Additionally, it helps reveal the impact of simplifying assumptions on model selection and parameterization, i.e. allowing for quaternary infections enables us to reveal a simpler model framework that outcompetes its 2-infection equivalent. Also, it sheds new light on the need for ADE in replicating dengue dynamics. The role of ADE is not supported when allowing quaternary and tertiary infections while it is preferred in the 2-infection case, with and without asymmetry in transmission rates.
The performance of the base-model is noteworthy, given that it does not include the explicit serotype interactions deemed necessary to replicate asynchronous serotype oscillations. However, there are two implicit serotype interactions that likely underlie this behaviour. First, in the 2-infection model, serotypes affect each other’s dynamics by causing complete immunity to all serotypes after recovery from the second infection. The observed collapse in model fit of the base-model when we allowed for tertiary and quaternary infections supports this hypothesis. However, the 4-infection base-model also generates desynchronized behaviour of serotypes albeit in a very sparse region of the parameter space. This may result from the other implicit serotype interaction as a result of constraining individuals from acquiring more than one infection at the same time. In other words, this second type of interaction arises because individuals infected with one serotype are cross-immune to the remaining serotypes for the duration of the infectious period. This interaction may be enough to underlie a few, sparse fits across the parameter space. Indeed, when the model is extended to include more than one concurrent infection, the out-of-sync oscillations observed in the 4-infection base-model disappear (S6 Fig).
An additional result revealed by the POM-approach is that model preference appears to be governed by the level of seasonal fluctuations. Namely, the support for models with CI is larger in low seasonal settings, whereas the inclusion of decomposed ADE is required to reproduce the observed dengue patterns in the presence of strong seasonal fluctuations (Fig 4B). However, when tertiary and quaternary infections are allowed, this pattern disappears and all models apart from the base-model reveal similar median values for seasonal forcing (S2bB Fig). Additionally, we observe that the estimates for the duration of cross-immune period differ markedly upon inclusion of ADE or when relaxing the two infection assumption. In fact, without the inclusion of ADE, the CI-only 2-infection model does not encapsulate the best estimate of the duration of the cross-immune period, as proposed by Reich et al. [21]. The CI-model in the 4-infection framework, does meet the values estimated. These findings highlight that improved understanding of the extent to which post-secondary dengue infections contribute to overall dengue transmission, may greatly aid in disentangling the dominant drivers of dengue dynamics.
The public health importance of knowing the processes governing dengue transmission in a specific setting is highlighted by our results on achieving transmission interruption by vector control. The results indicated that the vulnerability of the models to disruption in transmission at equal levels of R0, was driven by the immune interactions incorporated in the model, with CI increasing resistance in low transmission settings, while ADE has the opposite effect. It is common practice to favour the most parsimonious model when the candidate models are equally efficient; however, the differences in model resistance we found here suggest that it is prudent to be extra cautious while making such a decision. Given their decisive role in selecting and quantifying the predominant mechanisms as well as determining the projected effects of interventions, in addition to R0-estimates, obtaining improved, localised estimates of seasonal forcing and the duration of cross-immunity should be prioritized towards better-informing modelling endeavours.
While efforts to disentangle the extent to which internal and external drivers influence the dynamics of multi-serotype systems have been made [64], adequately incorporating both the complex serotype interactions as well as the effects of coupling and decoupling between seasonal forcing and incidence remains an important issue. This is more so because long time series for serotype-specific incidence and vector abundance are scarce and case data are distorted by misclassification and underreporting. The core of the POM approach lies in the appreciation that single data patterns (e.g. multi-annual signals) usually do not contain enough information to unambiguously identify the mechanism generating such patterns; additional patterns from data are needed to fit several model responses simultaneously [65]. As pointed out above, we have shown here that, even with sparse data and relatively wide criteria, POM can be a useful tool to distinguish between different conceptual models for capturing dengue dynamics and assessing their vulnerability to control.
While the use of multiple patterns enhances the process of model selection greatly, it is not always clear whether a model capable of replicating the observed patterns can react realistically to environmental perturbations. This may especially be pertinent here as the models are fitted to macroscopic data using the average behaviour of the dynamical system rather than lower level processes [24]. While the proposed framework could be extended to incorporate additional, lower level patterns, such as serotype driven variation in disease severity, age-distributions of sero-prevalence, or age at first infection, these are likely to vary across regions and would greatly enhance the parameter dimension to be studied, diminishing the transparency and insights gained into the distribution and behaviour of model parameters which is our main focus. Similarly, matching to multiple patterns may not be sufficient to overcome the suspicion that the models demonstrate unrealistic resistance to control, as over 40 weeks of interrupted transmission is required to bring about an 80% probability of success (Fig 6). The import factor prevents the models from showing unviable dynamic behaviour that results from unrealistically low levels of infections innate to ODE-systems in general and especially prevalent in models with ADE(x2), yet also enhances the resistance of models. While the absolute levels of control are thus of limited practical use, the overall conclusion of differential resistance is found to persist across models with a lower import factor as well (S8 Fig), highlighting a fundamental challenge arising from structural model uncertainty.
The criteria derived and used in this work may be subjective. By basing the criteria on current literature and the available data and keeping the characteristics broad, we aimed to limit such subjectivity. By focussing on patterns that are common across endemic regions, the derived patterns are inherently weaker than for a localised approach, yet the outcomes are more generalizable. The broadness of the characteristics does lead to decreased uniqueness (as model fits to dengue patterns can be found across the entire parameter space) [66] and a wide range of model behaviours (S4a–S4c Fig) [67]. To reduce subjectivity, we have used uniform distributions bounded by ranges informed by literature. For model calibration, too restricted ranges may underestimate the level of uncertainty around a parameter value, whereas in model selection, the proportion of passes is sensitive to the width of the range. Also, the comparison between the models with different numbers of sampled parameters has underlying difficulties. In more complex models, the passing parameter space may be underrepresented, giving rise to a local decrease in likelihood and wider parameter bounds [68]. However, given the small number of parameters and large number of parameter combinations examined, the severity of under-sampling in this exercise is limited. Finally, caution should be taken in judging the likelihood of models based on the number of passes, as no correction is made for the differential complexity between the models.
The six models examined were chosen based on their proven performance in the literature [13,15,19]. However, the models contain some inherent limitations. The limited persistence typical in highly seasonal models with (decomposed) ADE may in part result from the lack of stochasticity in the model [12,23]. Serotype persistence is also believed to be affected by the assumed symmetry in transmission rate and or virulence between serotypes [17]. We indeed observe less wild fluctuations upon the inclusion of asymmetry and a consequential increase in the fit of models with ADE (S4b Fig). Further, the inclusion of explicit vector dynamics has been found to increase the robustness of the system to changes in cross-immunity and ADE parameters, resulting in a larger parameter space with regular (1–2 year inter-epidemic periods) dynamics and moderate amplitude fluctuations [69]. Therefore, including vector population dynamics may affect the quantitative conclusions of this study, especially when high seasonal fluctuations are assumed. The inclusion of explicit vector dynamics would further allow for a more quantitative assessment of required control efforts, which will be a focus of future work.
Lastly, no long-term variation in parameter values was taken into account. Yet, fertility rates have decreased and life expectancy has gone up in most dengue endemic countries over the last decades [70]. Cummings et al. showed that a decrease in birth rate might result in a decrease in the force of infection and increase in the mean age of infection [71]. The same authors also demonstrate that this demographic shift may have induced prolonged multiannual oscillations [71]. Additionally, vector control has intensified over the years with varying success [72]. The on-and-off vector control is likely to act as a distorting factor in the estimation of the role of seasonality, as the climate driven signal in the incidence data may be weakened by these control measures. Therefore, ignoring on-going control measures may have had some influence in our model selection and predictions. Further research will focus on disentangling the complex interplay of dengue dynamics with non-stationary factors such as intervention efforts, demography and climate.
With the expanding spatial spread of dengue and the increase of frequency and size of outbreaks, understanding dengue disease dynamics and the consequences of control efforts (e.g. a near-future vaccine introduction) has become critically important. Indeed, the present work stresses that ignoring model uncertainty in prediction exercises can skew the impact of vector control substantially. It also emphasizes that the wider use of improved data-model assimilation approaches, such as the POM method, could play a significant role in overcoming this problem.
|
10.1371/journal.pbio.2000598 | Authorization of Animal Experiments Is Based on Confidence Rather than Evidence of Scientific Rigor | Accumulating evidence indicates high risk of bias in preclinical animal research, questioning the scientific validity and reproducibility of published research findings. Systematic reviews found low rates of reporting of measures against risks of bias in the published literature (e.g., randomization, blinding, sample size calculation) and a correlation between low reporting rates and inflated treatment effects. That most animal research undergoes peer review or ethical review would offer the possibility to detect risks of bias at an earlier stage, before the research has been conducted. For example, in Switzerland, animal experiments are licensed based on a detailed description of the study protocol and a harm–benefit analysis. We therefore screened applications for animal experiments submitted to Swiss authorities (n = 1,277) for the rates at which the use of seven basic measures against bias (allocation concealment, blinding, randomization, sample size calculation, inclusion/exclusion criteria, primary outcome variable, and statistical analysis plan) were described and compared them with the reporting rates of the same measures in a representative sub-sample of publications (n = 50) resulting from studies described in these applications. Measures against bias were described at very low rates, ranging on average from 2.4% for statistical analysis plan to 19% for primary outcome variable in applications for animal experiments, and from 0.0% for sample size calculation to 34% for statistical analysis plan in publications from these experiments. Calculating an internal validity score (IVS) based on the proportion of the seven measures against bias, we found a weak positive correlation between the IVS of applications and that of publications (Spearman’s rho = 0.34, p = 0.014), indicating that the rates of description of these measures in applications partly predict their rates of reporting in publications. These results indicate that the authorities licensing animal experiments are lacking important information about experimental conduct that determines the scientific validity of the findings, which may be critical for the weight attributed to the benefit of the research in the harm–benefit analysis. Similar to manuscripts getting accepted for publication despite poor reporting of measures against bias, applications for animal experiments may often be approved based on implicit confidence rather than explicit evidence of scientific rigor. Our findings shed serious doubt on the current authorization procedure for animal experiments, as well as the peer-review process for scientific publications, which in the long run may undermine the credibility of research. Developing existing authorization procedures that are already in place in many countries towards a preregistration system for animal research is one promising way to reform the system. This would not only benefit the scientific validity of findings from animal experiments but also help to avoid unnecessary harm to animals for inconclusive research.
| Scientific validity of research findings depends on scientific rigor, including measures to avoid bias, such as random allocation of animals to treatment groups (randomization) and assessing outcome measures without knowing to which treatment groups the animals belong (blinding). However, measures against bias are rarely reported in publications, and systematic reviews found that poor reporting was associated with larger treatment effects, suggesting bias. Here we studied whether risk of bias could be predicted from study protocols submitted for ethical review. We assessed mention of seven basic measures against bias in study protocols submitted for approval in Switzerland and in publications resulting from these studies. Measures against bias were mentioned at very low rates both in study protocols (2%–19%) and in publications (0%–34%). However, we found a weak positive correlation, indicating that the rates at which measures against bias were mentioned in study protocols predicted the rates at which they were reported in publications. Our results indicate that animal experiments are often licensed based on confidence rather than evidence of scientific rigor, which may compromise scientific validity and induce unnecessary harm to animals caused by inconclusive research.
| Reproducibility is a fundamental principle of the scientific method and distinguishes scientific evidence from mere anecdote. The advancement of basic as well as applied research depends on the reproducibility of the findings, and can be seriously hampered if reproducibility is poor. However, accumulating evidence indicates that reproducibility is poor in many disciplines across the life sciences [1]. For example, in a study on microarray gene expression, only 8 out of 18 studies could be reproduced [2]; Prinz and colleagues [3] found large inconsistencies (65%) between published and in-house data in the fields of oncology, women’s health, and cardiovascular diseases; oncologists from Amgen could confirm only 6 out of 53 published findings [4]; and, of more than 100 compounds that showed promising effects on amyotrophic lateral sclerosis (ALS) in preclinical trials, none displayed the same effect when retested by the ALS Therapy Development Institute in Cambridge [5]. Besides a waste of time and resources for inconclusive research [6–8], however, poor reproducibility also entails serious ethical problems. In clinical research, irreproducibility of preclinical research may expose patients to unnecessary risks [9,10], while in basic and preclinical animal research, it may cause unjustified harm to experimental animals [11].
Reproducibility critically depends on experimental design and conduct, which together account for the internal and external validity of experimental results [12]. External validity refers to how applicable results are to other environmental conditions, experimenters, study populations, and even to other strains or species of animals (including humans) [12]. Thus, it also determines reproducibility of the results across replicate studies (i.e., across different labs, different experimenters, different study populations, etc.) [11,13,14]. Internal validity refers to the extent to which a causal relation between experimental treatment and outcome is warranted, and critically depends on scientific rigor, i.e., the extent to which experimental design and conduct minimize systematic bias [12,15]. It has been suggested that poor internal validity due to a lack of scientific rigor may also be a major cause of poor reproducibility in animal research [16–18].
There are various sources of bias (e.g., selection bias, performance bias, detection bias), and specific measures exist to mitigate them (e.g., randomization, blinding, sample-size calculation; [12,15,19,20]). To assess the internal validity of studies, e.g., in the peer review process, and to facilitate replication of studies, publications must contain sufficiently detailed information about experimental design and conduct, including measures taken against risks of bias [20,21]. However, systematic reviews generally found a low prevalence of reporting of measures against risks of bias (further referred to as reporting) in animal research publications. Thus, reporting ranged from 8% to 55.6% for allocation concealment, from 3% to 61% for blinded outcome assessment, from 7% to 55% for randomization, and from 0% to 3% for sample size calculation [19,22–29].
Low rates of reporting have been interpreted as evidence for a lack of scientific rigor (e.g., [20]). Indeed, several systematic reviews found correlations between poor reporting and overstated treatment effects [19,29–31]. Reporting guidelines have thus become a major weapon in the fight against risks of bias in animal research [32]. However, although the ARRIVE guidelines (Animal Research: Reporting of In Vivo Experiments) by the United Kingdom-based organization NC3Rs (National Centre for the Replacement, Refinement & Reduction of Animals in Research) have been endorsed by over 1,000 journals, this did not lead to a substantial improvement of reporting in animal studies [33]. Nevertheless, awareness seems to rise, as Macleod and colleagues [28] recently found that reporting increased over the past decades, although there is still considerable scope for improvement.
Research on the internal validity of animal experiments has focused mainly on reporting in scientific publications. However, most published research has undergone peer review when submitted for funding, and in some countries (e.g., Switzerland, Germany), individual animal experiments are licensed by national or regional authorities. For example, in Switzerland, the licensing of animal experiments is based on an explicit harm–benefit analysis, whereby any harm imposed on the animals is gauged against the expected benefit (gain of knowledge) of the experiment. Because the gain of knowledge critically depends on the scientific validity of the findings, risks of bias may affect the weight attributed to the expected benefit of a study in the harm–benefit analysis. An accurate harm–benefit analysis thus depends on information regarding risks of bias and measures used to mitigate them.
In the present study, we therefore screened applications for animal experiments submitted to the cantonal authorities in Switzerland (n = 1,277) for evidence of the use of measures to avoid risks of bias, and compared the rates at which these measures were described in applications (for reasons of simplicity hereafter also referred to as reporting) with the rates of reporting of the same measures in a representative sub-sample of publications (n = 50) resulting from experiments described in these applications. This allowed us, for the first time, to compare evidence of scientific rigor available to the authorities when licensing animal experiments with the evidence reported in scientific publications, and to assess whether poor reporting in the scientific literature is predicted by poor reporting in applications for experiments.
Our database included a final sample of 1,277 applications for animal experiments approved by the cantonal authorities of Switzerland in the years 2008, 2010, and 2012, respectively. Evidence of scientific rigor was assessed based on seven common measures against risks of bias: allocation concealment, randomization, blinded outcome assessment, sample size calculation, inclusion and exclusion criteria, primary outcome, and a statistical analysis plan (S2 and S3 Texts). Besides analyzing each item separately, we also calculated an internal validity score (IVS; see Eq 1), which served as the primary outcome variable for the statistical analysis of effects of various study descriptors on rates of reporting. In addition, we calculated an accuracy score (AS; see Eq 2) based on six items of information explicitly asked for on the application form as a measure of how accurately the applicants had filled out the application form to control for effects of accuracy on the IVS.
Reporting rates were generally very low (Table 1); on average, less than one out of the seven items were reported in applications for animal experiments, with reporting rates varying among the seven items, ranging from 2.4% for the statistical analysis plan to 18.5% for the primary outcome variable (Table 1). However, reporting rates greatly differed between individual applications, with the IVS ranging from 0 (i.e., 0/7 items reported) to 0.857 (i.e., 6/7 items reported), whereby 711 out of the 1,277 applications (55.68%) scored 0 (S1 Fig).
We hypothesized that reporting rates and, thus, the IVS might depend on various characteristics of the studies, including the year of authorization (Year), the types of animals used (Species), the severity of the experimental procedures (Severity), the institution conducting the study (Institution), the canton authorizing the study (Canton), and the language in which the application was written (Language), as well as the AS of the application. Generalized linear models in a Bayesian information criterion selection process were used to identify which of the study descriptors best described our data, indicating that they were most likely to have influenced the IVS. The best fitting model included Year, Canton, Language, Institution, and the interaction between Species and AS (see Eq 4). According to the model output (S1 Data), however, none of the individual descriptors had a significant effect on the IVS except Language, as applications written in German had a significantly higher IVS compared to applications written in English (odds ratio [OR] = 0.79, 95% confidence interval [CI] = 0.64–0.98) and applications written in French (OR = 0.46, CI = 0.32–0.65), and the interaction between farm animals and AS (OR = 168.24, CI = 1.17–2,5571.31). Thus, below we report trends that were observed regarding effects of the descriptors that were included in the final model on the IVS.
The IVS was similar across all three years of authorization: 2012 (median = 0.0, range: 0–0.71), 2010 (0.0, 0–0.71), and 2008 (0.0, 0–0.85). At the level of individual items, trends of improvement across years were observed in the reporting rates of blinding, sample size calculation, and statistical analysis plan (Fig 1B–1E). While there was some variation in IVS across cantons, canton did not seem to have a strong effect (Fig 2). Among the different research institutions, academic institutions (i.e., universities, federal institutes of technology, or university hospitals) accounted by far for the largest part of applications, with 972 (76%) applications compared to 87 (7%) from industry, 56 (4%) from governmental institutions, and 162 (13%) from other private institutions. Overall, academic institutions (0.0, 0–0.86) tended to score lower on IVS than institutions from industry (0.14, 0–0.57), governmental institutions (0.14, 0–0.71), and other private institutions (0.14, 0–0.57; Fig 3A). At the level of individual items, similar trends were observed in the reporting rates of randomization and sample size calculation (Fig 3B–3E). There was also variation in IVS depending on the species of animals used (Fig 4A). Thus, applications for experiments on “higher” mammals (i.e., cats, dogs, rabbits, and primates [CDRP]) tended to score higher (0.17, 0–0.71) compared to experiments on farm animals (0.14, 0–0.86), other mammals (0.15, 0–0.29), laboratory rodents (0.0, 0–0.71), and non-mammals (0.0, 0–0.6), respectively. A similar trend was observed in the reporting rates of blinding, randomization, sample size calculation, and statistical analysis (Fig 4B–4E). Thus, applications for experiments on CDRP as well as farm animals scored higher compared to those involving laboratory rodents and non-mammals, while data from applications for experiments involving other mammals varied widely due to the small sample size (n = 8).
In contrast to the IVS, the AS was generally high, with a median score of 0.8, ranging from 0.11 to 1.00. Despite the low IVS and more than half of the applications scoring 0, there was a weak but positive correlation between AS and IVS (Spearman’s rho = 0.17, p < 0.001; Fig 5).
In order to ensure reliability of the data between the two investigators (TSR, LV) as well as across time, inter-rater and intra-rater reliability tests were conducted at regular intervals. Inter-rater reliability scores (see Eq 3) of the IVS ranged from 91.4% to 97.1%, while the respective intra-rater reliability scores ranged from 87.1% to 95.7% for TSR and from 94.3% to 97.1% for LV. Similarly, inter-rater reliability scores of the AS ranged from 91.3% to 96.3%, while the respective intra-rater reliability scores ranged from 87.5% to 97.5% for TSR and from 92.5% to 98.8% for LV (see S2 Data).
In order to relate the reporting rates obtained from applications for animal experiments to reporting rates found in the scientific literature, we selected 50 publications originating from 50 independent applications in our sample, screened them for the same seven internal validity criteria, and calculated the IVS for each publication using the same method.
Similar to what we found for applications, reporting rates in the 50 publications were generally low, albeit slightly higher than in the applications (Fig 6), resulting in a median IVS of 0.14. Reporting rates for the seven items ranged from 0% for sample size calculation to 34% for the statistical analysis plan. Again, reporting rates differed greatly between individual publications, with IVS ranging from 0 to 0.6, whereby 23 out of 50 publications (46%) scored 0.
Except for sample size calculation and the primary outcome variable, reporting rates for individual items were higher in publications than in applications (see Fig 6). Whereas IVS of applications and publications were the same in 27 cases (of which 21 scored 0), it was higher in 18 pairs (which was due to a statistical analysis plan in 12 cases) and lower in five cases. This increase was corroborated by a weak positive correlation between the IVS of applications and that of publications (Spearman’s rho = 0.34, p = 0.014).
Due to the smaller sample size, not all descriptors assessed for their effects on the IVS of applications could be analyzed here. Instead, we analyzed publication-specific descriptors, namely whether or not the journal in which the study was published had endorsed the ARRIVE guidelines and the impact factor of the journal (IF). There was no significant effect of ARRIVE on IVS (yes: median = 0.14, range: 0 to 0.57; no: median = 0, range: 0 to 0.60; p = 0.69; Fig 7A). In contrast, IF had a significant negative effect on IVS (Spearman’s rho = -0.49, p < 0.001; Fig 7B).
Based on the low reporting rates in publications of animal research and evidence suggesting that poor reporting may reflect a lack of scientific rigor [19,29–31], this study examined whether poor reporting in the scientific literature is predicted by poor reporting in applications for animal experiments, that is before the studies have actually been conducted. The study was restricted to animal experiments authorized in Switzerland for two reasons. First, Switzerland has an authorization system for animal experiments that requires detailed description of study protocols for every planned study. These study protocols form the basis of the harm–benefit analysis upon which the decision for or against authorization of individual studies is based. Second, the study was facilitated by the Swiss Federal Food Safety and Veterinary Office (FSVO) providing access to all applications for animal experiments via their online platform (e-tierversuche) through which scientists communicate with the authorities and submit their applications for animal experiments. Such unlimited access to application forms for animal experiments is unprecedented, and it is laudable that the FSVO supported this meta-research. This kind of support has notoriously proven difficult to obtain for reasons of confidentiality, as highlighted by Chan et al. [34], with respect to clinical trial protocols for meta-research. As described in the Materials and Methods, access to the application forms was possible without violating confidentiality.
We evaluated 1,277 applications for animal experiments and 50 publications derived thereof and found very low reporting rates in both applications and publications (Fig 6). Reporting rates in publications were within the range reported in previous studies (e.g., [19,20]). That reporting rates in applications were similar—even slightly lower—indicates that the authorities approving animal experiments are lacking important information about experimental conduct that may be critical for evaluating the expected benefit in a harm–benefit analysis. Risks of bias question the scientific validity of the results, which is a precondition for a study to achieve the expected benefit. Whether the authorities are unaware of risks of bias and measures to avoid them or whether they consider them as unimportant for the benefit of the research is unknown and warrants further study. As a result, however, animal experiments are authorized based on implicit confidence rather than explicit evidence of scientific rigor. Similarly, poor reporting in publications means that manuscripts are often accepted for publication in the absence of evidence of scientific rigor. This “trust me model” of science has been criticized before [1,35,36]. It sheds serious doubts on the current authorization procedure for animal experiments as well as the peer-review process for scientific publications, which in the long run may compromise the credibility of the research.
We found a weak positive correlation between the IVS of applications and that of the corresponding publications. This suggests that the reporting of bias avoidance measures in applications predicted, at least to some extent, the reporting of such measures in publications. If this reflects a consistent relationship, asking for more detailed information on experimental conduct in applications for animal experiments might help to promote better experimental conduct as well as better reporting in publications. Asking for more detailed information at the planning stage of the research might also reduce the danger of normative responses, whereby scientists simply satisfy the guidelines (e.g., ARRIVE) at a time when it is too late to take corrective actions on experimental conduct.
The increase in the IVS of publications compared to applications was largely due to better reporting of the statistical analysis plan (S2 Fig). This is likely due to the fact that journals (and reviewers) generally insist on a detailed description of the statistical analysis. It indicates that reporting guidelines (such as ARRIVE) could potentially increase scientific quality of animal research, if editors and reviewers helped to enforce them. However, as shown by Baker et al. [33] and confirmed by the present study (Fig 7A), this has not been the case so far; publications in journals having endorsed the ARRIVE guidelines did not score higher than publications in other journals. We also found a weak positive correlation between the accuracy of completing the application forms (AS) and the IVS. Thus, applicants who answered questions in the application form more accurately had a higher IVS. As shown by Minnerup et al. [37], this further confirms that enforcement of guidelines may be important in view of improving reporting standards.
In the final statistical model, language was the only descriptor having a significant effect on IVS of applications for animal experiments. Applications written in German had significantly higher IVS than applications written in English or French. Several explanations may account for this result. For example, the proportion of German native speakers may have been higher among authors of German applications; German may have been mostly used by native German speakers, while English may have been used by many non-native English speakers. Similarly, French may have been used by many non-native French speakers because, apparently, authorities in French-speaking cantons of Switzerland strongly encourage submission of applications in French (own observation). However, one might not necessarily expect language skills to affect such standardized terminology (randomization, blinding, etc.), but because these items are not explicitly asked for, applicants writing in their native language might be more likely to provide unsolicited detail. Alternatively, differences in regional policies of authorities between French- and German-speaking cantons, as well as the fact that all French applications were scored by only one experimenter (LV), may have contributed to this effect, but our data do not allow us to examine these explanations further.
Apart from language, all other explanatory variables in the final model had only weak effects on IVS that did not reach statistical significance (S1 Data). For example, there was a weak tendency for the reporting rates of blinding, sample size calculation, and statistical analysis to be higher in 2012 compared to those from previous years (Fig 1). This trend might reflect increasing awareness by both researchers and authorities of the importance of reporting, and it is consistent with recent evidence from a random sample of life sciences publications [28]. However, despite the many systematic reviews revealing flaws in experimental design and conduct since Ioannidis’ seminal opinion paper [38], and the wealth of solutions that have since been proposed [2,5,32,39], little progress has been made. Like Baker et al. in 2014 [33], we did not find convincing evidence that reporting had increased from applications authorized before (2008) to those authorized after (2012) publication of the ARRIVE guidelines. Again, the main reason for this might be a lack of enforcement of these guidelines by authorities as well as journal editors. However, our sample was mostly based on studies designed and authorized before the ARRIVE guidelines became widely known. That the endorsement of the ARRIVE guidelines had no effect on the IVS of publications may thus reflect the delay in such a change taking effect.
Recent evidence indicated that industry-sponsored research is less biased than academic research [40]. We therefore predicted higher rates of reporting of measures against risks of bias in applications from private compared to academic institutions. Although there was a weak tendency for applications from academic institutions to score lower on IVS compared to governmental and private institutions, we cannot exclude random variation as the source of this trend. If true, however, it might reflect the different incentives between institutions, favoring more conservative approaches in non-academic institutions [41].
An interesting tendency was found in relation to the type of animals being used. Thus, applications for experiments on CDRP, farm animals, and other mammals had slightly higher IVS than those for experiments on lab rodents and non-mammals. CDRP and, to a lesser extent, farm animals and other mammals may benefit from the attribution of a higher moral status, e.g., because they are close relatives (primates), social partners (dogs, cats, rabbits), or otherwise elicit more compassion (farm animals, other mammals) than lab rodents (that are also considered as “pest” species) and non-mammals (mostly fish; e.g., [42,43,44]). On the one hand, this might indicate that applications are assessed more carefully when the stakes are perceived as morally high, although it would remain unclear whether this effect is due to the applicants providing more information or to the authorities asking for more. On the other hand, IVS was low throughout, and the difference between species categories was not significant. In addition, there was no such trend with increasing degree of severity of studies. Importantly, however, the Swiss Animal Welfare Act does not provide a legal basis for such “speciesism” among vertebrates, and both authors and authorities should treat all vertebrates equally.
Finally, we found a weak but significant negative relationship between the IVS of publications and the IF of the journal in which it was published. That the journal IF does not necessarily reflect the quality of research has long been known (e.g., [45]), and a systematic review of a random sample of life sciences publications recently found no evidence for a positive relationship between IF and reporting [28]. Across the whole range of journal IF in our sample of publications, IVS of 0 clearly prevails, confirming that poor reporting of measures against risk of bias is common throughout the scientific literature.
According to the Animal Protection Index (API) by World Animal Protection, Switzerland (together with the United Kingdom, Austria, and New Zealand) ranked top in an international comparison of animal protection policy among 50 countries (http://api.worldanimalprotection.org/). In particular, authorization of animal experiments is based on a harm–benefit analysis, and authorization is denied if, in relation to the anticipated gain in knowledge, they inflict disproportionate harm on the animals (Article 19(4), [46]). Because the anticipated gain in knowledge critically depends on experimental design and conduct, the lack of information on measures against risks of bias in applications means that, in Switzerland, authorization of animal experiments is based on implicit confidence rather than explicit evidence of scientific rigor.
Several arguments may be held against this interpretation of our results, namely (i) that the measures against risks of bias assessed here are not important determinants of scientific validity, (ii) that they are not explicitly asked for on the application form for animal experiments, (iii) that, as the system currently works, it is not the authorities’ duty to assess the scientific validity of the experiments, and (iv) that the authorities’ confidence in scientific rigor is well justified. First, it is certainly the case that the authorities assess the scientific rationale underlying the proposed studies, thereby assessing several important aspects of scientific validity, although these are not specified explicitly. Also, there may be other, even more important risks of bias (e.g., use of inappropriate control group) that were not included in our evaluation. However, all seven items included here are considered as relevant measures against risks of bias that may compromise scientific validity in important ways; they have therefore been included in reporting guidelines such as the ARRIVE guidelines. Second, while it is also true that the application form does not explicitly ask for allocation concealment, randomization, blinding, and inclusion or exclusion criteria, it does ask explicitly for the primary and secondary outcome variables, sample size calculation, and a detailed statistical analysis plan. Moreover, the first example of how to describe procedures presented in the explanatory notes to the application form by the FSVO starts with “The dogs are divided randomly into 3 groups,” indicating that randomization is also considered a relevant aspect of the description of procedures. Even if only those measures explicitly asked for on the application form were enforced, all applications would score IVS ≥ 0.42 (i.e., 3/7). Third, authorities may argue that it is the peers’ duty to assess and guarantee scientific rigor, while the authorities’ duties (and those of their advisory committees) should be limited to assessing the scope for applying the 3Rs (replacement of animal experiments, reduction of animal use, and refinement of procedures) and whether the expected benefits (as declared by the applicants) outweigh the harms inflicted on the animals. However, it is important to note that not all experiments are based on project proposals that have undergone scientific peer review (e.g., most applications from the private sector), and that peer review does not seem to guarantee good scientific practice [47]. Finally, whether the authorities’ implicit confidence in the scientific validity of the results of licensed experiments is justified is an empirical question. Concerns that such confidence may not be warranted is largely based on studies showing a negative relationship between reporting of measures against risks of bias and inflation of treatment effect size in preclinical studies (e.g., [19,25]). Together with accumulating evidence of poor reproducibility of in vivo research, these findings have shed doubts on the quality of experimental design and conduct. However, there is clearly a need for more research on the actual implementation of measures against risks of bias in experimental animal research. We have recently conducted an online survey amongst all Swiss animal researchers to elucidate actual implementation of the same seven measures against risks of bias assessed here. Our findings suggest that although reporting rates found in the literature tend to underestimate actual implementation of these measures, there is considerable scope for improvement [48].
Lack of scientific rigor in experimental conduct is widely considered to be an important determinant of poor reproducibility of in vivo research [16,17,18]. However, this assumption is based on the indirect evidence outlined above, and has never been tested directly. Randomization, blinding, sample size calculation, and all the other measures against risks of bias assessed here mainly affect the internal validity of experiments. Although the reproducibility of results can be affected by the internal validity of studies, reproducibility depends more on the external validity of studies [11–13]. Reproducibility may thus be enhanced mainly by using design features aimed to increase the external validity of results, such as more heterogeneous study populations, independent replicate cohorts, or multicenter study designs [14,49,50]. Thus, there is also a need for more research on the relative contribution of experimental conduct and experimental design, respectively, to the reproducibility of results.
Last, but not least, besides experimental design and experimental conduct, several other factors introduce bias into the scientific literature, in particular “hypothesizing after results are known” (HARKing, [51]), p-hacking [52], selective reporting [53], and publication bias [54]. The most effective way of eliminating all of these biases would be prospective registration of preclinical animal experiments similar to preregistration of clinical trials [55]. Further research is certainly needed on how to facilitate practical implementation of preregistration in the face of several contentious issues such as confidentiality, property rights, and theft of ideas. However, the authorization procedure for animal experiments already in place in Switzerland (and other countries, e.g., Germany), provides an ideal basis for implementing preregistration of animal experiments, which would not only benefit the scientific validity of results from animal experiments but also minimize unnecessary harm to animals for inconclusive research. By this, Switzerland could consolidate its position as a leader in animal protection as well as extend its leadership to scientific rigor.
Applications for animal experiments (Form A, S1 Text) were selected from an anonymized database obtained from the FSVO, containing all applications submitted in Switzerland since 1983. Access to applications archived by the FSVO was based on a contract between the FSVO and the authors of this study, which guaranteed confidentiality to the applicants. Applications were selected based on predefined inclusion and exclusion criteria. Thus, only new applications submitted during the years 2008, 2010, and 2012 were included, of which applications related to (i) diagnosis of disease, (ii) education and training, and (iii) the protection of humans, animals, and the environment by toxicological or other safety tests required by law were excluded a priori (S3 Fig). A total of 1590 applications met these criteria and were subjected to formal screening.
In order to assess risks of bias in the experiments described in the applications, a checklist was elaborated (S2 Text) based on checklists used in previous studies assessing the use of measures to reduce risks of biases as reported in the published literature [19,20,56]. We restricted our checklist to items that (i) are essentially applicable to all kinds of experimental studies and (ii) can be assessed objectively without specific expertise of the research topic, and included those seven items that we encountered most often in the literature: (1) allocation concealment, (2) blinded outcome assessment, (3) randomization, (4) formal sample size calculation, (5) inclusion and exclusion criteria, (6) a primary outcome variable, and (7) a statistical analysis plan. These seven items were also used to calculate an IVS based on the number of items that were reported in the application divided by the total number of items applicable to the study (max = 7).
Additional items were assessed that were, however, not included in the IVS. These included additional aspects of study conduct (blinded conduct of study, randomized conduct of study, termination criteria, references for the sample size, and general statements on statistical analysis; S2 Text).
In addition, we assessed the accuracy with which the application forms (Form A) were filled out, using items that were explicitly asked for on Form A, and for which the content to be filled in was explicitly specified in the accompanying guidelines to Form A on the FSVO webpage (https://www.blv.admin.ch/dam/blv/en/dokumente/tiere/publikationen-und-forschung/tierversuche/erlaeuterungen-form-a.pdf.download.pdf/erlaeuterungen-form-a.pdf). Furthermore, we chose items that are relevant for the harm–benefit analysis and could be determined with high reliability. The following six items were included: (1) description and justification of the methods used (e.g., by indicating references, previous results, or results from a pilot study); (2) information about the identification of individual animals; (3) the total number of animals used, the number of treatment groups, and the number of animals per treatment group; (4) reference to a score sheet for the assessment of animal welfare; (5) the degrees of severity for all animals involved in the experiments; and (6) the fate of the animals at the end of the experiments. These six items were used to calculate an AS based on the number of items reported divided by the total number of items applicable to the study (max = 6).
The AS was constructed as a control measure, to control for variation in IVS induced by variation in the accuracy with which the form was filled out. Both IVS and AS were assessed by scoring whether or not the respective items were reported in any of the experiments included in an application form. Thus, a “YES” was recorded if an item was reported in at least one of the described experiments and a “NO” if an item was either not reported at all or if it was unclear. If an item was not applicable to the experiment described in the application form, “NA” was recorded (more details are given in the S3 Text).
The 1590 applications were randomly allocated to two investigators (LV, TSR) for formal screening (leading to two lists of 795 applications each, one for each investigator). During screening, 94 applications were excluded because they were either incomplete or not available in the archives of the FSVO. A further 36 applications were excluded because they met one or more of the exclusion criteria reported above. This left 1,460 applications that were deemed suitable for screening. Applications written in French (n = 423) or Italian (n = 5) were screened by the investigator with better knowledge of these languages (LV), regardless of their assignment to the two investigators, while applications written in German (n = 430) or English (n = 602) were screened according to their assignments to the two investigators. Therefore, a total sample of n = 935 was screened by investigator LV while a total sample of n = 525 applications was screened by investigator TSR.
To restrict analysis to experimental in vivo studies, a further 183 applications were excluded in the course of the screening process because they referred to in vitro studies (if the animals were killed before the experimental treatment was applied; n = 106), monitoring studies (if the animals were observed in the wild; n = 28), or other exceptions (e.g., breeding studies, post-mortem studies; n = 49), resulting in a final sample size of n = 1,277 applications used for analysis (see S3 Fig).
Based on information provided by the applicants on Form A and used for the annual statistics of animal use by the FSVO, we also recorded several descriptors that might influence the reporting of internal validity items; these included (i) year of authorization (2008, 2010, 2012), (ii) language (English, German, French), (iii) canton (the six largest cantons of Basel, Bern, Freiburg, Geneva, Vaud, Zurich, and the group of the remaining small cantons), (iv) type of institution (academic institutions [i.e., universities, federal institutes of technology, hospitals], industry, governmental institutions [national and cantonal], other [e.g., private institutions, foundations]), (v) animal species (laboratory rodents, higher mammals [CDRP], farm animals, other mammals, non-mammals), (vi) genetically modified animals (yes, no), and (vii) the prospective degree of severity of the planned procedures as defined by the FSVO (0, 1, 2, 3).
Prior to the screening of the selected Form A, two pilot studies on separate applications (i.e., applications authorized in 2009) were conducted to ensure the applicability of the checklist and to ensure consistency of scoring within and between investigators. To ensure consistent scoring of applications between the two investigators, both investigators screened the same 10 applications, and discrepancies were checked at the end of the day. Inter-rater reliability (Eq 3) was assessed at regular intervals (on day 1 and then after the 100th, 300th, 500th, and 700th application on the investigators’ list, respectively) by assessing the proportion of agreement between the two investigators. For this, the first five applications on each investigator’s list were screened by both investigators.
Only applications written in either German or English were used for inter-rater reliability tests. Overall, 50 applications were screened twice in the course of these inter-rater reliability tests. Inter-rater reliability never dropped below 85% (S2 Data).
To ensure that both investigators scored applications consistently over time, samples of 10 applications were re-scored at regular intervals (after 50, 150, 350, and 550 listed applications, respectively). In addition, each investigator conducted a final intra-rater reliability test on 10 randomly chosen applications from the whole list after completing the screening procedure. If systematic discrepancies would have occurred, the applications previously scored would have been re-scored. However, as in the case of inter-rater reliability, intra-rater reliability never dropped below 85% (S2 Data).
No a priori sample size calculation was performed, as all applications were included in our sample that fulfilled the inclusion/exclusion criteria. However, once the sample size was determined, we verified that it was suitable for the planned statistical analysis (see model description below).
The screening data from the checklists were transferred to a tabulating program (Microsoft Excel 2010.Ink, Redmond, WA, USA) and imported into the statistical software R [57].
We used descriptive statistics to represent reporting rates for individual criteria of internal validity (allocation concealment, blinded outcome assessment, randomization, sample size calculation, inclusion and exclusion criteria, primary outcome, and statistical analysis). Furthermore, influences of relevant descriptors (year, canton, institution, and animal species) were represented graphically, with median and mean IVS of the group, and overall mean IVS.
For the statistical analysis of the overall internal validity score of applications, we used generalized linear models to evaluate the influence of the a priori stated descriptors on the internal validity score. The analyses were performed in R [57] using the built in function glm with a binomial error distribution to account for the data structure (primary outcome as proportions). As a first step, we compared univariate models (model with one descriptor) with an intercept-only model (modelling the intercept of the internal validity score) based on significant (p < 0.05) likelihood ratio test of the package lmtest [58] in order to identify descriptors to be included in the further modelling process. The descriptors to be retained were language, canton, species category, institutions, authorization year, and accuracy of the application. In a second step, by means of an information theoretic approach to model selection using the Bayesian Information Criterion (BIC), we identified the model that best fit our data. For an automated model selection procedure, the package MuMln [59] with the function dredge was used to compare all models with all possible combinations of the retained descriptors (full model included also the interaction term for species category and accuracy; see Eq 4).
The dredge function ranks all descriptor combinations according to their BIC; the model with the lowest BIC was assumed to be the one representing our data best. The final model included the following main effects (descriptors): language (3 levels), cantons (7 levels), species category (5 levels), accuracy (continuous), institution (4 levels), and authorization year (3 levels). In addition to these main effects, the candidate model included the two-way interaction between species category and accuracy (corresponds to full model, cf. Eq 4). The model parameters were retrieved after correction for over dispersion (see S1 Data).
In order to relate the reporting rates of internal validity criteria assessed here by scoring applications for animal experiments with the reporting rates of such criteria in the published literature [19,22–28,60], we also scored a sub-sample of publications originating from studies based on applications in our study sample. These were identified by searching through grant numbers mentioned in the applications and references listed as output in the annual reports to the FSVO. For 155 applications (12.1%) we identified one or more corresponding publications. This number was reduced to 139 after excluding reviews and publications that were clearly unrelated to the study described in the applications (mismatch in animal species, general topic, or methods). This low number can be explained by the fact that studies licensed in 2012 and also many of those licensed in 2010 were not yet published, and that the search for publications had to rely on grant numbers mentioned in both application and publication (often grant numbers were not mentioned on applications) or on publications listed in the final reports required by the authorities upon completion of licensed studies (for most studies licensed in 2012 and also many of those licensed in 2010, final reports were not yet available).
For the comparison of the internal validity scores between applications and publications, we aimed to detect a medium effect size (0.3) with a statistical power of 0.8 at a significance level of p < 0.05. Based on this, we chose a sample size of n = 50, which allowed us to detect an effect size of 0.276 (G*power for correlations, bivariate normal model) [61]. A stratified random sampling procedure was used to select 50 publications from the 139 available publications, so as to select publications derived from a representative sample of all applications with respect to canton and type of animals used. Because this sample of publications was biased towards older applications, we compared the IVS of the sub-sample of 50 applications from which these 50 publications originated with the IVS of the entire sample of applications and found no significant difference; median IVS of the entire sample of applications (n = 1,277) was 0.0 (range 0 to 0.857), compared to 0.0 (range 0 to 0.714) for the sub-sample of applications (n = 50) from which the 50 publications were derived.
The publications were screened for reporting of internal validity criteria with a checklist containing the same seven internal validity criteria as were used for applications. The screening of all 50 publications was performed by one single investigator (LV). Publications were randomly allocated to one of the 10 d of screening (five publications per day). Days of screening were separated by two non-screening days. For the publications, descriptors were impact factor of the journal and endorsement of the ARRIVE guidelines by the journal. To determine the descriptors, the impact factor for the year of the publication as well as the ARRIVE status of the journal were assessed. If it was not possible to determine the ARRIVE status of a journal for the date of publication, given that all publications were published in 2012 or later, we used the ARRIVE status of the journal in 2015. Whether or not the ARRIVE status affected the internal validity score of publications was tested with a univariate generalized linear model (binomial error distribution), with IVS as dependent and the descriptor (endorsement of ARRIVE yes or no) as independent variables.
Whether or not the internal validity score of publications was correlated with the impact factor of the journal was investigated using a spearman rank correlation test.
To ensure that the investigator scored the publications constantly over time, an independent person randomly chose one publication per five publications screened (i.e., one per day of screening) for an intra-rater reliability test. The chosen publication was re-screened on the second following day. The reliability (Eq 3) never dropped below the threshold of 85%.
|
10.1371/journal.pgen.1005493 | The Dynamic Genome and Transcriptome of the Human Fungal Pathogen Blastomyces and Close Relative Emmonsia | Three closely related thermally dimorphic pathogens are causal agents of major fungal diseases affecting humans in the Americas: blastomycosis, histoplasmosis and paracoccidioidomycosis. Here we report the genome sequence and analysis of four strains of the etiological agent of blastomycosis, Blastomyces, and two species of the related genus Emmonsia, typically pathogens of small mammals. Compared to related species, Blastomyces genomes are highly expanded, with long, often sharply demarcated tracts of low GC-content sequence. These GC-poor isochore-like regions are enriched for gypsy elements, are variable in total size between isolates, and are least expanded in the avirulent B. dermatitidis strain ER-3 as compared with the virulent B. gilchristii strain SLH14081. The lack of similar regions in related species suggests these isochore-like regions originated recently in the ancestor of the Blastomyces lineage. While gene content is highly conserved between Blastomyces and related fungi, we identified changes in copy number of genes potentially involved in host interaction, including proteases and characterized antigens. In addition, we studied gene expression changes of B. dermatitidis during the interaction of the infectious yeast form with macrophages and in a mouse model. Both experiments highlight a strong antioxidant defense response in Blastomyces, and upregulation of dioxygenases in vivo suggests that dioxide produced by antioxidants may be further utilized for amino acid metabolism. We identify a number of functional categories upregulated exclusively in vivo, such as secreted proteins, zinc acquisition proteins, and cysteine and tryptophan metabolism, which may include critical virulence factors missed before in in vitro studies. Across the dimorphic fungi, loss of certain zinc acquisition genes and differences in amino acid metabolism suggest unique adaptations of Blastomyces to its host environment. These results reveal the dynamics of genome evolution and of factors contributing to virulence in Blastomyces.
| Dimorphic fungal pathogens including Blastomyces are the cause of major fungal diseases in North and South America. The genus Emmonsia includes species infecting small mammals as well as a newly emerging pathogenic species recently reported in HIV-positive patients in South Africa. Here, we synthesize both genome sequencing of four isolates of Blastomyces and two species of Emmonsia as well as deep sequencing of Blastomyces RNA to draw major new insights into the evolution of this group and the pathogen response to infection. We investigate the trajectory of genome evolution of this group, characterizing the phylogenetic relationships of these species, a remarkable genome expansion that formed large isochore-like regions of low GC content in Blastomyces, and variation of gene content, related to host interaction, among the dimorphic fungal pathogens. Using RNA-Seq, we profile the response of Blastomyces to macrophage and mouse pulmonary infection, identifying key pathways and novel virulence factors. The identification of key fungal genes involved in adaptation to the host suggests targets for further study and therapeutic intervention in Blastomyces and related dimorphic fungal pathogens.
| Blastomyces is a genus of a thermally dimorphic fungal pathogen, which is the etiological agent of blastomycosis, a lung infection that can become a systemic mycosis. In North America, Blastomyces is endemic in the Ohio and Mississippi river valleys, the Great Lakes region, and the St. Lawrence River [1]. Within Blastomyces, two lineages of B. dermatitidis have been recognized [2], with recent work providing evidence that one lineage is a distinct species, B. gilchristii [3]. Both species can infect humans, and vary in morphology, virulence and immune responses by the host. The primary mode of infection is inhalation of conidia and the subsequent conversion of these conidia into parasitic yeast [4,5]. Clinical manifestations range from asymptomatic infection to symptomatic disease and include pneumonia, acute respiratory distress syndrome, and a rapidly progressive dissemination involving multiple organ systems that is often fatal [5,6]. Diagnosis is often complicated by the similarity of symptoms to those of viral or bacterial respiratory infection and by the aforementioned variety of manifestations [7].
As a thermally dimorphic fungus, Blastomyces has the remarkable ability to switch between two different morphologies in response to external stimuli, predominantly temperature [5]. At 22–25°C, Blastomyces grows as septate hyphae that produce infectious conidia and at 37°C it grows as a budding yeast [8]. Blastomyces is part of a larger group of dimorphic fungal pathogens, including Histoplasma, Paracoccidioides, and Coccidioides, all belonging to the order Onygenales. The dimorphic fungi collectively are the most common cause of invasive fungal disease worldwide and account for several million infections each year [8]. Unlike opportunistic fungi, such as Candida albicans, Cryptococcus neoformans, or Aspergillus fumigatus, the dimorphic fungi can infect immunocompetent and immunocompromised hosts [6,9–11].
Previous work has shown that in Blastomyces, the temperature-dependent switch from hyphae to yeast along with upregulation of yeast-phase specific genes is critical for virulence [12–14]. The dimorphism-regulating kinase-1 (DRK1) promotes the temperature-dependent conversion from mold to yeast, and its deletion renders Blastomyces avirulent during experimental murine pulmonary infection [12]. The upregulation of yeast-phase specific genes, such as the Blastomyces yeast-phase specific gene 1 (BYS1) [15] and the Blastomyces adhesion-1 gene (BAD1) [13,14], is also important for the adaptive response of the yeast cells in the host environment. BAD1 is considered an essential virulence factor in Blastomyces, since it binds yeast cells to host tissue and impairs host immune defenses by inhibiting the production of tumor necrosis factor-α and blocking CD4+ T lymphocyte activation [13].
Within the Onygenales, Blastomyces, Histoplasma and Paracoccidioides belong to the family Ajellomycetaceae. Also within Ajellomycetaceae is the genus Emmonsia, which includes E. crescens and E. parva, the etiological agents of adiaspiromycosis, a pulmonary disease of small mammals and occasionally of humans [16]. Recently, a cluster of systemic infections of HIV-positive patients in South Africa were shown to be caused by Emmonsia isolates [17]. While E. crescens and E. parva also undergo a dimorphic shift at high temperature, they transform into large, thick-walled adiaspores rather than yeast cells [18] (S1 Table). Two phylogenetic studies using 18S ribosomal DNA sequences found that E. parva was the sister species to Blastomyces [19,20]. The positioning of E. crescens was less clear; in one analysis it was a sister group to Paracoccidioides [19] while in the other analysis it was grouped with Blastomyces and E. parva [20]. In neither phylogeny was the alternative positioning of E. crescens strongly supported.
To further investigate the genomic basis of differences observed among the Ajellomycetaceae in terms of pathogenicity, morphology, and the infection process, we sequenced six genomes of Blastomyces and Emmonsia, as well as sequencing the B. dermatitidis transcriptome during macrophage co-cultivation and in vivo pulmonary infection. The newly sequenced genomes included three representative strains of B. dermatitidis (ER-3, ATCC18188, and ATCC26199), and one strain of each of B. gilchristii (SLH14081), E. parva (UAMH139), and E. crescens (UAMH3008). Blastomyces dermatitidis ER-3 was isolated from a woodpile located in a highly endemic region of Wisconsin and is hypovirulent in mice [21,22]. The ATCC18188 strain is the only current example of the 'a' mating type (MAT1-1 locus) available for B. dermatitidis [23]. ATCC26199 is a clinical isolate from South Carolina that is commonly used for in vitro and in vivo laboratory assays [14]. Blastomyces gilchristii SLH14081 is a human clinical isolate that is highly virulent in a murine model of blastomycosis [22,24]. Both Emmonsia strains were isolated from small mammals, E. parva from a weasel in Ravelli County, Montana, and E. crescens from lungs of a rodent (Arvicola terrestris) in Norway.
Utilizing this genomic data, we find that the Blastomyces genomes are much larger than those of their close relatives, and are characterized by large, isochore-like GC-poor regions overrun by repetitive elements. Our whole-genome analyses provide further evidence for the phylogenetic relationships between Blastomyces and Emmonsia and other Onygenales. Finally, we identify novel sets of candidate virulence factors through comparison of the Blastomyces transcription during in vivo pulmonary infection to growth in co-culture with macrophages or in different media or temperature. This combination of genomic and transcriptomic analysis provides a foundation and new candidate genes to further characterize the underlying molecular differences that determine the infectious potency of Blastomyces strains and give rise to the clinical profiles attributable to blastomycosis.
We sequenced and assembled the genomes of three Blastomyces dermatitidis strains and one B. gilchristii strain, and representatives of two Emmonsia species. The Blastomyces strains were sequenced using either Sanger technology or a hybrid of Sanger and 454 technologies. The Emmonsia strains were sequenced using Illumina technology, and de novo assemblies were generated for each strain (Methods). Comparison of the genomes of four Blastomyces strains, SLH14081, ER-3, ATCC18188 and ATCC26199, revealed they were over twice the size of all other Onygenales. The Blastomyces assemblies range in size from 66.6 Mb for B. dermatitidis strain ER-3 to 75.4 Mb B. gilchristii strain SLH14081 (Table 1). These assemblies were over twice as large as those of other dimorphic pathogens in the order Onygenales including the Emmonsia species (30.4 Mb), although the use of only short reads from a single library for the two Emmonsia may under-represent repetitive sequence (Fig 1). The assemblies of two Blastomyces strains, SLH14081 and ER-3, were sequenced to a higher depth than the other two strains, and as a result contain nearly all of the assembled sequence in a relatively small number of scaffolds, 100 and 25 scaffolds respectively. As an independent assessment of genome size and structure, we generated an optical map of the SLH14081 strain (S1 Fig). Consistent with our assembly of this strain, the map had an estimated size of 79.6 Mb, arranged in eighteen linkage groups. In addition, a total of 65.9 Mb of the 75.4 Mb of the SLH14081 assembly was anchored to the optical map (S2 Table).
The total number of predicted genes in Blastomyces, Emmonsia, and other related fungi was similar despite the large difference in genome size. In Blastomyces, the number of predicted genes varied between 9,180 in ATCC26199 to 10,187 in ATCC18188; for E. parva and E. crescens the counts were similar, 8,563 and 9,444, respectively (Table 1), as were those of other sequenced Onygenales (Fig 1). High representation of core eukaryotic genes in each genome provides evidence that their gene sets are nearly complete; E. parva includes 88% of core eukaryotic genes, while the E. crescens and Blastomyces gene sets include 96–98% (S2 Fig).
To compare gene content and conservation, we identified orthologous gene clusters in the six genomes sequenced here, 10 additional Onygenales genomes, including three other pathogenic species (Histoplasma, Paracoccidioides, and Coccidioides), and three Aspergillus genomes. Using 2,062 single copy core genes present in all strains, we estimated a phylogeny of these organisms using RAxML ([25]; Fig 1). This analysis strongly supports the clustering of Blastomyces with E. parva (100% of bootstrap replicates and 100% Gene Support Frequency (GSF) [26]) as previously reported [19,20]. In contrast to prior work, Histoplasma is strongly supported as sister group to Blastomyces and E. parva (100% of bootstrap replicates and 90% GSF), with E. crescens strongly supported as a sister group to that clade (100% of bootstrap replicates and 100% GSF), and with Paracoccidioides in a basal position (Fig 1). The polyphyletic nature of Emmonsia suggests that the Ajellomycetaceae have undergone multiple evolutionary transitions allowing the infection of humans and other mammals. Within Blastomyces, we found support for strain SLH14081 as an outgroup relative to the other three strains (S3 Fig). This is consistent with the placement of strain SLH14081 within the newly described species B. gilchristii [3]; the other three strains sequenced here are still classified as B. dermatitidis.
A bimodal distribution of GC-content observed in all Blastomyces sequenced, which was less pronounced in E. parva and E. crescens and absent in other Ajellomycetaceae, suggests that these genomes are organized in large isochore-like regions of high and low GC-content. This finding for nuclear DNA explains the GC-poor fraction of the Blastomyces genome initially identified using CsCl gradient analytical ultracentrifugation [27], which the authors hypothesized was due to a large proportion of GC-poor mitochondrial DNA in Blastomyces cells. Examining the genome wide GC content revealed a bimodal distribution for all strains of Blastomyces including ER-3 and SLH14081, the smallest and largest assembly, respectively (Fig 2), and was observed for all window sizes ranging from 2 kb to 256 kb (S4 Fig). The detection of a bimodal signal in larger windows supports the organization of the genomes in large isochore-like regions, with average GC content of 29.6% and 31.0% in GC-poor regions and 45.9% and 46.6% for the rest of the genome in B. gilchristii strain SLH14081 and B. dermatitidis strain ER-3, respectively (Table 2). Analysis of the related pathogens H. capsulatum, P. lutzii, and C. immitis showed no evidence for bimodality of GC content, while both E. parva and E. crescens revealed small peaks of low GC sequence. Read-based analysis and using smaller window sizes (e.g. 128 bp) supported these findings, suggesting they are not due to differences in assembly completeness (S5 Fig).
To further examine the organization of GC-content across the genome, we next defined the boundaries of low GC content regions in Blastomyces. In the smallest assembly, of the ER-3 strain, we identified 221 GC-poor tracts with an average size of 186.0 kb, encompassing a total size of 41.1 Mb (Tables 2 and S3). In the largest assembly, of the SLH14081 strain, we identified 350 GC-poor tracts with an average size of 140.2 kb, encompassing a total size 49.1 Mb (Tables 2 and S3). The 8 Mb difference between the total size of GC-tracts in the genomes of B. dermatitidis ER-3 and B. gilchristii SLH14081 accounts for nearly all of the 8.8 Mb difference in assembly size. Notably, GC-poor tracts in Blastomyces can be quite long, and reach maximal lengths of 1.3 Mb. In the assemblies of E. parva, E. crescens and other Ajellomycetaceae, long GC-poor tracts were rarely observed (e.g., a total of only 4 GC-poor regions larger than 10 kb in E. parva were found adjacent to a long GC-rich region in the same scaffold, and just 1 in E. crescens), corresponding to the less pronounced bimodal GC distribution of the genome assembly. However, more contiguous assemblies would be needed to reveal the overall extent of long GC-poor tracts. The only other fungal genome noted to have an isochore-like structure, Leptosphaeria maculans [28], contains a smaller expansion of GC-poor regions (Fig 2); individual tracts were on average half the size (70.4 kb) of those in Blastomyces, and encompassed a smaller fraction (36%) of the L. maculans genome [28]. This difference is consistent with the lower fraction of long AT blocks we observe by comparing windows of different sizes in Blastomyces and L. maculans (S4 Fig).
The GC-poor regions include nearly all the repetitive elements in the genome and consequently have a lower density of predicted genes (e.g., see Fig 3). In ER-3, 93.7% of repetitive sequence is found in GC-poor regions (Table 2). The gypsy elements that dominate repetitive sequence in the Blastomyces genomes have low GC-content; on average those in ER-3 and SLH14081 have respective GC-content of 31.0% and 29.9%, matching the overall GC level of the GC-poor regions (Table 2). GC-poor tracts of Blastomyces contain only approximately one fifth of the predicted protein-coding gene set, including some notable genes such as 1,3-beta-glucan synthase component (FKS1), Blastomyces yeast phase-specific gene (BYS1), and one of two BYS1-like proteins we identified (S6 Fig and S4 Table). By contrast, BAD1, which encodes an essential virulence factor involved in host cell interaction and immune evasion [13], is found within a GC-rich region. Intergenic regions are also larger here than for other genes in the genome; the average intergenic region for ER-3 is 18.5 kb in GC-poor regions, a 3-fold expansion compared to the 6.0 kb genome-wide average (Table 2 and Figs 3 and S6).
The GC-poor regions also show lower synteny between the Blastomyces genomes compared to other regions with more typical GC content (e.g., see Fig 3). Overall, B. dermatitidis strain ER-3 and B. gilchristii strain SLH14081 shared 125 syntenic blocks including 93.8% and 94.5% of genes, encompassing only 69.5% and 69.3% of each assembly. These percentages are much smaller than those observed among strains of related species (such as 95% and 93% synteny between strains of P. brasiliensis [29]). The lower synteny among Blastomyces strains is largely explained by the proportion of genes found in repeat-rich, GC-poor regions (Table 2 and Fig 3). Nearly all (99%) of genes in GC-rich regions are highly syntenic across Blastomyces strains, even between B. dermatitidis strain ER-3 and B. gilchristii strain SLH14081. However, the GC-poor regions have more limited synteny even within strains of Blastomyces encompassing 74 to 76% of genes in those regions (Table 2 and Fig 3).
Overall, the function, expression, and selective pressure of genes in GC-poor regions appear similar to those genes found elsewhere in the genome. Despite the lower synteny, GC-poor regions are not significantly enriched for Blastomyces-specific genes, nor did they show much functional enrichment (S1 Text, S5 Table). Comparing selection pressure on the 7,228 single copy orthologs present in all four Blastomyces genomes also did not find a significant difference in the number of genes with high omega values (omega > 1) (Methods). These analyses suggest that despite the dynamic reorganization due to invading gypsy elements, the GC-poor regions do not appear to be fast evolving by these measures. Furthermore, there is no large-scale difference in the expression levels of genes in GC-poor regions. Comparing transcript levels for genes in GC-poor and GC-rich regions, we found that genes in both GC classes show similar expression levels (S7 Fig), again supporting the general similarity of genes found in these two genomic neighborhoods.
The 2-fold larger size of the Blastomyces genomes compared to other dimorphic fungi is due largely to an expansion of repetitive sequence. The proportions of the Blastomyces genome assemblies that were covered by repeats ranged from 56.6% (41.6 Mb) for B. dermatitidis ATCC18188 to 63.0% (47.5 Mb) for B. gilchristii SLH14081. SLH14081 had the highest repeat content and the largest assembly size. The E. parva and E. crescens assemblies both had a lower repeat content, 9.9% (3.0 Mb) and 5.4% (1.6 Mb), respectively (Table 1). In all genomes, a small number of transposable element classes as well as AT-rich simple sequence regions were highly represented (Fig 4A).
More specifically, the genome expansion in Blastomyces strains has resulted from a proliferation of gypsy LTR retrotransposons, including both ancestral and lineage-specific proliferation. In the Blastomyces genomes, Gypsy elements account for almost all repetitive DNA, with a lower frequency of sequences similar to the non-LTR Tad1 and copia LTR retroelements (Figs 4A and S8). In all Blastomyces and Emmonsia genomes the most frequent Gypsy element subtype matches the “ACa” (Ajellomyces or Histoplasma capsulatum) Gypsy element from Repbase [30] (Fig 4A and 4B). Further phylogenetic characterization of 2,331 Gypsy elements identified four subtypes that appear specific to Blastomyces (S1 Text and Figs 4 and S9). Some subtypes had diversities that were primarily the result of ancestral duplication, resulting in large numbers of orthologous elements between strains (e.g., Fig 4B), while other subtypes appeared to predominantly contain strain-specific paralogous expansions, consistent with the cryptic speciation in the Blastomyces genus (e.g., Fig 4C). Gypsy elements were also detected in the Emmonsia and Histoplasma assemblies, but in far fewer copies (Figs 3 and 4), consistent with the recent expansion in Blastomyces. Gypsy elements are frequently observed in fungal genomes [31], including Coccidioides [32] and Paracoccidioides [29] but in far fewer copies.
To identify gene content that could play a role in the evolution of the dimorphism and pathogenesis within the family Ajellomycetaceae, we searched for expansions or contractions in functionally classified genes compared to the other fungi from the order Onygenales (S6 Table). We identified PFAM domains, KEGG pathways, Gene Ontology (GO) terms, or kinase families that were significantly enriched or depleted. Domains associated with polyketide synthases were depleted in the Ajellomycetaceae, and an independent prediction of secondary metabolite enzymes confirmed that Blastomyces and other fungi from the Ajellomycetaceae contain fewer PKS gene clusters than other Onygenales (S7 Table, S1 Text). Other differences between the Ajellomycetaceae and other Onygenales include fewer copies of multiple classes of peptidases (M36, M43, S8) as well as an associated inhibitor (I9, inhibitor of S8 protease), variable copy number of LysM-domain proteins potentially involved in chitin binding, which are most expanded in dermatophytes but at next highest levels among the human pathogens in Blastomyces, and a higher number of protein kinases (S6 Table and S10 Fig), including an expansion of the FunK1 family similar to that previously noted in Paracoccidioides [29].
We next identified 140 gene clusters conserved in Blastomyces, Emmonsia, Histoplasma, and Paracoccidioides, but absent from other Onygenales and Aspergillus (S8 Table). These gene clusters include a predicted heme oxygenase (BDBG_02689), which could produce free iron from host heme. In addition to the 140 gene clusters, we also identified conserved genes in subsets of the Ajellomycetaceae including homologs of two previously typed antigens; a gene similar to the 27 kDa antigen of Paracoccidioides [33] is present in Blastomyces and one Histoplasma genome, and a gene cluster specific to Blastomyces and Paracoccidioides shares similarity with the antigenic Aspergillus cell wall mannoprotein [34].
To identify potential genetic features of the Ajellomycetaceae pathogenic to immunocompetent humans (Blastomyces, Histoplasma, and Paracoccidioides) relative to E. parva and E. crescens, we conducted a second enrichment analysis as described above (S9 Table). The primary pathogens showed enrichment of only two PFAM domains, a phosphorylase and endonuclease (S9 Table). The phosphorylase domain over-represented in Blastomyces is associated with nucleoside phosphorylases; many of these proteins also contain Ankyrin repeats and NACHT domains. Phosphorylases are involved in nucleotide and amino acid salvage, and could allow pathogens greater metabolic versatility when certain building blocks are unavailable. The absence of any larger pattern of gain or loss of functional classes suggests that smaller changes in gene content, independent gain and loss between the species, or expression differences may account for differences in pathogenesis.
We then identified specific orthologs present in all four strains of Blastomyces but absent from both non-pathogenic Emmonsia species. Comparing the ortholog set of Blastomyces to E. parva and E. crescens, we found a total of 552 ortholog clusters that were present in all Blastomyces strains but absent in both Emmonsia genomes (S10 Table). Most of these (393 clusters) were present only in Blastomyces, and while most of these proteins (92% in B. gilchristii strain SLH14081) had no PFAM domain assignment, the set did include the Blastomyces yeast phase-specific protein 1 (BYS1). This gene is a marker of the phase transition to and from the yeast phase [15], although it has recently been shown not to be required for virulence in studied strains [24].
While both E. parva and E. crescens are not reported to be primary human pathogens, phylogenetic analysis suggests that the transition to this lifestyle may have been independent, resulting in differential gene loss. One of the genes absent only in E. crescens is the siderophore iron transporter mirB (MIRB). Many pathogenic microorganisms have evolved high affinity iron acquisition mechanisms, which include the production and uptake of siderophores. In B. dermatitidis, the expression of genes involved in the biosynthesis of siderophores and uptake of siderophores, including two iron transporters (MIRB and MIRC), are induced under iron-poor conditions [35]. While MIRB appears to be absent in E. crescens, siderophore uptake may be still enabled by the second transporter, MIRC, which is conserved in this species.
To better understand which Blastomyces genes play roles in pathogenicity and virulence, we carried out RNA-Seq of B. dermatitidis strain ATCC26199 to profile expression under varying temperature, nutrient availability, and infection status. Combining this data allowed us to disambiguate expression variability due solely to differences in temperature and media-specific nutrient availability from those specific to macrophage interactions in vitro or host infection in vivo. Five conditions were sampled: 37°C with macrophages in RPMI media, 37°C in RPMI media, 37°C in HMM media, 22°C in HMM media, and in vivo pulmonary infection with yeast in a mouse model (Fig 5A). For each condition, two biological replicates were performed, and the read counts per transcript were highly correlated between replicates (R> 0.99, S11 Fig). Gene expression levels and mapping statistics are presented in S11 and S12 Tables, respectively.
When B. dermatitidis yeast cells were co-cultured with human bone marrow derived macrophages, the majority of yeast cells (59%) were internalized by macrophages. Comparison of yeast co-cultured with and without macrophages identified 140 genes differentially expressed between these two conditions, 112 of which were upregulated in the presence of macrophages (S13 Table). This upregulation suggested a small, specific response to macrophages in this experiment. Examination of this set of genes revealed numerous genes that have the potential to facilitate adaptation to the host environment. The 20 most significantly upregulated genes (Table 3) include a predicted secreted endo-1,3(4)-β-glucanase (BDFG_03060) involved in cell separation after cytokinesis in C. albicans [36], transporters, including an ABC transporter (BDFG_05060) homologous to Aspergillus fumigatus MDR1 and a zinc transporter (BDFG_02462) similar to the vacuolar zinc transporter ZRT3 in S. cerevisiae, dehydrogenases involved in amino acid catabolism, and antioxidants peroxisomal catalase (CATP, BDFG_02965) and superoxide dismutases (SOD3, BDFG_01204; SOD2, BDFG_07895), which may protect against reactive oxygen species (ROS). The induction of endo-1,3(4)-β-glucanase and CATP in the presence of macrophages was also confirmed by qRT-PCR (S12 Fig and Methods).
We also identified gene expression changes specific to in vivo murine pulmonary infection from our transcriptomic data of B. dermatitidis strain ATCC26199. By k-means clustering of expression values, we identified a set of 72 genes that are upregulated in vivo during mouse infection relative to all other conditions, regardless of temperature, media, and in vitro macrophage interactions (Fig 5B and S14 Table). Using all conditions for this comparison helped eliminate from consideration differences observed, for example, between the yeast samples cultured in different media. Genes in this set with greater than 2-fold upregulation in vivo are highlighted in Table 4, and primarily fell into five functional categories: 1) secreted proteins, 2) zinc acquisition, 3) antioxidants and oxygenases, 4) amino acid metabolism, and 5) transporters.
The most highly expressed gene in vivo was BAD1 (BDFG_03370; S11 Table), which encodes a yeast-phase specific virulence factor that facilitates adhesion to host cells and evasion of host immune defenses [13]. BAD1 also had the highest transcript abundance for yeast co-cultured with macrophages and yeast without macrophages at 37°C (S13 Table). Thus, BAD1 was not identified among the set of 72 differentially expressed genes because the transcription of BAD1 is influenced by temperature [37]. The effect of temperature during the mold to yeast transition on the transcriptome of dimorphic fungal pathogens has been the topic of previous studies [38–41] and was therefore not further evaluated here.
A total of nine secreted proteins were identified in this set of 72, including five of the ten most highly upregulated genes by fold change, potentially playing roles in host-pathogen interactions. Another highly up-regulated secreted protein (BDFG_00717) contains a predicted CFEM domain as well as a GPI-anchor; these features, as well as small size (236 amino acids), are shared with member of the haemoglobin-receptor gene family in C. albicans [42]. The most highly upregulated gene, BDFG_05357, encodes a HRXXH domain-containing secreted protein that may function as a zinc scavenging protein (Tables 4 and S14). This gene is present in the genomes of Blastomyces and Coccidioides, but absent from those of Emmonsia, Histoplasma and Paracoccidioides. BDFG_05357 is a homolog of C. albicans PRA1 (pH-regulated antigen-1) [43] and S. cerevisiae ZPS1 (zinc-pH-regulated protein). In C. albicans, the transcription of PRA1 and ZPS1 is induced under alkaline pH and zinc-deplete conditions [44,45], and PRA1 is co-regulated with its upstream gene, ZRT1, which encodes a high-affinity zinc transporter that interacts with zinc-bound PRA1 [45]. Similarly, the B. dermatitidis homolog of ZRT1, BDFG_09159, is highly expressed in vivo; the induced expression of both PRA1 and ZRT1 was confirmed by qRT-PCR (S12 Fig). However unlike in C. albicans, ZRT1 is not adjacent to PRA1 in the B. dermatitidis genome. While PRA1 is conserved in all four Blastomyces genomes, there is no copy of this gene in Histoplasma as previously noted [45], nor in Emmonsia or Paracoccidioides, suggesting differences in how zinc is acquired within the Ajellomycetaceae.
In addition to PRA1/ZPS1 and ZRT1, a larger module of genes that regulate zinc acquisition is co-regulated in Blastomyces. The transcript abundance of BDFG_07269, which encodes a low-affinity zinc transporter (ZRT2), is also significantly upregulated in the mouse model. In S. cerevisiae, the zinc-responsive transcription factor ZAP1 regulates expression of ZRT1 and ZRT2, along with ZPS1. We identified the ortholog of ZAP1 in strain ATCC26199 as BDFG_07048, which was also significantly upregulated in vivo relative to all other conditions (S14 Table) despite not being identified by k-means clustering. These results suggest that zinc acquisition and homeostasis may play a critical role for survival of B. dermatitidis in vivo.
Genes that convert reactive oxygen species to dioxygen and dioxygen to metabolites were also highly upregulated in vivo. These include two superoxide dismutases (SOD3: BDFG_01204 and SOD2: BDFG_07895), which were even more upregulated in vivo than in macrophages. Four dioxygenases (BDFG_04184, BDFG_04185, BDFG_08059, BDFG_06504) were also upregulated in vivo, representing almost half of the dioxygenases found in the genome, which utilize dioxygen to drive amino acid catabolism. This set includes 4-hydroxyphenylpyruvate dioxygenase, (4-HPPD; BDFG_04184) and homogentisate 1,2-dioxygenase (BDFG_04185), which are involved with tyrosine catabolism [46]. Other upregulated oxygenases include indoleamine 2,3-dioxygenase (BDFG_06504) and squalene monooxygenase (ERG1—BDFG_07857), which are involved with tryptophan catabolism and ergosterol biosynthesis respectively. ERG1 is a target of current antifungal drugs, including terbinafine. High in vivo expression of this gene may suggest that drugs targeting it may be more effective in vivo than in vitro.
Genes involved in cysteine biosynthesis and catabolism were highly upregulated during infection including cysteine synthase A (BDFG_02039) and cysteine dioxygenase (BDFG_08059). Cysteine synthase A may provide a large pool of synthesized cysteine for B. dermatitidis metabolism; the induced expression during infection was confirmed by qRT-PCR (S12 Fig). Based on orthology analysis, cysteine synthase A is absent from the genome of H. capsulatum, and previous studies have shown that Histoplasma yeast are auxotrophic for cysteine [47,48]. Cysteine dioxygenase catabolizes cysteine to L-cysteinesulfinic acid, an intermediate that can be used for taurine biosynthesis or metabolized to sulfite and pyruvate. A homolog of C. albicans SSU1 (BDFG_06814), which encodes a sulfite efflux pump and is co-regulated with cysteine dioxygenase in C. albicans [49], was also upregulated in B. dermatitidis.
Transporters in fungi have been associated with an enhanced ability to remove harmful products as well as to survive on diverse nutrient sources, both of which could contribute to virulence and pathogenicity. Of the 72 genes upregulated in vivo during mouse infection, 11 are predicted transporters. These included the major facilitator type (MFS; BDFG_06068 –unknown function, BDFG_06042 –glycose transport, BDFG_02038 –unknown function), amino acid transporters (BDFG_02310, BDFG_07447) and metal transporters (zinc/iron transporters discussed above, BDFG_09159, BDFG_07269, and NIC1 nickel transporter, BDFG_02449; S14 Table). This upregulation potentially reflects differences in metabolite and cofactor availability in the host relative to in vitro conditions.
Our whole-genome based phylogenetic analysis recovered a sister-group relationship between Blastomyces spp. and Emmonsia parva, as previously reported from ribosomal DNA sequences [19,20]. However, our study placed Histoplasma as the next most basal species, and uniquely placed E. crescens between Histoplasma and the basal Paracoccidioides with strong bootstrap support. This more external position of Paracoccidioides compared to Histoplasma agrees with an earlier rDNA tree without Emmonsia [50]. Furthermore, gene support frequencies (GSF) were relatively high, and increased when we subsampled only well-supported genes, providing additional support for the topology presented here.
The polyphyletic nature of the non-human pathogen Emmonsia suggests substantial plasticity in regard to human pathogenesis in this group. Ancestral variation in the ability of these species to infect other mammals could then be associated with exaptation to human hosts. Additional diversity of Emmonsia, including the third described species, E. pasteuriana [51,52] and other closely related isolates [17] suggests that the full breadth of the Emmonsia genus may not be captured by the two isolates sequenced here. Interestingly, both E. pasteuriana and related isolates produce yeast cells at high temperature, rather than the adiaspores produced by E. parva and E. crescens. Further sequencing of Emmonsia species and other related strains may reveal additional patterns and trends in the evolution of the dimorphic fungi.
The mosaicism observed here in the genome of Blastomyces differs substantially from that observed in other fungi and larger eukaryote genomes. While isochore-like GC-poor regions are unprecedented at this scale in fungal genomes described to date, there are parallels to the organization of L. maculans, though GC-poor regions occupy a smaller fraction of that genome [28]. Longer GC-poor isochores of more than 300 kb are commonly found in mammals and other vertebrates [53–55]. GC-poor isochores in vertebrates are often more stable over long evolutionary times [55,56] and have other properties such as lower gene expression [55] that do not appear to be shared by the GC-poor tracts of B. dermatitidis and B. gilchristii (S1 Text).
Characterization of repetitive sequence in GC-poor regions suggests these originated with a dramatic expansion of elements of the LTR/Gypsy category. Phylogenetic analysis suggests these elements swept through a lineage leading to the present-day B. dermatitidis and B. gilchristii, and to a lesser extent Emmonsia parva, and have further expanded during the diversification of Blastomyces. While H. capsulatum does not have such an expanded genome, or a sizable GC-poor component, and so appears less affected by gypsy expansion, Histoplasma may alternatively have more robust defense against repetitive elements or be less able to accommodate large amounts of repeats in its genome.
While GC-poor tracts have been particularly dynamic areas due to Gypsy element insertions during the recent evolution of Blastomyces, these regions appear typical in gene content and expression. Perhaps due to their recent origin, the GC-poor regions do not appear to have sequestered particular classes of genes such as secreted proteins or have other hallmarks of rapidly evolving gene content. The long GC-poor regions also include some well characterized genes involved in phase transitions and pathogenesis, including the Blastomyces yeast-specific gene BYS1, a marker of the phase transition to and from the yeast phase [15,24]. Reduced levels of synteny in the GC poor regions between B. dermatitidis and B. gilchristii could prevent effective meiotic recombination between the two lineages, further supporting their designation as separate species.
Despite the large increase in genome size in Blastomyces, the total number of protein coding genes is only modestly expanded. Blastomyces and other sequenced species from the Ajellomycetaceae family, including the human primary pathogens Histoplasma and Paracoccidioides, have similar gene content with only a few gene loss or gain events that map to common functional classes. This stability suggests that more modest differences in gene content and sequence, as well as potential divergence of gene regulation, contribute to phenotypic differences between the species. Larger differences exist between the Ajellomycetaceae and other more divergent members of the Onygenales. There is no expansion of degradative proteases as previously noted for Coccidioides [57]; in fact, three peptidase families (M36, M43, and S8) are present at lower copy number in Blastomyces and the other Ajellomycetaceae. While Blastomyces contains a higher number of LysM proteins than the dimorphic Onygenales, the number is small relative to that found in Dermatophytes [58]. This analysis also identified candidate genes involved in host interaction, including proteins homologous to antigens in related fungi and a heme oxygenase that could release iron from host heme.
For yeast co-cultured with macrophages and yeast in vivo, some aspects of the transcriptional response were shared including response to oxidative stress and amino acid catabolism. Yeast co-cultured with macrophages showed upregulation of numerous genes involved in oxidative reduction, which may play a major role in protecting Blastomyces from ROS. The macrophage phagosome is poor in glucose and amino acids, but rich in ROS [59,60]. Blastomyces is relatively resistant to ROS and virulence correlates with the ability to minimize ROS generation in innate immune cells [61,62]. The upregulation of superoxide dismutases (SOD3, SOD2) and catalase P may protect B. dermatitidis yeast against oxidative stress. In H. capsulatum, extracellular SOD3 and intracellular catalase P, contribute to survival within macrophages [63,64]. Moreover, SOD3 promotes H. capsulatum virulence in a murine model of pulmonary infection [63]. The upregulation of 4-HPPD, which is involved with pyomelanin biosynthesis, contributes to antioxidant defense and intracellular survival of Penicillium marneffei [65]. Inhibition of 4-HPPD in P. brasiliensis and P. marneffei blocks the phase transition to yeast at 37°C [65,66]. Furthermore, in vivo numerous dioxygenases were upregulated, suggesting that dioxide produced in response to ROS may be utilized for amino acid metabolism.
Changes in amino acid metabolism were prevalent in both the macrophage co-cultured and in vivo Blastomyces, suggesting the recycling of amino acids as an energy source (Results, S1 Text). In particular, the very specific increase in cysteine catabolism (cysteine dioxygenase) and biosynthesis (cysteine synthase A) during in vivo infection suggests that cysteine may be critical to virulence. In mice, deletion of cysteine dioxygenase (CDG1) in C. albicans results in attenuated virulence [49]. Furthermore, upregulation of sulfite efflux pump (SSU1), which is co-regulated with CDG1 in C. albicans, could play a role in B. dermatitidis virulence during in vivo infection. Exported sulfite can destabilize host proteins by reducing disulfide bonds and facilitates the growth of dermatophytes on keratinized tissue [67]. How breakdown of tryptophan by indoleamine 2,3-dioxygenase (IDO), which can supply de novo nicotinamide adenine dinucleotide (NAD+), alters the fungal-host interaction is unknown. In cancer, tumor cells with increased expression of IDO may facilitate tryptophan depletion in the microenvironment to suppress the host immune response [68]. Infection with H. capsulatum, P. brasiliensis, and C. albicans upregulates host IDO activity, reduces fungal growth, impairs Th17 T-cell differentiation, and blunts excessive tissue inflammation [69–71].
The specific in vivo upregulation of genes that encode secreted proteins as well as those involved with transmembrane transport (e.g., amino acids, glucose), amino acid metabolism (e.g., cysteine), and metal acquisition (e.g., zinc, nickel) highlights virulence factors potentially being missed by in vitro studies and the importance of understanding nutrient and co-factor availability in any study system. Uptake of zinc and nickel, which serve as enzyme co-factors, contribute to virulence in C. albicans and Cryptococcus neoformans respectively [45,72]. PRA1 encodes a secreted “zincophore” under alkaline and zinc-poor conditions that acts in concert with ZRT1 to promote zinc acquisition and facilitate endothelial cell damage by C. albicans [45]. NIC1-mediated nickel uptake is critical for urease activity, which contributes to C. neoformans invasion of the central nervous system [72]. In C. posadasii, urease induces host tissue damage [73]. While genes involved with the acquisition of zinc (e.g., ZRT1, ZRT2, ZAP1 homologs) and nickel (e.g., NIC1 homolog) are largely conserved with other fungi, the absence of PRA1 in Histoplasma, Paracoccidioides, and Emmonsia highlights recent evolutionary changes in zinc acquisition mechanisms in the family Ajellomycetaceae. This, in conjunction with differences in cysteine metabolism between Blastomyces and Histoplasma, suggest that despite the many common elements of dimorphism and pathogenesis, each genus of dimorphic fungi likely has unique nutritional requirements.
Four strains of Blastomyces were sequenced: SLH14081 representing the new species B. gilchristii, and ER-3, ATCC18188 and ATCC26199 representing B. dermatitidis. The SLH14081 strain is a highly virulent, clinical isolate that can cause disease in immunocompetent persons, while ER-3 was isolated from a woodpile and is hypovirulent in mice [21,22]. The remaining two strains are strain ATCC18188, a representative MAT 'alpha' isolate, and ATCC26199, a commonly used laboratory isolate.
Two species that are closely related to Blastomyces, that can cause pulmonary disease in rodents (adiaspiromycosis), were also sequenced: Emmonsia parva UAMH139 and Emmonsia crescens UAMH3008. These isolates were chosen for comparison as these species are not typically human pathogens, yet they are closely related to the three pathogenic dimorphic genera Blastomyces, Histoplasma and Paracoccidioides, with which they form a clade that is nested within the order Onygenales and represents the Ajellomycetaceae family [20].
Genomic DNA for sequencing was prepared from mycelial or yeast culture, using phenol/chloroform extraction. For the Blastomyces SLH14081 and ER-3 strains, whole genome shotgun sequence was obtained using Sanger technology on an ABI 3730 from a Fosmid (epiFOS) and two plasmid (pJAN and pOT) libraries. For B. dermatitidis ATCC18188, whole genome shotgun sequence was obtained from two small insert libraries (fragment and 1.5 kb) using Roche 454 technology and from a Fosmid library using Sanger technology. For B. dermatitidis ATCC26199 20X of sequence was generated using 454 technology from a small insert fragment library. In addition, a plasmid (pOT) and Fosmid (epiFOS) library were constructed and sequenced using Sanger technology by the Washington University Genome Center, generating a total of roughly 3.6X coverage.
For each Emmonsia species, a single library was used to generate 101 bp paired-end reads using Illumina technology on a Genome Analyzer II generating a total of 116X coverage for E. parva UAMH139 and 163X coverage for E. crescens UAMH3009. Libraries of average insert size of 639 bp for E. parva and of 686 bp for E. crescens were chosen based on the electropherograms obtained from Bioanalyzer. Sequencing of both Emmonsia genomes was performed at the Genomic Sequencing Laboratory, University of California, Berkeley.
Blastomyces strains SLH14081 and ER-3 were assembled with Arachne [74] (Assemblez Build 20080911). For B. dermatitidis ATCC18188, a hybrid assembly was generated with Newbler version 2.3. For B. dermatitidis ATCC26199, a hybrid assembly of the Sanger and 454 data was generated with Newbler version "MapAsmResearch-03/15/2010" with options-rip and -scaffold.
For the Emmonsia genomes, assemblies were generated using multiple programs, including the SOAPdenovo / GapCloser package [75], ABYSS [76] and Velvet [77]. SOAPdenovo assemblies were selected based on quality metrics. While assemblies with high k values increased the fraction of GC-poor regions represented in the assembly, optimal assembly of gene sequences were achieved using lower k values, based on comparing each assembly to gene sets of Blastomyces and other related dimorphic fungi using TBlastN. The assemblies for the Emmonsia genomes (k = 27 for E. parva and k = 29 for E. crescens) were processed using the program GAEMR (http://www.broadinstitute.org/software/gaemr/), where overall assembly metrics were used to select the best assembly considering both continuity and completeness, though these measures were lower than for genomes assembled from multiple libraries.
To validate the assembly of strain SLH14081 and anchor it onto chromosomes, we constructed an optical map, a single-molecule based ordered restriction map. The map of B. gilchristii strain SLH14081 was constructed by OpGen using the restriction enzyme BsiWI (C^GTACG). The optical map consists of 16 linkage groups, with size ranging from 9.728 Mb to 730 kb. The total size of the map was estimated as 79.64 Mb in size, slightly larger than the 75.35 Mb genome assembly, likely due to repetitive element sequence missing from the assembly. A total of 36 assembly scaffolds covering 68.9 Mb were mapped based on shared restriction sites to the optical linkage groups (S2 Table).
To enable more accurate gene prediction and analyze gene expression, RNA was prepared and deeply sequenced from five conditions (yeast with or without macrophages in RPMI media, in vivo during murine pulmonary infection, and in vitro yeast and mold in Histoplasma macrophage media (HMM)) with two biological replicates per condition.
ATCC26199 yeast cells were co-cultured with bone marrow derived murine macrophages from C57BL/6 mice in RPMI media with 10% heat inactivated FBS and supplemented with penicillin (100 U) and streptomycin (100 ug) or incubated in this media alone. Yeast and macrophages were co-cultured using a ratio of one yeast for every two macrophages (MOI 0.5). The use of alveolar macrophages was precluded due to the large numbers of mice that would be needed to harvest these cells. Following inoculation of cell culture flasks with B. dermatitidis yeast, the co-cultures were incubated at 37°C for 24 hrs. The majority of the yeast were either single cells or cells with one bud (average 89%). The extent of macrophage internalization of yeast was measured using Uvitex staining to differentiate between extracellular and intracellular yeast. A total of 1,976 cells were counted across seven individual fields of view, with an average of 59% Uvitex negative (intracellular) and 41% Uvitex positive (extracellular). The majority of B. dermatitidis cells exhibited yeast morphology (> 96%); pseudohyphal growth occurred in 2.4% of co-cultured yeast and 3.7% of yeast grown in RPMI media without macrophages. Harvested yeast cells were flash frozen in liquid nitrogen, ground with a mortar and pestle into a fine powder, and RNA extracted using the phenol-guanidium thiocyanate-1-bromo-3-chloropropane extraction method [78].
For in vivo transcriptional profiling, C57BL/6 mice were infected with 2 x 103 B. dermatitidis ATCC26199 yeast cells intratracheally and monitored for signs and symptoms of infection [79]. Mice with euthanized by carbon dioxide at 17 days post infection and yeast were isolated from murine lung tissue using the technique developed by Marty et al. [80]. Briefly, excised lungs were homogenized in pre-chilled (4°C) double-distilled, sterile water (ddH2O) supplemented with DNase I 10 μg/ml (Roche) using an Omni TH tissue homogenizer (Omni International, Kennesaw, GA), passed through a 40 μm cell strainer (ThermoFisher Scientific, Waltham, MA), and centrifuged at 770g for 5 minutes at 4°C. The supernatant and interface were removed using a serologic pipette and yeast pellet was washed with ice-cold ddH2O and rapidly frozen in liquid nitrogen for RNA extraction. Time ex vivo was less than 30 minutes and samples were near-freezing (4°C) during all isolation steps. Quality control analyses using qRT-PCR demonstrated that the short ex vivo time (< 30 minutes) at 4°C minimized changes in transcript abundance that would have occurred if the samples were processed at higher temperatures or for a longer duration [80]. Total RNA isolated from B. dermatitidis yeast during pulmonary infection was divided into 2 pools of 5 mice each (pool #1 and pool #2).
In vitro yeast were incubated in liquid Histoplasma macrophage media (HMM) at 37°C while shaking [81]. The majority of cells had yeast morphology; less than 3.25% of cells grew as pseudohyphae. To generate mycelia, yeast cells were incubated in liquid HMM for 14 days at 22°C while shaking. Harvested yeast and mycelial cells were flash frozen in liquid nitrogen, ground with a mortar and pestle into a fine powder, and RNA extracted using the phenol-guanidium thiocyanate-1-bromo-3-chloropropane extraction method [78].
Total B. dermatitidis RNA from all samples (in vivo, in vitro, co-cultures) was treated with TurboDNase (Bio-Rad, Hurcules, CA) and cleaned using an RNeasy column (Qiagen). RNA integrity and quality was assessed using Nanodrop spectrophotometry, 0.8% agarose gel electrophoresis, and an Agilent Bioanalyzer (Agilent Technologies, Santa Clara, CA). RNA integrity numbers (RIN) for in vivo samples were > 7.5 (7.6 for pool #1, 7.8 for pool #2). RIN values for in vitro and co-cultures (including yeast only RPMI) were ≥ 8.7.
For RNA-Seq, poly-A mRNA was purified for each total RNA sample and strand-specific libraries prepared as previously described [82,83]; each library was sequenced using Illumina Technology, generating an average of 65,174,908 101 bp reads per sample. RNA-Seq was incorporated into gene prediction and used to detect differentially expressed transcripts as described below.
For initial gene sets, a total of 38,405 ESTs generated from yeast and mycelial samples of ATCC26199 (Washington University) and from a normalized cDNA library of SLH14081 (Broad Institute) were used for gene prediction. To achieve better transcript coverage, strand-specific RNA-Seq data from 10 samples representing the above yeast, mold, and infection stages was assembled with the Inchworm component of Trinity [84] and processed with PASA [85] to generate a set of transcripts for gene prediction. Gene sets were generated by using EvidenceModeler (EVM) [85] to select the best gene call for a given locus from the gene prediction programs SNAP, Augustus, Geneid, and Genewise and from PASA RNA-Seq transcripts as previously described [85,86].
Project numbers and locus tag prefixes assigned to gene sets are as follows: B. gilchristii SLH14081 (PRJNA41099, locus tag prefix BDBG), B. dermatitidis ER-3 (PRJNA29171, prefix BDCG), ATCC18188 (PRJNA39265, prefix BDDG), and ATCC26199 (PRJNA39263, prefix BDFG); the E. parva UAMH139 (PRJNA178178, prefix EMPG) and E. crescens UAMH3008 (PRJNA178252, EMCG).
RNA-Seq reads were aligned to the transcript sequences of B. dermatitidis strain ATCC26199 using Bowtie [87]. Transcript abundance was estimated using RSEM [88], TMM-normalized FPKM for each transcript were calculated, and differentially expressed transcripts were identified using edgeR [89], all as implemented in the Trinity package version r2013-2-25 [90]. To identify GO term enrichment of differentially expressed genes, we classified transcripts using Blast2GO [91] and then performed comparisons with Fisher’s exact test. A 2-fold difference in FPKM values and a false discovery rate below 0.05 were used as a criteria for significant differential expression. To identify possible functions of the gene products of significantly differentially expressed parasitic-phase genes, protein homologs were assigned based on BLAST, Gene Ontology (GO) terms and protein family domains (PFAM, TIGRFAM).
Total RNA was extracted from B. dermatitidis yeast as described above. One microgram of DNase-treated total RNA was converted to cDNA using iScript cDNA synthesis kit (Bio-Rad). qRT-PCR was performed with SsOFast EvaGreen Supermix (Bio-Rad) using a MyiQ real-time PCR detection system (Bio-Rad). Reactions were performed in triplicate using the following conditions: 1 cycle 95°C x 30 sec, followed by 40 cycles at 95°C for 5 sec, 60°C for 10 sec. Transcript abundance for genes of interest were normalized relative to the transcript abundance of GAPDH. Relative expression (RE) was calculated as RE = 2-ΔCt, ΔCt = Ctgene of interest−CtGAPDH [92].
Primer sequences used were as follows: AATCCTTTGACAGTGAAAC (forward) and CCATAAATCTGCTACAACAG (reverse) for BDFG_03060, ACTGTCGGTGGAGAGAAG (forward) and ACTGGGGTGTTGTTGAAG (reverse) for BDFG 02965, GACTATCCCATCCACAAC (forward) and TACAGAGCGGAATCTTTG (reverse) for BDFG 05357, TTTGGCACTGGAGTTATG (forward) and TGCTTCGTAGTCTAAAGTC (reverse) for BDFG 09159, GTGCTACAACGGAGATAC (forward) and GATAACCACCACGAACAC (reverse) for BDFG 02039, ACCCCCGCTCCTCCATCTTC (forward) and GAGTAGCCCCACTCGTTGTCATACC (reverse) for BDBG_07959 (GAPDH).
We used the IsoFinder GC segmentation program (http://bioinfo2.ugr.es/oliver/isofinder; [93]) to segment all ER-3 and SLH14081 scaffolds into long homogeneous genomic regions (LHGRs). The option p2 (parametric/student t-test with different variances), a window size of 5 kb and a p value cutoff of 0.01 (P parameter 0.99) were chosen after evaluating P cutoffs between 0.95 and 0.99, and window sizes ranging between 3 and 5 kb. The final settings were chosen as they accommodated gene synteny between ER-3 and SLH14081 in the GC-poor segments, obviating the need to manually remove narrow GC peaks caused by short genic regions.
To identify the coordinates of the longer GC-poor and GC-rich tracts on the assemblies of Blastomyces strains ER-3 and SLH14081, we set the boundary between GC-poor and GC-rich at 38% GC, a value that is in the deep valley between the two components of these genomes’ bimodal GC distribution. The deep valley is robust and persists over a wide range of window/segment sizes ranging up to > 60 kb (S4 Fig). We then grouped adjacent segments located between successive transitions (regime switches) across the 38% GC border. Islands of N’s in the interior of the GC-poor tracts were retained, but those at the tract fringes (i.e., next to a jump across the 38% GC threshold) were not. This procedure yields a large-scale segmentation of all scaffolds into strictly alternating “GC-poor” and “GC-rich” tracts. The GC-poor tracts and genes in those regions are listed in S3 and S4 Tables, respectively; GC-rich tracts form the remainder of the assemblies. We performed MySQL joins to identify the genes and repeats (GFF files produced by RepeatMasker of elements from RepeatModeler) located entirely or partly in the GC-poor tracts.
DAGchainer [94] was used to identify syntenic blocks with a minimum of 6 genes, which were required to be in the same order and orientations in the compared genomes. Synteny plots were generated using a custom perl script, using the GDgraph library; code is available at https://github.com/gustavo11/syntenia. Geneious Pro was used to visualize smaller-scale syntenies within and among genome assemblies.
De novo repetitive sequence in each assembly was identified using RepeatModeler version open-1.0.7 (www.repeatmasker.org/RepeatModeler.html). Copies of de novo repeats and fungal sequences from RepBase [95] were mapped in each assembly using RepeatMasker version open-3.2.8 (www.repeatmasker.org/). For phylogenetic analysis of gypsy elements, reverse transcriptase domains were identified from each element; matches to the PFAM RVT_1 domain were identified with HMMER (version 3.1b1) [96] for 6-frame translations of each element generated by EMBOSS transeq (version 6.5.7 with parameters-frame 6-clean Y) [97]. The best domain match for each element was selected, requiring 50% alignment coverage and c-Evalue < 1e-5. The domains identified in Blastomyces SLH14081 (991 total) and ER-3 (1,296 total), E. parva (40 total), and similar Repbase gypsy elements (4 total) were aligned with MAFFT (version 6.717) [98], and a phylogeny estimated using FastTreeDP (version 2.1.8) [99]. Four large subgroups were identified and visualized using iTOL [100].
A total of 16 genomes from the Onygenales order and three Aspergillus genomes were chosen for comparative analyses (S15 Table). These include the four Blastomyces (SLH14081, ATCC26199, ATCC18188, ER-3) and two Emmonsia species (UAMH139, UAMH3008) as well as the following: Histoplasma capsulatum WU24 (AAJI01000000), H. capsulatum G186AR (ABBS01000000), Paracoccidioides lutzii Pb01 (ABKH02000000), P. brasiliensis Pb03 (ABHV02000000), and P. brasiliensis Pb18 (ABKI02000000), Coccidioides immitis RS (AAEC00000000), C. posadasii C735 delta SOWgp (ACFW00000000), Uncinocarpus reesii 1704 (AAIW00000000), Microsporum gypseum CBS118893 (ABQE00000000), Trichophyton rubrum CBS118892 (ACPH01000000), Aspergillus nidulans FGSC A4 (AACD00000000), A. flavus NRRL3357 (AAIH00000000), A. fumigatus Af293 (AAHF01000000). OrthoMCL was used to cluster the protein-coding genes of the 19 chosen genomes by similarity.
To estimate the species phylogeny, a total of 2,062 orthologs present in a single copy in all of the 19 genomes were identified. Protein sequences of the 2,062 genes were aligned using MUSCLE, and a phylogeny was estimated from the concatenated alignments using RAxML v7.7.8 with model PROTCATWAG. To more closely examine the relationship of the Blastomyces isolates, single copy orthologs were identified in all four strains of Blastomyces and E. parva; the protein sequences of a total of 6,605 single copy orthologs were aligned using MUSCLE, and the resulting sequences replaced with the corresponding codons. A phylogeny was estimated from this nucleotide alignment using RAxML v7.3.3 with model GTRCAT. A total of 1,000 bootstrap replicates were used for each analysis. The level of support for the best RAxML tree was also evaluated using individual gene trees, by calculating the gene support frequency (GSF, [26]). A phylogeny was estimated and bootstrapped using the same parameters as for the concatenated sequence matrix, and gene trees with high bootstrap support at all nodes were then selected. A total of 162 gene trees were supported by at least 70% of bootstrap replicates at all nodes; the percent of gene trees supporting the RAxML best tree was calculated using RAxML and is shown in Fig 1. We also evaluated larger subsets of trees including those with 60% bootstrap support at all nodes, 50% bootstrap support, or all trees regardless of support, and found lower support respectively in each subset for our best tree.
To examine selective pressure on genes in GC-poor regions, we identified 7228 genes that were single copy in the four Blastomyces genomes from the OrthoMCL run. dN/dS values for each gene were computed on codon-based nucleotide alignments with the codeml module of PAML [101], using the one-ratio (M0) model.
Genes were functionally annotated by assigning PFAM domains, GO terms, and KEGG classification. HMMER3 [96] was used to identify PFAM domains using release 27. GO terms were assigned using Blast2GO [91], with a minimum e-value of 1x10-10. Protein kinases were identified using Kinannote [102] and divergent FunK1 kinases were further identified using HMMER3. Secondary metabolite gene clusters were predicted with antiSMASH version 2.0.2 [103]. Genes were clustered using OrthoMCL [104] with a Markov inflation index of 1.5 and a maximum e-value of 1x10-5.
To identify functional enrichments in Blastomyces and other subsets of the 19 compared genomes, we used four gene classifications: OrthoMCL similarity clusters (see above), PFAM domains, KEGG pathways, and Gene Ontology (GO), including different hierarchy levels. A gene was considered to be a member of a given gene class when, respectively, the gene (a) belonged to the given OrthoMCL cluster, (b) contained at least one instance of the given PFAM domain in the encoded protein, (c) belonged to the given KEGG pathway, or (d) was tagged by the given GO label. Using a matrix of gene class counts for each classification type, we identified enrichment comparing two subsets of queried genomes using Fisher’s exact test. Fisher’s exact test was used to detect enrichment of PFAM, KEGG, or GO terms between groups of interest, and p-values were corrected for multiple comparisons [105]. Significant (corrected p-value < 0.05) PFAM and GO terms expansion or depletion was examined for three comparisons: Ajellomycetaceae compared to other Onygenales (S6 Table), pathogenic compared to non-pathogenic from Ajellomycetaceae (S9 Table), and Blastomyces compared to other Ajellomycetaceae; the only terms found to be expanded only in Blastomyces included nucleosome and zinc ion binding. No significant enrichment in KEGG terms was detected for these comparisons.
|
10.1371/journal.pgen.1006273 | Glucose Sensor MdHXK1 Phosphorylates and Stabilizes MdbHLH3 to Promote Anthocyanin Biosynthesis in Apple | Glucose induces anthocyanin accumulation in many plant species; however, the molecular mechanism involved in this process remains largely unknown. Here, we found that apple hexokinase MdHXK1, a glucose sensor, was involved in sensing exogenous glucose and regulating anthocyanin biosynthesis. In vitro and in vivo assays suggested that MdHXK1 interacted directly with and phosphorylated an anthocyanin-associated bHLH transcription factor (TF) MdbHLH3 at its Ser361 site in response to glucose. Furthermore, both the hexokinase_2 domain and signal peptide are crucial for the MdHXK1-mediated phosphorylation of MdbHLH3. Moreover, phosphorylation modification stabilized MdbHLH3 protein and enhanced its transcription of the anthocyanin biosynthesis genes, thereby increasing anthocyanin biosynthesis. Finally, a series of transgenic analyses in apple calli and fruits demonstrated that MdHXK1 controlled glucose-induced anthocyanin accumulation at least partially, if not completely, via regulating MdbHLH3. Overall, our findings provide new insights into the mechanism of the glucose sensor HXK1 modulation of anthocyanin accumulation, which occur by directly regulating the anthocyanin-related bHLH TFs in response to a glucose signal in plants.
| Glucose is considered as a major regulatory molecule in addition to being essential metabolic nutrients and structural components in higher plants. As is well known, hexokinase1 (HXK1) is a glucose sensor that integrates diverse signals to govern gene expression and plant growth in response to environmental cues. Previously, it is reported that the nuclear HXK1 forms a glucose signaling complex core with the vacuolar H+-ATPase B1 (VHA-B1) and the 19S regulatory particle of proteasome subunit (RPT5B), which influences the transcription of target genes. However, it is yet unknown if and how HXK1 directly targets TFs to modulate their function in the nucleus in plants. Our results reveal the important roles of MdHXK1 protein kinase in phosphorylating MdbHLH3 TF to modulate anthocyanins accumulation in response to glucose in apple.
| In higher plants, sugars function as major regulatory molecules in addition to being essential metabolic nutrients and structural components. Sugars control gene expression to affect developmental and metabolic processes during the entire plant life cycle and function in response to biotic and abiotic stresses [1–3]. Therefore, rigorous sugar-sensing and sugar-signaling systems are critical for coordinating photosynthesis and carbon metabolism and for adapting to changes in environmental conditions to sustain normal plant growth and development.
Among the myriad of sugars in photosynthesis, glucose is the preferred carbon and energy source. Glucose is involved in many metabolic pathways, including the glycolytic process, in organisms ranging from unicellular microbes to plants and animals [4,5]. In addition to its metabolic function, glucose is the most intensively studied sugar molecule and functions in specific regulatory pathways to modulate plant growth and development [6,7]. Glucose signaling modulates the gene expression of enzymes in the glyoxylate cycle and photosynthesis pathway, and is also involved in the decision of whether to initiate the normal seedling establishment after seed germination [8,9].
Hexokinase 1 (HXK1) is the first plant sugar sensor identified [9,10]. The genetic evidence for HXK1 as a sugar sensor is the isolation of two Arabidopsis gin2 (glucose insensitive 2) mutants, both of which are mapped to the HXK1 gene [11]. In the Arabidopsis genome, there are three HXKs and three HXK-like (HKLs) genes, which execute a variety of physiological functions, including controlling subcellular localization, protein complex formation and tissue-specific expression patterns [12–14]. Moreover, five orthologous HXKs have been identified in the apple genome. Among them, MdHXK1, a well-known apple hexokinase, is highly homologous with Arabidopsis AtHXK1 [15]. Generally, HXKs are located on the outer mitochondrial membrane, plastids and even in the nucleus [13,14,16].
The regulatory role of HXK1 in sugar signaling has been identified and characterized in plants in the past two decades. In Arabidopsis, HXK1 forms a high-molecular-weight complex together with the V-ATPase subunit VHA-B1 and the proteasome 19S regulatory subunit RPT5B in the nucleus. This complex directly binds to the promoters of CAB2 (chlorophyll a/b binding protein 2) and CAB3 genes to confer glucose-mediated transcriptional regulation independent of glucose metabolism in the cytosol [17]. Both seedlings and adult plants of vha-b1 and rpt5b mutants display similar phenotypes as the gin2 mutant, demonstrating the crucial role of the interaction with HXK1 in glucose signaling [11,17]. In addition, glucose signaling mediated by HXK1 shows crosstalk with ABA, ethylene, auxin, cytokinin and brassinosteroid signaling [18–20]. However, whether HXK1-mediated signaling is involved in the regulation of anthocyanin biosynthesis in plants remains unclear.
Anthocyanins are ubiquitously present in various tissues and organs of plants, especially in the fruit, leaf and flower of ornamental crops. They are responsible for the red, purple and blue coloration of tissues and organs depending on the cellular conditions, such as pH value [21]. Colored organs, such as flowers and fruits, attract pollinators and seed-dispersing animals [22]. Anthocyanins are also antioxidant molecules that protect plants from damage by reactive oxygen species (ROS) [23–25]. These properties also make them interesting as food ingredients for human and animal nutrition. Anthocyanins are biosynthesized via the flavonoid pathway in the cytosol and are transported into the vacuole by vacuolar transporters, including ABC and MATE-type transporters [26,27].
The flavonoid biosynthetic pathway is transcriptionally controlled by a regulatory MYB-bHLH-WDR (WBM) complex containing WD-repeat proteins, basic helix-loop-helix bHLH and MYB transcription factors (TFs), which are highly conserved among higher plant species [28–31]. As the crucial components of the WBM complex, bHLH TFs promote anthocyanin biosynthesis by directly binding to the promoters of not only anthocyanin structural genes, such as DFR and UFGT, but also anthocyanin-associated MYB TF genes to activate their expression [31–34]. Interestingly, MdbHLH3 protein promotes anthocyanin accumulation partially through a putative phosphorylation modification in response to low temperature in apple [32]. However, the protein kinase that mediates the phosphorylation of MdbHLH3 protein is unknown.
Sugars induce anthocyanin biosynthesis in various plant species [35–37]. First, they provide carbon sources, skeletons and glucosides for anthocyanin biosynthesis [38,39]. Second, they increase the expression levels of biosynthetic structural genes and regulatory MYB genes [37,40]; however, the precise mechanism by which sugars modulate these genes remains unknown. The present study found that a protein kinase, MdHXK1, is involved in the regulation of anthocyanin biosynthesis in response to glucose by interacting with the phosphorylating and stabilizing MdbHLH3 protein. Subsequently, the function of MdHXK1 in the modulation of anthocyanin accumulation was characterized in apple calli and fruits. Finally, the potential application of HXK1-mediated glucose signaling in the genetic improvement of horticultural traits is discussed.
Previous studies have verified that glucose significantly induces anthocyanin biosynthesis in Arabidopsis seedlings [36]. Similarly, the effect of different concentrations of glucose (0–6%, w/v) and the HXK inhibitor glucosamine on anthocyanins accumulation was tested in in vitro shoot cultures of the ‘Gala’ apple cultivar. The results showed that glucose promotes anthocyanin accumulation in an HXK-dependent manner in apple (S1 Fig; S5 Text).
Because glucose controls anthocyanin accumulation in an HXK-dependent pathway in apple, it is reasonable to propose that this process is regulated by the catalytic or signaling function of HXK. We isolated the MdHXK1 gene from apple to investigate this possibility. The predicted MdHXK1 protein is highly homologous with AtHXK1, which functions as not only a catalytically active kinase but also a glucose sensor in Arabidopsis [11,41]. Two catalytically inactive HXK1 mutants have been identified in Arabidopsis, namely, HXK1S177A and HXK1G104D; these mutants retain signaling functions but not catalytic activities [11]. To investigate whether Ser and Gly at positions 177 and 104 of HXK1, respectively, are conserved between apple and Arabidopsis, an alignment of the amino acid sequence of MdHXK1 with AtHXK1 was performed. The result showed that apple MdHXK1 protein exhibited a high sequence similarity (77.31% identity) to Arabidopsis AtHXK1 (S2 Fig). The positions of 177 and 104 of apple MdHXK1 are the Ser and Gly residues, respectively, which were the same as those of Arabidopsis AtHXK1 (Fig 1A). These results suggest that the catalytically inactive apple MdHXK1 proteins MdHXK1S177A and MdHXK1G104D also exercised their functions in signaling but not catalytic activities, similar to Arabidopsis.
To rapidly determine whether the catalytically important G104 and S177 is involved in glucose-induced anthocyanin accumulation, two point mutations, i.e., MdHXK1G104D and MdHXK1S177A, were made to the MdHXK1 protein. A total of three types of 35S-driven vectors of 35S::MdHXK1, 35S::MdHXK1S177A and 35S::MdHXK1G104D were generated and genetically transformed into apple calli of the 'Orin' cultivar (Fig 1B). Subsequently, these three transgenic apple calli and the wild-type (WT) control were used for immunoblotting assays with an anti-MdHXK1 antibody. The result demonstrated that the protein abundance of MdHXK1 was increased by 4.2-, 3.9- and 4.0-fold in the 35S::MdHXK1, 35S::MdHXK1G104D and 35S::MdHXK1S177A transgenic apple calli, respectively, compared with the WT control (Fig 1C), indicating that the target genes were successfully transformed into and expressed in the transgenic apple calli. In addition, qPCR assays showed that MdHXK1 repressed two classes of photosynthesis genes including MdSBP and MdCAA in 35S::MdHXK1, 35S::MdHXK1G104D and 35S::MdHXK1S177A transgenic apple calli but not WT apple calli (Fig 1D), suggesting that Gly104 and Ser177 mutations have similar effect on MdHXK1 as Arabidopsis AtHXK1.
As a result, the three transgenic apple calli produced nearly the same levels of anthocyanins to each other but at a considerably higher level than in the WT control under 6% glucose conditions (Fig 1B and 1E), indicating that MdHXK1 and two point mutants successfully function to promote anthocyanin accumulation in these transgenic apple calli. In addition, the glucose phosphorylation activities were determined for the WT and these transgenic apple calli. The results showed that the 35S::MdHXK1 transgenic calli exhibited higher glucose phosphorylation activity than the 35S::MdHXK1S177A and 35S::MdHXK1G104D calli and the WT control (Fig 1F). However, there was no significant difference in glucose phosphorylation activities among the 35S::MdHXK1S177A and 35S::MdHXK1G104D transgenic calli and the WT control (Fig 1F). Collectively, these results suggest that MdHXK1 modulates anthocyanin accumulation mainly through glucose signaling, but not the catalytic pathway, under the high-glucose (6%) conditions.
Furthermore, the WT and aforementioned three transgenic calli were also treated with a low glucose concentration (1%) to induce anthocyanin accumulation. The result demonstrated that the 35S::MdHXK1S177A and 35S::MdHXK1G104D transgenic apple calli produced more anthocyanins than the WT controls but less anthocyanins than the 35S::MdHXK1 transgenic calli (S3A and S3B Fig). However, the glucose phosphorylation activities of the 35S::MdHXK1G104D and 35S::MdHXK1S177A transgenic apple calli showed no significant difference compared with the WT control but were considerably lower than the activities for the 35S::MdHXK1 transgenic calli (S3C Fig).
Taken together, these results indicate that MdHXK1 induces anthocyanin accumulation depending on both the catalytic activity and signaling under low-glucose conditions but mainly depending on signaling under high-glucose conditions.
To screen the target protein of MdHXK1 in its signal pathway, the 35S::MdHXK1-Myc vector was constructed and genetically transformed into apple calli (S4A Fig). The 35S::MdHXK1-Myc transgenic calli were used for co-immunoprecipitation (Co-IP) against the monoclonal anti-Myc antibody (S4B Fig). Subsequently, the Co-IPed proteins were analyzed with LC/MS to identify the potential proteins that interact with the MdHXK1 protein. The results showed that the anthocyanin-associated bHLH TF MdbHLH3 is a candidate (S1 Text).
To determine whether MdHXK1 interacts with MdbHLH3 protein, yeast two-hybrid (Y2H) assays were performed. MdHXK1 protein contains two conserved hexokinase domains, i.e., hexokinase_1 and hexokinase_2 (S4C Fig). Therefore, the full-length cDNA of MdHXK1 gene was divided into two fragments, i.e., MdHXK11-245aa and MdHXK1245-498aa. Subsequently, the full-length cDNA and two truncated mutants of the MdHXK1 gene were inserted into the pGBT9 vector, independently, as the bait vectors. Moreover, the full-length MdbHLH3 cDNA and its serially truncated mutants, as previously reported by Xie et al. [32], were inserted into the pGAD424 vector as the prey vectors. The different combinations of bait and prey vectors were transformed into yeast for Y2H assays. The results indicated that the full-length MdHXK1 strongly interacted with the full-length MdbHLH3 proteins. Furthermore, the truncated peptide MdbHLH3346-709aa, i.e., the C-terminus of MdbHLH3, interacted with MdHXK1 proteins at the hexokinase_2 domain MdHXK1245-498aa but not at the hexokinase_1 domain MdHXK11-245aa (Fig 2A).
To further verify the interaction between MdHXK1 and MdbHLH3, an in vivo Co-IP assay using 35S::MdbHLH3-GFP transgenic apple calli was conducted. The result indicated that the MdbHLH3-GFP fusion protein, but not the GFP negative control, interacted with MdHXK1 in apple calli (Fig 2B). In addition, a GST pull-down assay showed that a GST-tagged MdbHLH3 physically interacted with a His-tagged MdHXK1 in vitro (Fig 2C). These results indicate that the hexokinase_2 domain of MdHXK1 physically interacts with the C-terminus of the MdbHLH3 protein.
To examine how MdbHLH3 protein responds to glucose, an expression vector 35S::MdbHLH3-Myc was constructed and genetically transformed into apple calli (S5A Fig). After treatment with or without 6% glucose, the 35S::MdbHLH3-Myc overexpressing calli were used for Western blotting with the anti-Myc antibody. The results showed that the position of the MdbHLH3 proteins shifted from a faster- to a slower-migrating band in the transgenic apple calli treated with 6% glucose compared to those without glucose (Fig 3A), indicating that the MdbHLH3 protein was post-translationally modified in response to glucose. Furthermore, treatment with calf intestine alkaline phosphatase (CIP), which cleaves exposed phosphate residues from ribonucleotides and deoxyribonucleotides, converted the slower-migrating form of MdbHLH3 to the faster-migrating form (Fig 3A), indicating that the glucose-induced post-translational modification for the MdbHLH3 protein in apple calli is predominantly a phosphorylation.
To examine the potential phosphorylation sites of the MdbHLH3 protein, the glucose-induced phosphorylated MdbHLH3 protein in the 35S::MdbHLH3-Myc overexpressing calli was captured with anti-Myc antibody-conjugated agarose beads and separated in an SDS-PAGE gel. After proteolytic digestion and purification, the protein sample was subjected to liquid chromatography-tandem mass spectrometry (LC-MS/MS) to detect the phosphorylation sites. The serine at residue 361 (Ser361) of the MdbHLH3 protein exhibited a high phosphopeptide signal intensity (Fig 3B; S2 and S4 Texts), suggesting that it is a potential phosphorylation site.
Subsequently, a monoclonal antibody specifically against the MdbHLH3 phosphorylation site at residue 361 was prepared and named as the anti-MdbHLH3S361 antibody (S5B Fig). This antibody specifically recognized the glucose-induced phosphorylation of MdbHLH3 protein in the WT apple calli (Fig 3C), which was consistent with the results shown in Fig 3A. These results indicate that glucose induces the phosphorylation of the MdbHLH3 protein at the Ser361 site in apple calli.
To examine whether the glucose concentration influences the phosphorylation of the MdbHLH3 protein, the WT apple calli were treated for 30 min with glucose concentrations of 0%, 1%, 3% and 6% and then used for immunoblotting with the anti-MdbHLH3S361 antibody. The results showed that the MdbHLH3 protein was not phosphorylated when the calli grew in absence of glucose, whereas the phosphorylation intensity of the MdbHLH3 protein increased gradually with glucose concentration (Fig 3D). Moreover, apple calli were treated with 6% glucose for different times (0, 10, 20, 30 and 60 min) to examine whether treatment time affects the phosphorylation of the MdbHLH3 protein. The results showed that the phosphorylation intensity of MdbHLH3 proteins in the calli gradually increased with the treatment duration (Fig 3E). These results indicate that the MdbHLH3 protein is phosphorylated in response to glucose and that this modification is positively associated with glucose concentration and treatment time.
In addition, glucose-induced phosphorylation of the MdbHLH3 protein could be observed in apple leaves (S5C Fig), indicating that the glucose-induced phosphorylation of the MdbHLH3 protein occurred in different apple tissues and organs.
Considering the interaction between MdHXK1 and MdbHLH3 proteins, it is reasonable to hypothesize that the MdHXK1 protein kinase mediates the phosphorylation of MdbHLH3 protein in apple calli. To verify this hypothesis, new transgenic apple calli, 35S::antiMdHXK1, were obtained, which contained an antisense fragment specific to MdHXK1 cDNA and exhibited considerably lower transcript and protein levels of MdHXK1 than the WT control (S6A and S6B Fig). Subsequently, immunoblotting assays with the anti-MdbHLH3S361 antibody were performed using the WT control and the 35S::MdHXK1 and 35S::antiMdHXK1 transgenic apple calli after treatment with or without glucose. The result showed that the 35S::MdHXK1 overexpressing calli exhibited a considerably higher phosphorylation level of the MdbHLH3 protein, whereas that of the 35S::antiMdHXK1-suppressing calli were lower than the WT control in response to glucose treatment (Fig 4A). This result suggests that the MdHXK1 protein kinase is necessary, if not sufficient, for the glucose-induced phosphorylation of the MdbHLH3 protein in apple calli.
To further verify that MdHXK1 directly phosphorylates the MdbHLH3 protein, in-gel assays were conducted using prokaryon-expressed and purified MdHXK1-GST and MdbHLH3-His fusion proteins. As a result, MdbHLH3 protein was phosphorylated by the recombined MdHXK1 (Fig 4B). Furthermore, this in vitro phosphorylation assays were performed with anti-MdbHLH3S361 antibody. The result showed that MdbHLH3-His proteins were phosphorylated by MdHXK1, while MdbHLH3 mutation MdbHLH3S361A-His were not (S7 Fig). These results demonstrated that the MdbHLH3 protein is a direct substrate of the MdHXK1 protein kinase.
In addition, the phosphorylation status of the MdbHLH3 protein was determined in the glucose-treated 35S::MdHXK1-, 35S::MdHXK1G104D- and 35S::MdHXK1S177A-overexpressing apple calli lines. Interestingly, there was no visible difference in the phosphorylation levels of these three transgenic calli (Fig 4C), indicating that the abolishment of MdHXK1 catalytic function as indicated by the phosphorylation activity is unable to affect the phosphorylation level of the MdbHLH3 protein.
To further examine which kinase domain functions to phosphorylate the MdbHLH3 protein, vectors were constructed to contain the truncated MdHXK1 cDNA fragments MdHXK135-242aa and MdHXK1245-491aa, which encode hexokinase_1 and hexokinase_2 domains, respectively. The resulting vectors 35S::MdHXK135-242aa-Myc and 35S::MdHXK1245-491aa-Myc were genetically transformed into the WT apple calli, independently. Subsequently, the 35S::MdHXK135-242aa-Myc and 35S::MdHXK1245-491aa-Myc transgenic apple calli were used for immunoblotting assays with anti-Myc and anti-MdbHLH3S361 antibodies, respectively. The results showed that the truncated proteins MdHXK135-242aa and MdHXK1245-491aa were successfully expressed in the 2 transgenic calli. However, there was no visible difference in the phosphorylation level of MdbHLH3 between the WT control and 2 transgenic calli, i.e., 35S::MdHXK135-242aa-Myc and 35S::MdHXK1245-491aa-Myc (Fig 4D).
In addition to the hexokinase_1 and hexokinase_2 domains, MdHXK1 also contains a signal peptide ranging from 1 to 22 amino acid residues at the N-terminus (S4C Fig). Given that signal peptides are polypeptide chains that are used as ‘address labels’ for sorting proteins to their correct subcellular destinations, it was hypothesized that the signal peptide of MdHXK1 is involved in the MdbHLH3 phosphorylation process. To verify this hypothesis, three vectors of MdHXK1 cDNA including the signal peptide domain, i.e., 35S::MdHXK11-21aa+35-242aa-Myc, 35S::MdHXK11-21aa+245-491aa-Myc and 35S::MdHXK11-499aa-Myc, were constructed and successfully transformed into the WT apple calli (Fig 4D). The resulting transgenic calli were used for immunoblotting assays with anti-Myc and anti-MdbHLH3S361 antibodies. The results showed that the phosphorylation intensities of MdbHLH3 proteins were considerably higher in the 35S::MdHXK11-21aa+245-491aa-Myc and 35S::MdHXK11-499aa-Myc transgenic calli than in the WT control. However, the level of MdbHLH3 phosphorylation was highly similar in the 35S::MdHXK11-21aa+35-242aa-Myc transgenic calli as in the WT control (Fig 4D). Therefore, the signal peptide and hexokinase_2 domain are crucial for MdHXK1-mediated phosphorylation of the MdbHLH3 protein.
To further verify the roles of the signal peptide and hexokinase_2 domain in the MdHXK1 protein on the MdbHLH3 phosphorylation process, a series of 35S promoter-driven vectors that express fluorescence-tagged fusion proteins, including MdHXK135-242aa-GFP, MdHXK1245-491aa-GFP, MdHXK11-21aa+35-242aa-GFP, MdHXK11-21aa+245-491aa-GFP, MdHXK11-499aa-GFP and MdbHLH3-RFP, were constructed and used to determine their cellular distribution using an apple protoplast system. Upon co-transfection of the MdHXK1-related GFP fusion genes together with the MdbHLH3-RFP fusion gene into the apple protoplasts, the transformant protoplasts were observed in a subcellular localization assay using a laser confocal microscope. The results showed that MdHXK11-499aa-GFP was co-localized with MdbHLH3-RFP in the nucleus (Fig 4E). Moreover, similar to MdHXK11-499aa-GFP, MdHXK11-21aa+245-491aa-GFP together with MdbHLH3-RFP resided in the nucleus, whereas other truncated peptides, including MdHXK135-242aa-GFP, MdHXK1245-491aa-GFP and MdHXK11-21aa+35-242aa-GFP, were not co-localized with MdbHLH3-RFP in the nucleus (Fig 4E).
Taken together, the signal peptide and hexokinase_2 domain of the MdHXK1 protein are essential for its nuclear co-localization together with the MdbHLH3 protein, which is crucial for MdHXK1-mediated phosphorylation of the MdbHLH3 protein.
To examine whether MdHXK1 influences the stability of MdbHLH3 proteins, the prokaryon-expressed and purified MdbHLH3-GST fusion proteins were incubated with plant total proteins that were extracted from the WT control and the 35S::MdHXK1 and 35S::antiMdHXK1 transgenic apple calli. Subsequently, protein degradation assays were performed. The results showed that MdbHLH3-GST proteins were more stable in the protein extracts of the 35S::MdHXK1 transgenic calli than in those of the WT control (Fig 5A, 5B and 5E), whereas they were degraded at a more rapid speed in the protein extracts of 35S::antiMdHXK1 transgenic calli compared to those of the WT control (Fig 5C and 5E). These results suggest that MdHXK1-mediated phosphorylation of the MdbHLH3 protein may increase its stability.
To further verify that phosphorylation influences the stability of the MdbHLH3 protein, a site-directed S361A mutation was introduced into the MdbHLH3 protein. The mutated cDNA MdbHLH3S361A was inserted into the expression vector for prokaryon-expression and purification of MdbHLH3S361A-GST fusion proteins, which were then incubated with the total proteins extracted from the WT calli. The protein sample was used for Western blotting with the anti-GST antibody. The results showed that the MdbHLH3S361A-GST proteins degraded at a rapid speed compared with the wild-type MdbHLH3-GST proteins (Figs 4B, 5D and 5E), indicating that the inhibition of phosphorylation promoted the degradation of MdbHLH3 proteins. In addition, MdHXK1 also enhanced the stability of the endogenous MdbHLH3 proteins (S8A–S8D Fig).
To examine whether phosphorylation of the MdbHLH3 protein influences its binding capacity to the downstream genes, such as MdMYB1, MdANS and MdUFGT, the 35S::MdbHLH3-Myc and 35S::MdbHLH3S361A-Myc transgenic apple calli were used for ChIP-PCR analysis (Fig 5F; S9 Fig). The results showed that the phosphorylated MdbHLH3-Myc protein exhibited a higher enrichment in the promoters of MdMYB1 and anthocyanins biosynthetic structural genes than the non-phosphorylated MdbHLH3S361A-Myc (Fig 5G). As a result, those genes showed higher expression levels in the 35S::MdbHLH3-Myc transgenic apple calli than the MdbHLH3S361A-Myc apple calli (Fig 5H). Furthermore, the abundance of the endogenous MdbHLH3 and MdMYB1 proteins were higher in 35S::MdHXK1 overexpressing calli but lower in 35S::antiMdHXK1 suppressing calli than in the WT control (Fig 5I).
Therefore, phosphorylation modification stabilizes the MdbHLH3 protein and enhances its transcriptional activation of downstream genes.
To examine whether and how MdHXK1 influences anthocyanin accumulation, the full-length sense ORFs and antisense cDNA fragments of MdHXK1 and MdbHLH3 (or MdbHLH3S361A) were inserted into the expression vectors downstream of 35S promoters independently. The resulting vectors were then transformed into apple calli. In the present study, we obtained nine types of transgenic apple calli, namely, 35S::MdHXK1, 35S::MdbHLH3, 35S::MdbHLH3S361A, 35S::MdHXK1+35S::MdbHLH3, 35S::MdHXK1+35S::MdbHLH3S361A, 35S::MdHXK1+35S::antiMdbHLH3, 35S::antiMdHXK1+35S::MdbHLH3, 35S::antiMdHXK1+35S::MdbHLH3S361A and 35S::antiMdHXK1+35S::antiMdbHLH3 (Fig 6A). The MdHXK1 and MdbHLH3 genes were successfully overexpressed or suppressed in the corresponding calli compared with the WT control (Fig 6B), indicating that the genetic transformation was successful in apple calli. As downstream genes, the transcript levels of MdANS and MdUFGT genes were positively correlated with that of the MdbHLH3 gene; however, MdANS and MdUFGT were considerably lower in the 35S::MdbHLH3S361A transgenic calli than in the 35S::MdbHLH3 calli (Fig 6B). In addition, the transcription activity of the MdANS promoter was positively associated with the transcript level of MdHXK1 genes (S10 Fig).
These transgenic apple calli were used to determine the anthocyanin content. The results showed that overexpression of MdHXK1 and MdbHLH3, either alone or together, noticeably enhanced the anthocyanin content in the corresponding transgenic calli compared with the WT control (Fig 6C). Moreover, the 35S::MdbHLH3S361A transgenic calli produced less anthocyanins than the 35S::MdbHLH3 calli (Fig 6C), indicating that the phosphorylation site Ser361 is crucial for MdbHLH3 to regulate the biosynthesis of anthocyanins.
Furthermore, the 35S::MdHXK1 transgenic calli produced more anthocyanins, but the 35S::MdHXK1+35S::antiMdbHLH3 calli produced less than the WT control (Fig 6A and 6C), indicating that the suppression of the MdbHLH3 gene inhibited the MdHXK1-mediated increase of anthocyanin biosynthesis. Therefore, MdHXK1 regulates anthocyanin accumulation at least partially, if not completely, depending on the presence of MdbHLH3.
To investigate whether MdHXK1 and MdbHLH3 regulate anthocyanin accumulation in apple fruits in a similar manner as in calli, a viral vector-based method was applied to alter their expression using vector pRI for overexpression and vector TRV for suppression. Four viral constructs, including pRI-MdHXK1, TRV-MdHXK1, pRI-MdbHLH3 and TRV-MdbHLH3, were obtained. Each construct and two combinations, i.e., TRV-MdHXK1+pRI-MdbHLH3 and pRI-MdHXK1+TRV-MdbHLH3, were used for fruit infiltration, with the empty vectors as controls (Fig 7A). The results showed that the transcript levels of MdHXK1 and MdbHLH3 genes were enhanced after being infiltrated with pRI-MdHXK1 and pRI-MdbHLH3 but decreased with TRV-MdHXK1 and TRV-MdbHLH3, respectively (Fig 7B).
Subsequently, anthocyanin levels were measured in apple peel tissues around the sites infiltrated with the different viral constructs. The results showed that both MdHXK1 and MdbHLH3 positively regulate anthocyanin accumulation and that the MdHXK1-mediated anthocyanin accumulation required MdbHLH3 in apple fruits (Fig 7C), similar to the apple calli (Fig 6C).
Sugar-induced anthocyanin accumulation is important for not only proper cell function [23,24] but also the quality formation of ornamental crops and fresh fruits [37,40,42]. Therefore, it is critical to elucidate the molecular mechanism underlying sugar-induced anthocyanin accumulation. The present study found that the glucose sensor MdHXK1, a hexokinase protein, stabilized the bHLH TF MdbHLH3 by phosphorylation modification, leading to an enhanced anthocyanin accumulation in apple.
Sugars are the major sources of carbon and energy metabolites and play key roles in plant growth and development. Sugars also act as effective signaling molecules throughout plant life [43,44]. In Arabidopsis, HXK1 is a crucial enzyme in glucose catabolism; HXK1 senses glucose and initiates its signaling pathway [11]. Glucose-promoted aliphatic glucosinolate biosynthesis is regulated by HXK1-mediated signaling via the MYB TFs MYB28 and MYB29 [45]. Most recently, it was reported that glucose treatment greatly enhances anthocyanin content and induces the expression of PsWD40-2, PsMYB2, PsCHS1, PsCHI1 and PsF3’H1 through glucose signaling in Paeonia suffruticosa cut flowers [37]. Among the WBM genes, MYB and WD40 genes, but not bHLH genes, are induced at the transcriptional level by glucose. The present study found that a glucose-dependent signaling pathway is involved in the regulation of anthocyanin accumulation in apple. This process depended on functional MdHXK1, which directly phosphorylated and stabilized the WBM component MdbHLH3 protein at the post-translational level (Figs 6A–6C and 7A–7C). MdbHLH3 modulates both anthocyanin biosynthetic structural genes and the regulatory MdMYB1 gene, thereby promoting anthocyanin accumulation [32].
Furthermore, anthocyanin accumulation is induced by glucose, which is not due to the osmotic effects of glucose (S1C and S1D Fig) [46]. MdHXK1 promoted anthocyanins accumulation mainly via the glucose signaling pathway under the high-glucose condition (Fig 1B–1F) and via both glucose metabolism and the signaling pathway under the low-glucose condition (S3 Fig). Given that sugars are the major sources of carbon and energy, higher plants require sugars for normal metabolism [9]. Glucose provides carbon skeletons for anthocyanin biosynthesis via its HXK1-dependent catalytic metabolism pathway, especially under low-glucose conditions (S3 Fig) [47]. Moreover, the MdHXK1-dependent glucose signaling pathway also plays a vital role in anthocyanin biosynthesis (Fig 1B–1E; S3 Fig). Therefore, glucose promotes anthocyanin biosynthesis depending on both signaling and metabolism under low-glucose conditions in apple. However, when carbohydrates are derived from glucose to meet the needs of anthocyanin synthesis under high-glucose conditions (e.g., 6% glucose), the glucose mainly served as signaling molecules to initiate anthocyanin biosynthesis (Fig 1B–1E). Taken together, glucose-induced anthocyanin accumulation is the result of MdHXK1-dependent glucose signaling together with catalytic metabolism pathways.
In the present study, the catalytic and signaling functions of MdHXK1 were characterized using its two catalytically inactive mutants, MdHXK1S177A and MdHXK1G104D, in apple (Fig 1), both of which retain signaling functions but not catalytic activities, similar to their activities in Arabidopsis [11]. Most recently, Feng et al. [48] successfully resolved the crystal structures of two AtHXK1 inactive forms, AtHXK1S177A and AtHXK1G104D, and analyzed the biochemical properties of AtHXK1 in Arabidopsis. These findings provide biochemical and structural insights into how HXK1 functions at the atomic level, thereby providing a structural explanation for the dual functions of HXK1 in plants.
As the most important glucose sensor, HXK1 is involved in diverse signaling functions, particularly in the regulation of gene expression. In plants, HXK1 is mainly localized in the cytosol; however, a minor degree of HXK1 is also present in the nucleus [49]. In the nucleus, this minor portion of HXK1 interacts with the B1 subunit of the V-ATPase (VHA-B1) and with a 19S regulatory particle of the proteasome subunit (RPTB5), leading to the formation of unexpected nuclear HXK1 complexes [49]. A large number of putative TFs identified in the nuclear HXK1 complexes interact directly with VHA-B1 and/or RPT5B but not directly with HXK1 [49]. In addition, nuclear-localized HXK1 has also been implicated in the control of the transcriptional activity and proteasome-mediated degradation of EIN3 (ethylene-insensitive3), a key transcriptional regulator in ethylene signaling [50]. The present study found that HXK1 directly interacted with MdbHLH3 (Fig 2B–2D), a key bHLH transcriptional regulator in anthocyanin biosynthesis [32]. However, it is unclear whether the HXK1/VHA-B1/RPT5B nuclear complex is also involved in these processes.
Furthermore, a R2R3 MYB regulator MdMYB1 interacts with the N-terminus of MdbHLH3 to regulate anthocyanin biosynthesis [32]. The present study found that the hexokinase_2 domain of MdHXK1 strongly interacted with the C-terminus of MdbHLH3 to modulate anthocyanin accumulation (Fig 2B). Therefore, there is no competition for the interaction of the MdbHLH3 protein with MdMYB1 and MdHXK1 in the regulation of anthocyanin biosynthesis.
In apples, a putative phosphorylation modification is involved in the MdbHLH3-mediated anthocyanin accumulation in response to low temperature [32]. However, the protein kinase that mediates the phosphorylation of MdbHLH3 protein is not yet identified. The present study found that the MdHXK1 protein kinase is directly involved in the glucose-induced phosphorylation of MdbHLH3 protein, thereby modulating anthocyanin biosynthesis (Fig 4A and 4B). In addition, the hexokinase_2 domain of MdHXK1, which may be required for signal peptide cleavage based on its functions in protein secretion and subcellular localization [51,52], plays a key role in the phosphorylation of the MdbHLH3 protein (Fig 4D and 4E).
Additionally, several bHLH TFs are phosphorylated by external environmental stimuli. For example, multiple light-induced Ser/Thr phosphorylation sites are found in the phyB-interacting bHLH TF PIF3 in Arabidopsis [53]. Multisite light-induced phosphorylation of the bHLH TFs PIF1 and PIF5 has been confirmed using photobiological and genetic approaches [54,55]. In addition to PIFs, another bHLH TF, TWIST1, is phosphorylated at Thr125 and Ser127 to control pro-metastatic functions in prostate cancer cells [56]. In contrast to the aforementioned bHLH TFs, the bHLH TF speechless is phosphorylated to promote stomatal development at a single serine 186 site in Arabidopsis [57]. Similarly to the bHLH TF speechless, only a single phosphorylation site in the bHLH TF MdbHLH3 protein was detected in apple (Fig 3B and 3C; S2 Text), suggesting that MdbHLH3 phosphorylation may be a single-site phosphorylation event in apple or at least that its Serine 361 plays a crucial role in anthocyanin biosynthesis (Figs 5D–5H, 6 and 7).
As is well known, the MYB-bHLH-WDR (MBW) complex plays an important role in regulating anthocyanin and proanthocyanidin biosynthesis. In apple, MdbHLH3 physically interacts with MdMYB1 and specifically binds to the promoters of anthocyanin structural genes, such as MdDFR and MdUFGT, to promote anthocyanin accumulation [32]. Moreover, MdbHLH3 interacts with MdMYB9 and MdMYB11 to regulate the JA-induced biosynthesis of anthocyanin and proanthocyanidin [58]. In the present study, MdbHLH3 promoted anthocyanin accumulation in apple calli and apple fruits (Figs 6 and 7). In addition, MdbHLH3 also promotes malate accumulation in the vacuole by indirectly regulating the vacuolar transport system in apple [59]. Similarly, the increase of malate content in 35S::MdHXK1-overexpressing apple calli accumulated more malate than the WT control (S11 Fig), possibly due to the MdHXK1-mediated stabilization of MdbHLH3.
The glucose supply promotes anthocyanin biosynthesis and organ coloration in different plant species, such as Arabidopsis, grape, and Paeonia suffruticosa [36,37,40]. However, the mechanism underlying the glucose signaling-mediated regulation of MYB TFs, WDR and anthocyanin structural genes remains unclear [37,40]. Here, a working model is proposed to illuminate how glucose regulates anthocyanin accumulation in apple (Fig 8). Under glucose deprivation conditions, the kinase activity of MdHXK1 is inactivated and fails to phosphorylate MdbHLH3 protein (Fig 3A and 3C). As a result, a small amount of MdbHLH3 protein binds to the promoters of anthocyanin structural genes, leading to reduced anthocyanin accumulation (Figs 6 and 7). When exposed to glucose, the kinase activity of MdHXK1 is activated, and then, MdHXK1 phosphorylates and stabilizes the MdbHLH3 protein, which further regulates the expression of the anthocyanin biosynthetic genes and the regulatory MYB genes (Figs 4A, 4B, 5 and 6B), ultimately enhancing anthocyanin biosynthesis in apple (Figs 6A, 6C, 7A and 7C). In addition, ectopic expression of the apple MdHXK1 gene also increased anthocyanin accumulation in the transgenic Arabidopsis (S12 Fig), suggesting that the mechanism by which HXK1 controls anthocyanin accumulation in response to glucose is conserved in different species.
In summary, the current study provides new insights into the molecular mechanism of MdHXK1 stabilization of the MdbHLH3 protein, which occurs via phosphorylation, thereby promoting the accumulation of anthocyanins in plant cells in response to glucose signals. Because color is one of the most eye-catching traits for fresh fruits and ornamental plants [60,61], there is considerable interest for the organ coloration in the breeding programs for these economically important plants. Taken together, the regulatory mechanism uncovered in the present study is also useful for the development of novel biotechnological strategies for improving the quality of apple fruit and other horticultural crops.
The in vitro shoot cultures of apple were obtained from detoxified buds of ‘Gala’ apples. They were maintained at 25°C under long-day conditions (16 h light/8 h dark) on Murashige and Skoog (MS) medium supplemented with 0.8 mg L-1 6-BA and 0.2 mg L-1 IAA and subcultured at a 4-week interval before being used for further studies.
The apple calli used in this study were induced from the young embryos of the ‘Orin’ apple (Malus domestica Borkh.). The calli were grown on MS medium containing 0.5 mg of L-1 indole-3-acetic acid (IAA) and 1.5 mg of L-1 6-benzylaminopurine (6-BA) at 25°C in the dark. The apple calli were subcultured three times at 15-day intervals before being used for genetic transformation and in other assays. Additionally, all the apple calli were suffered from a dark (24 hours dark)-induced glucose starvation before being treated with exogenous glucose in this study, unless noted otherwise.
The apple fruits used for the injection of viral vectors were collected from mature trees of the cultivar ‘Red Delicious’ that were grown in a commercial orchard near Tai-An City. Fruits were bagged at 35 DAB (days after blooming); the bagged fruits were harvested at 140 DAB and de-bagged before injection.
The present study used the Arabidopsis (Arabidopsis thaliana) ecotype ‘Columbia,’ the MdHXK1 overexpression line MdHXK1-OVX1, the glucose-insensitive mutant gin2, and the function-complementary line MdHXK1-R1 (overexpression of MdHXK1 in a gin2 mutant background). Seeds were surface sterilized with 70% (v/v) ethanol and sown on 0.8% (w/v) agar plates containing half-strength MS medium and different glucose concentrations. The seeds were stratified for three days at 4°C and transferred into constant light (100 μmol m2 s-1) at 20°C for 2 weeks of growth. Before being used for exogenous glucose treatment, 2-weeks-old Arabidopsis plants were suffered from a dark (24 hours dark)-induced glucose starvation.
To construct MdHXK1 and MdbHLH3 sense overexpressing and antisense suppressing vectors, the full-length cDNA of MdHXK1 and MdbHLH3, a specific fragment of MdHXK1 and a conserved fragment of MdbHLH3 were isolated from ‘Gala’ apple using RT-PCR. Furthermore, truncated sense overexpression vectors, including MdHXK135-242aa, MdHXK1245-491aa, MdHXK11-21aa+35-242aa and MdHXK11-21aa+245-491aa, were also isolated from ‘Gala’ apple using RT-PCR. All of the cDNA were digested with EcoRI/BamHI and cloned into the pRI plant transformation vector downstream of the CaMV 35S promoter. All of the primers used in this study are listed in S3 Text.
In addition, two point mutants of MdHXK1, namely, G104D (altering Glycine to Aspartate at position 104) and S177A (altering Serine to Alanine at position 177), and the MdbHLH3 point mutant Ser361A (mutation of Serine to Alanine at position 361), were obtained using site-directed mutagenesis methods. The resulting cDNA were digested with EcoRI/BamHI and cloned into the pRI plant transformation vector downstream of the CaMV 35S promoter. The primers used in this study are listed in S3 Text.
In addition, the full-length cDNA of MdHXK1 and MdbHLH3 were also cloned into the PRI plant transformation vector with a Myc tag downstream of the CaMV 35S promoter, and subsequently, the recombined expression vectors MdHXK1-Myc and MdbHLH3-Myc were used for genetic transformation.
For apple calli transformation, the constructs, including 35S::MdHXK1, 35S::MdbHLH3, 35S::MdbHLH3S361A, 35S::MdHXK1+35S::MdbHLH3, 35S::MdHXK1+35S::MdbHLH3S361A, 35S::MdHXK1+35S::antiMdbHLH3, 35S::antiMdHXK1+35S::MdbHLH3, 35S::antiMdHXK1+35S::MdbHLH3S361A, 35S::antiMdHXK1+35S::antiMdbHLH3, and 35S::MdbHLH3-Myc, were introduced into ‘Orin’ apple calli using an Agrobacterium-mediated method as described by Hu et al. [59].
For Arabidopsis transformation, the 35S::MdHXK1 vector plasmid was introduced into WT (Col-0) and the glucose-insensitive mutant gin2 via the Agrobacterium strain GV3101 using a floral dip method [59]. The seeds of the transgenic plants were individually harvested and subsequently selfed. Homozygous transgenic lines were used for further investigation.
RNA extraction and quantitative RT-PCR (qRT-PCR) assays were performed with the methods described by Hu et al. [59]. All of the primers used for qRT-PCR are listed in S3 Text.
Protein extraction and Western blotting assays were conducted as described by Hu et al. [59]. The monoclonal antibody of anti-MdHXK1, anti-MdbHLH3S361 (specifically against the MdbHLH3 phosphorylation site at residue 361) and anti-GST antibody were prepared by the Abmart Company (Shanghai, China).
Yeast two-hybrid assays were performed using the Matchmaker GAL4-based two-hybrid system (Clontech, Palo Alto, CA, USA). Full-length cDNA and truncated mutants of MdHXK1, including MdHXK11-245 aa and MdHXK1245-498aa, were inserted into the pGBT9 vector. The associated yeast two-hybrid vectors of MdbHLH3, which were inserted into vector pGAD424, are detailed in Xie et al. [32]. All of the constructs were transformed into yeast strain AH109 using a lithium acetate method. Yeast cells were cultured on minimal medium -Leu/-Trp according to the manufacturer’s instructions. Transformed colonies were plated onto minimal medium -Leu/-Trp/-His/-Ade with or without β-galactosidase to test for possible interactions.
The WT and 35S::MdbHLH3-GFP transgenic apple calli were treated with 50 μM MG132 for 16 h to stabilize the MdbHLH3-GFP and MdHXK1 proteins. The Co-IP was carried out as described by Oh et al. [62]. The eluted samples were immunoblotted using anti-GFP and anti-MdHXK1 antibodies.
For the GST pull-down assays, full-length cDNA of MdbHLH3 were inserted into the pGEX-4T-1 vector, whereas that of MdHXK1 was inserted into pET-32a. All of the recombinant proteins were used to perform GST pull-down assays as described by Oh et al. [62].
MdbHLH3 proteins were immunoprecipitated with anti-MdbHLH3 antibody-conjugated agarose beads and then separated on an SDS-PAGE gel and stained with Coomassie brilliant blue (CBB). The band containing phosphorylated MdbHLH3 protein was cut from the stained SDS-PAGE gel. The protein digestion, phosphopeptide enrichment, mass spectrometry data acquisition, data analysis, and label-free quantitation were carried out as described by Wang et al. [63].
The MdbHLH3-Myc transgenic apple calli were pre-incubated in MS medium plus 6% glucose with or without 5 U of calf intestine alkaline phosphatase (CIP) for 1 and 3 h. Subsequently, proteins extraction was performed for Western blotting assays with an anti-Myc antibody. Actin served as a protein-loading control.
A total of 0.2 μg of recombinant His-tagged protein kinase MdHXK1 and 1 μg of MdbHLH3S361A-GST and normal MdbHLH3-GST proteins were incubated in 25 μL of reaction buffer [20 mM Tris (pH 7.5), 5 mM MgCl2, 10 mM NaCl and 2 mM DTT] with 100 μM ATP and [λ-32P]ATP (0.2 mCi per reaction) at room temperature for 30 min. Recombinant MdbHLH3S361A-GST was served as a negative control in the in vitro kinase assay. The phosphorylated proteins were visualized using autoradiography after separation on a 12% SDS-PAGE gel.
To construct antisense expression viral vectors, a specific fragment of MdHXK1 and a conserved fragment of MdbHLH3 were amplified with RT-PCR using apple fruit cDNA as the template. The resulting products were cloned into the tobacco rattle virus (TRV) vector in the antisense orientation under the control of the dual 35S promoter. The vectors were named TRV-MdHXK1 and TRV-MdbHLH3. To generate overexpression viral vectors, full-length cDNA of MdHXK1 and MdbHLH3 were inserted into the IL-60 vector under the control of the 35S promoter. The resulting vectors were named MdHXK1-IL and MdbHLH3-IL.
The antisense expression viral vectors were transformed into Agrobacterium tumefaciens strain GV3101 for inoculations. Fruit infiltrations were performed as previously described [59]. The injected apple fruits were kept in the dark at room temperature for two days and subsequently placed in the highlight at 10°C for one week. The peel of the injected part was then harvested for gene expression analysis and anthocyanin content determination.
Glucose phosphorylation activity was measured using an enzyme-linked assay according to Schaffer and Petreikov [64]. The assays contained a total volume of 1 mL of 30 mM HEPES-NaOH, pH 7.5, 2 mM MgCl2, 0.6 mM EDTA, 9 mM KCl, 1 mM NAD, 1 mM ATP, and 1 unit of NAD-dependent glucose-6-phosphate dehydrogenase (G6PDH). To assay glucose phosphorylation, 25 μL of the desalted extract was added to start the reaction under 25°C. Reduction of NAD within 5 min was monitored by the increase in absorption at 340 nm. Activity was calculated in terms of μmol of NAD reduced per minute.
Protoplasts isolated apple calli cells were prepared and transformed as described by Sheen [64]. The fluorescence in transformed cells was detected with a confocal laser scanning microscope (Zeiss LSM 510 META), with excitation wavelengths of 488 nm and 543 nm using an argon laser and an emission wavelength of 505–530 nm and over 560 nm using a BP filter or excitation wavelengths of 458 nm and 514 nm using an argon laser, and an emission wavelength of 475–525 nm and 530–600 nm using a BP filter. A total of 20–30 cells were imaged for each experiment. Co-expressed proteins in the same protoplasts of apple calli cells were detected in the same Pinhole.
The total anthocyanins were extracted using a methanol-HCl method and detected as described by Hu et al. [59].
Samples were analyzed in triplicates, and the data are expressed as the mean ± standard deviation unless noted otherwise. Statistical significance was determined using Student’s t-test. A difference at P≤0.01 was considered significant, and P≤0.001 was considered extremely significant.
|
10.1371/journal.pntd.0002059 | Dog Bite Histories and Response to Incidents in Canine Rabies-Enzootic KwaZulu-Natal, South Africa | The objective of this paper is to report evaluated observations from survey records captured through a cross-sectional observational study regarding canine populations and dog owners in rabies enzootic KwaZulu-Natal province, South Africa. Our aim was to evaluate respondent knowledge of canine rabies and response to dog bite incidents towards improved rabies control. Six communities consisting of three land use types were randomly sampled from September 2009 to January 2011, using a cluster design. A total of 1992 household records were analyzed using descriptive statistics and regression modeling to evaluate source of rabies knowledge, experiences with dog bites, and factors affecting treatment received within respective households that occurred within the 365 day period prior to the surveys. 86% of the population surveyed had heard of rabies. Non-dog owners were 1.6 times more likely to have heard of rabies than dog owners; however, fear of rabies was not a reason for not owning a dog. Government veterinary services were reported most frequently as respondent source of rabies knowledge. Nearly 13% of households had a member bitten by a dog within the year prior to the surveys with 82% of the victims visiting a clinic as a response to the bite. 35% of these clinic visitors received at least one rabies vaccination. Regression modeling determined that the only response variable that significantly reflected the likelihood of a patient receiving rabies vaccination or not was the term for the area surveyed. Overall the survey showed that most respondents have heard of dog associated rabies and seek medical assistance at a clinic in response to a dog bite regardless of offending dog identification. An in-depth study involving factors associated within area clinics may highlight the area dependency for patients receiving rabies post exposure prophylaxis shown by this model.
| Canine rabies has been enzootic to KwaZulu-Natal province, South Africa since the mid-1970's. Vaccination requirements for domestic species and animal control laws enforced in industrialized countries frequently eliminate the need for rabies post exposure prophylaxis (PEP) when an animal bite occurs. Rabies deaths in Africa are frequently linked to poverty and ignorance resulting in a lack of urgency for PEP in an environment where less than 70% of the domestic dog population is vaccinated against the disease. The results presented here are part of a larger canine ecology study conducted in KwaZulu-Natal from September 2009 through January 2011. The six surveyed areas consisted of three land use types: three rural villages, two urban townships and one peri-urban township. The findings show that although a large portion of the population has awareness of rabies, there is a lack of understanding in the response to dog bites. Regression modeling of data suggests that there is an effect of area upon the result of a bite victim receiving PEP as part of treatment. Detailed retrospective study of dog bite incidence and an introspective study of clinics and treatment centers within the province may help explain the results found in this study.
| Rabies kills tens of thousands of people in developing countries each year, and it is estimated that almost half of global rabies incidences occur in Africa [1]–[2]. However, one major factor compounding the problems of rabies is a high probability of disease underreporting. Studies in Tanzania, for example, indicated that there are ten cases for every one officially reported [1]. Once clinical signs of encephalitis become apparent, human rabies is virtually untreatable [3]. Although there has been at least one bona fide case of survival using intensive care treatment, much remains to be understood about factors determining the outcome of such treatment procedures while the required facilities and cost of procedure put such interventions outside the reach of those countries where dog and human rabies is most prevalent [4]. Reliance on proper wound management, and timely post exposure prophylaxis (PEP) [appropriate administration of vaccine and immunoglobulin], is crucial to the prevention of human rabies in exposed persons.
From the above perspective, rabies deaths in Africa are linked to ignorance and poverty. People from rural areas and young children, lacking knowledge of rabies and thus the requirement and urgency of PEP, are most frequently affected [5]. In just one example, from a relatively progressive African state, viz. South Africa, it was shown that half of the laboratory confirmed cases from 2008 did not seek any medical intervention after dog bite exposure [6]. Although rabies PEP is free of charge for bite victims in South Africa, the full post exposure treatment with vaccine and immunoglobulin G costs the South African health system more than USD $152 per individual [7].
KwaZulu-Natal (KZN), one of nine provinces, is located on the east coast of South Africa (Figure 1) with an area of 94,361 km2 and a human population last estimated at 10,819,130 with a growth rate of 1.2% for all races of people [8]. KZN contains just over 21% of the total human population for South Africa despite the province being only 7.7% of the country's land mass. Over 84% of the population is black African, mostly of Zulu cultural origin [8]. It is thought that canine rabies spread to KZN from adjacent Mozambique during the 1960's. Although the disease was then eradicated through the use of mass vaccination campaigns and dog control, rabies was reintroduced in the mid 1970's and has been enzootic to KZN ever since [9]. Historically, most of the human rabies cases in South Africa over past decades have been from KZN. Many dogs in South Africa are not immunized against rabies despite laws mandating vaccination, and KZN is no exception. High dog population turnover, lax enforcement of government regulations and interruptions in vaccination campaigns are all likely factors that contribute to low rabies immunization coverage. In this regard, rabies does not generally appear to enjoy appropriate public health priority in African countries. Poor reporting and poor surveillance, resulting in an apparent lack of political commitment to rabies control, seems to be common practice. In this study it was our objective to better quantify issues such as the above, for the particular region of KZN. Here we analyze the responses to dog bites in an area that has been dog rabies-enzootic for decades, and where control has been attempted for an equal period of time. A better understanding of the societies and practices involved, including knowledge and awareness, would be crucial in improving a disease situation that has been ongoing for the past 40 years. As part of a comprehensive rabies control program, we have queried a representative sample of several population segments of the KZN province about rabies knowledge, interest in enforcement of animal control laws and in community based surveillance.
From September 2009 through January 2011, household surveys were conducted in six different communities across KZN province, covering three land use types: rural, urban and peri-urban (Figure 1). Distribution of the 1992 households completing the surveys was 52% rural, 33% urban and 15% peri-urban. Rabies was enzootic in all areas, with the exception of the peri-urban community of Wembezi. Affluent urban and suburban areas where people keep dogs in confined spaces most likely have lower rabies risk due to fewer affective contacts between animals and easier access to veterinary services and were therefore excluded [10]. Poorer urban townships and rural villages most frequently represent the areas from where canine rabies is reported (KZNDAERD unpublished data). The study areas were selected with the assistance of the KZN Department of Agriculture and Environmental Affairs and Rural Development (KZNDAERD), Veterinary Services division.
Simple random sampling and systematic surveys are difficult in developing countries due to logistical and sometimes adverse cultural reasons [11]. Random sampling using a cluster or ‘area’ design was used because homesteads in rural areas are not numbered and informal housing settlements within townships frequently are arranged haphazardly [12].
Based upon World Health Organization guidelines [13] the questionnaires were composed of two parts; a household survey for collecting demographics and community opinions, and an individual dog survey for descriptive statistics of the owned dog population. Though the primary objective of the surveys was to gain provincial dog demographics, respondents were asked exploratory questions regarding their knowledge of rabies, histories of dog bites and response to those incidents in consideration of possible future studies. Respondents were also queried about their interest in animal-law enforcement, animal ownership regulations and community based surveillance. The surveys were translated into isiZulu and then back translated to English before being piloted in a township with similar human demographics and a history of canine rabies. The survey tool was refined prior to use in this study and the final questionnaires were well received by both the surveyors and the target population with no further improvements or modifications required. KZNDAERD Animal Health Technicians and students, Department of Health workers, Environmental Health workers and SPCA employees were trained to perform the surveys. Surveyors were instructed to introduce themselves to household respondents and explain the purpose of the questionnaire prior to asking their permission to carry out the survey. Surveyors wore name tags which had ‘Rabies Surveillance Team’ printed in large block letters with a bright red border and the Departmental insignia as an identification aid. Prior permission to conduct the surveys had been sought from municipal counselors. All interviews were conducted between the hours of 9 am and 3 pm.
The data from each area was entered into a Microsoft Excel spreadsheet and then imported into SAS version 9.3 (SAS Institute, Inc., Cary, North Carolina, USA). Descriptive statistics were generated, and cross tabulations calculating Pearson's Chi Square (χ2) were performed in tests of association. A logistic regression model was built using SAS to predict outcomes of human dog bite victims receiving rabies PEP [14].
The study was approved by the University of Pretoria, Veterinary Faculty Research Committee at Onderstepoort campus. An application for the non-experimental use of animals was approved by the Animal Use and Care Committee from the University of Pretoria. Interviewed subjects were provided informed consent orally in their native language as was stated in the research proposal approved by the Research Committee. The purpose of the survey interview was explained to each participant by the interviewer, who could either accept or decline to participate in the survey. If the respondent declined to be interviewed it was marked at the top of the survey form, whereas those who agreed to be interviewed had a third party witness to this verbal agreement, and the interview was continued.
A total of 1992 households consisting of 13,756 people (range 1–34, median = 6) completed the surveys within the three targeted community types. Surveys were answered by a person defined as head of the household in 68% (1361/1992) of the cases across the province (range 63–76%). The sex of the respondent was not recorded. Of the remaining cases, 11 comprised interviews with children under the age of fourteen years of age in the presence of an older relative who consented to the child answering questions. Another 183 children over the age of fourteen years were interviewed at homes where adults were not present. In 435 surveys, an adult other than the head of the household responded to interview questions. In 2 cases the category of the respondent was missing.
Eighty-six percent of the population (1716/1992) surveyed across the province had heard of the disease called rabies (Figure 2). No attempt was made to evaluate the individual's depth of rabies knowledge. Some respondents stated that they did not truly know the source of rabies or how to prevent it. However, it was clear that some respondents knew that vaccination of dogs was important to the safety of people in the community based upon their remarks.
There was no significant relationship found between area surveyed and respondent knowledge of rabies (χ2 = 10.864, df = 5, p = 0.541). When surveyed areas were grouped by land use, 81% of the peri-urban society had some knowledge of rabies, whereas 87% of rural and 88% of urban citizens had knowledge of rabies. Surprisingly, non-dog owners were 1.6 times more likely to have heard of rabies compared with dog owners. No respondents stated that fear of rabies infection was a reason for not owning a dog.
Persons who responded that they did have some knowledge of rabies were further queried as to the source of their knowledge. Government Veterinary Services Animal Health Technicians have the role of visiting schools and educating children about rabies. Among other resources, they utilize a government prepared video entitled “If I Only Knew” and various informative pamphlets discussing rabies. We found that schools and school children accounted for 19% of the population's knowledge source across the province. However, there was not a significant relationship between the presence of school children and knowledge of rabies in individual households (χ2 = 0.027, df = 1, p = 0.868).
Less than two percent of the surveyed population indicated that they have acquired rabies knowledge from the local health clinic. Since one objective of our research was to understand where citizens were able to receive valuable information about rabies, any viable source was noted and utilized as an element of feedback for the development of future programs employed by the government departments responsible for human and animal health. Public print and broadcast media were cited highest among non-dog owners as their source of rabies knowledge, whereas government sponsored campaigns were most frequently cited from dog owners (Figure 3).
12.7% (95% CI 11.3–14.2) of the 1992 households in the areas surveyed reported that at least one member of the household had been bitten by a dog in the past year. Age of bite victim was not recorded. The lowest incident rate was in Esikhawini (urban) and the highest incidence occurred in Wembezi (peri-urban). Across KZN, significantly more people who did not own dogs had been bitten by a dog than those who did own a dog (χ2 = 9.477, df = 1, p = 0.002). Among victims who were dog owners the number of dogs owned did not make a difference in dog bite incidence. Although 33% (667/1992) of households reported feeding of dogs that they did not own (on their property), there was not a significant relationship between being bitten by a dog and feeding other dogs on the property (χ2 = 3.424, df = 1, p = 0.064).
As a follow up question, households with recent dog bite victims were asked if they knew the aberrant dog involved in the incident (Figure 4). Identification of the offending dog was missing in 10% (26/253) of recorded cases. In 71% of records where the dog was identified, the neighbor's dog had bitten the victim. Only 12% of victims had been bitten by their own dog and 17% of victims had been bitten by a dog with which they were unfamiliar. Unknown dogs were referred to as strange dogs rather than stray dogs, as the animal could be owned but unrestricted.
Historically there have been concerns about people in rural areas visiting traditional healers rather than attending a clinic after a dog bite event. Sudarshan et al [15] showed that 60% of dog bite victims in India who succumbed to rabies had sought some kind of indigenous treatment following the incident, receiving either magico-religious practices or some kind of herbal therapy. After a recent emerging rabies epidemic in the Limpopo province of South Africa (2005–2006), it was established that 20% of the fatal human rabies case patients saw a traditional healer prior to attending hospital [16]. In this survey in KZN, less than 2% (4/253) of dog bite victims, all of whom were from rural areas, reported resultant visits to a traditional healer.
With regard to the washing of bite wounds as a first response to incident, only 8% of victims mentioned washing the wound. We found that 56% (1115/1992) of domiciles visited in our surveys had either a pit latrine or no toilet facilities, indicative of a lack of running water at the household level. In some rural areas a public tap was available some distance from the house. In other rural areas, people made use of rivers or streams for daily water.
In the six areas surveyed, over 80% (207/253) of victims visited a clinic as a response to dog bite incident except in the rural area of St. Chad's (Figure 5). This area alleged the highest rate of rabies knowledge (88%) (Figure 2), but had the least number of visits to the clinic (54%) as a response to bite incidence. St. Chad's has both a community health center and close access to neighboring community clinics and hospitals. Detailed questions that would uncover the decisions made by bite victims, actions taken and reasons for those actions were not asked.
Those households with victims who had visited a clinic in response to a dog bite were asked what injections they had received (Table 1, Figure 6). Twenty-four percent (50/207) of clinic visitors reported receiving no injections. Thirty-four percent (70/207) of respondents did not know the extent of treatment received, as it was either not explained to them, they could not remember or they did not attend the clinic with the victim. Thirty-five percent (73/207) of persons visiting the clinic received rabies vaccine, 5% received tetanus only and 1.4% received both rabies and a tetanus vaccine. Those victims who said they received rabies vaccine were not asked if they returned to the clinic to complete the World Health Organization recommended (Essen schedule in the case of South Africa) four injection series or if they received immunoglobulin in the case of category III bites [17]. Severity and location of bite wounds was not queried of respondents.
An effort was made to determine if there was an association between those clinics where patients had received rabies vaccine and the area surveyed, whether the victim owned dogs, respondent knowledge of rabies and identification of offending dog. Fifty-two responses (20%) were deleted from the model due to missing values for either the response or explanatory variables. In the final logistic regression model only the area surveyed significantly contributed to human rabies vaccination outcome (Table 2).
The model appeared to fit the data (Somer's D = 0.395). The urban township of Esikhawini was the area with the most dog bite victims receiving rabies vaccine, while victims in urban Umlazi Township received the least (Table 3).
Respondents were queried if they were interested in learning more about illnesses that could be shared between people and animals. Ninety-four percent of the population surveyed across the province (1865/1992) said they would be interested in gaining information about zoonotic disease potential in their community. The urban area of Umlazi, where only 88% of the respondents answered agreeably, stood out as the only area where there are a significant number of respondents who were disinterested in zoonoses (χ2 = 30.581, df = 10, p = 0.001).
Respondents were asked if they would, as witnesses, be interested in reporting ill dogs, strange behavior or dog bite incidents occurring in their communities. The majority of respondents across all communities were interested in reporting such sightings; however, the peri-urban and urban communities had significantly less interest than the rural areas in community based reporting (χ2 = 22.120, df = 5, p = 0.000) (Figure 7).
Respondents answering in the affirmative were further queried as to whom in the community they would want to report these incidents. Nearly 40% identified government veterinary services when considering reporting sick dogs and possible rabies cases (Figure 8). Other parties mentioned were the local clinic, a teacher, the dog's owner, or SPCA. However, community members regularly stated that despite their desire to report, they had no contact information for either veterinary services or the SPCA.
Other than dogs, 2 potential rabies maintenance hosts present in KZN are mongooses and jackals. Bat eared foxes, which maintain rabies in the western provinces of South Africa [18], are rarely seen in KZN. Questioning respondents about sightings in their community served as a screening tool for the possibility of further studies concerning wildlife and the spread of rabies in KZN. Overall, less than 22% of the respondents across the province encountered either of these wildlife species in their communities (Figure 9). The rural tribal authority area outside of Pongola has dense flora and is located on the Swaziland border which could explain why there were so many more jackal sightings in this rural area versus any other.
Despite South Africa possessing laws requiring vaccination and licensure of dogs, these regulations are rarely enforced. Respondents were asked if they desire law enforcement regarding removal of stray or unsupervised dogs (Figure 10). There was a significant difference in area type and desire for animal control law enforcement, with the least concern reported from the rural areas (p = 0.0001).
Surveys respondents were asked if they desired laws that would limit the number of dogs that one household could own (Figure 11). The number of dogs owned by the dog owning households surveyed was similar between the rural and peri-urban households with an average of 2.47 dogs per dog owning household (range 2.16–2.64). Urban dog owning households had fewer dogs with the average being 1.66 (range 1.64–1.68). Some households in rural areas were recorded as owning up to 19 dogs. There was a significant difference between rural areas and the urban/peri-urban areas in desire for limitations on the number of dogs one household could own (p = 0.0001).
The results from this survey indicate that 86% of persons in high risk canine rabies areas of KwaZulu-Natal have at least heard of the disease even if they are unaware of the details surrounding transmission and consequences of exposure. The long history of enzootic canine rabies in the province and the continuous efforts put forth by the KZNDAERD- Government Veterinary Services appear to contribute the most (33%) to this public awareness. Although 100% would be ideal, an 86% knowledge rate is better than was reported from other studies. In dog rabies enzootic Zimbabwe, for example, 74% of pet owners interviewed in Harare animal clinics were aware that rabies was transmitted to people by dogs [19]. Zimbabwean respondents also reported gaining information about other zoonotic diseases from their veterinarian. Respondents in KZN do not have local veterinary clinics as a resource from which to gain this knowledge. In the current study, the peri-urban working community of Wembezi had the lowest rate of rabies knowledge at 81%. This is interesting in that it is also the only community identified as being free of canine rabies for greater than 10 years per surveillance records from KZNDAERD. Thirty-nine percent of households in Wembezi owns at least one dog and has an estimated dog population of 2, 916 (Hergert unpublished data). Calculated dog density figures for Wembezi are similar to the urban areas surveyed in this study, where a slightly higher level of rabies knowledge was recorded. Non-dog owners were 1.6 times more likely to have heard of rabies versus dog owners. People who do not own dogs are gaining information about rabies from media sources, which had a value of almost 30% from respondents in all areas. This is similar to findings in the developed world – e.g. in Texas, USA 43% of non-pet owners reported learning about zoonotic diseases from the media or newspaper [20]. Media sources for rabies information may actually be viewed by both the dog owning and non-dog owning public; however, dog owners may be more likely to respond that their source of rabies knowledge was from government vaccination campaigns, as they would be attending these events. Therefore, a certain amount of bias may weigh towards government campaigns as a source for dog owners due to their familiarity. Understanding why more non-dog owners report knowing about rabies versus dog owners is unclear from this survey and would require further study. South African Government Veterinary Services has the task of informing people about rabies through vaccination campaigns and schools. When these two reported sources of knowledge are combined it is evident that Veterinary Services is responsible for 52% of the information gained by both the dog and non-dog owning public. However, there was not a significant relationship between household knowledge of rabies and knowledge source originating from schools. Veterinary Services of KZN might take into consideration when planning educational campaigns in their communities, that schools are not heavily targeted. Eighty-two percent of interviewed households contained school aged children; therefore, schools appear to be a viable outlet for the dissemination of rabies information. Human health clinics were reported as a knowledge source in less than 2% of responses. This result may support the findings from Francophone countries of Africa where medical authorities and health practitioners are reported to be under educated in the perils of rabies [5]. In Texas, USA, the family doctor was reported as the source of zoonotic disease information in only 6% of both pet and non-pet owning households [20]. Doctors were also indicated well below veterinarians and the media as a source of disease information in Zimbabwe [19]. Detailed examination into what transpires in KwaZulu-Natal clinics for dog bite case patients should be explored in the face of a One Health environment.
Twelve percent of the households in the areas surveyed had someone bitten by a dog in the last year. Other animal bite victims in African countries have been identified through retrospective studies starting at the clinic or hospital level using a trace back system in order to locate and interview the victims in depth [21]–[22]. This type of retrospective study should be conducted in KwaZulu-Natal in order to gain further descriptive information of the dog bite incidence.
The neighbor's dog was identified as the offending canine in 71% of bite cases in this survey. Only 12% of the people had been bitten by their own dog. However, 17% of victims had been bitten by a dog with which they were unfamiliar. These dogs were identified as strange rather than strays or feral dogs, as they could not be differentiated from owned free roaming dogs. Eighty-three percent of dogs in this study were identified as being fully or partially unrestricted, being allowed to wander at will. Over 96% of the human rabies cases in India from 1992 to 2002 resulted from a dog bite, with 75.2% resulting from stray dogs and 11.1% from pets [15].
In response to the dog bites more than 80% of people went to the clinic in all areas except for rural St. Chad's where only 54% of bite victims reported clinic visits. St. Chad's residents reported a high awareness of rabies (88%), which could be explained by the rabies epidemic experienced in the area a few months prior to the survey. This area has a community clinic, as well as other nearby clinics and hospitals reachable by taxi. Without in depth queries, the reason why more persons did not visit a clinic remains unknown. Despite concerns about delayed treatment after dog bites, less than 2% of victims in this study visited a traditional healer and all of those cases were from rural areas. Herbal therapy and magico-religious practices were sought by rabies bite victims in India in 60% of fatal cases [15]. Respondents in KZN may be more informed about rabies than persons in other developing countries. One survey respondent said that the reason he did not go to the clinic after being bitten by his neighbor's dog was because the neighbor could prove to him that his dog had been previously vaccinated against rabies. Therefore, the victim felt he was safe to treat the wound at home. The investigation, follow through and cognition shown by this respondent is not something that should be expected from most bite victims. Dog bite victims in this study tended to visit the clinic regardless of their familiarity with the dog that bit them.
Of those victims that did attend a clinic 22–75% received at least one rabies vaccine. The lower end of this spectrum is similar to what was seen in India where only 21% of rabies victims had received at least one rabies vaccine [15]. The area with the lowest rabies vaccination treatments was urban Umlazi Township and the highest was urban Esikhawini Township. In the regression model predicting what factors had an important impact on the victim receiving a rabies vaccine, only the area surveyed was found to be significant (p = 0.0001). Health facilities in South Africa where rabies vaccine and immunoglobulin (RIG) are available are listed with telephone contact numbers in the national rabies guideline [23]. However, a nationwide telephonic survey, which included 50% of the facilities identified for KwaZulu-Natal, was conducted in order to confirm the availability of these products. Only 68% of all the sites surveyed across the country were contactable by telephone. Forty-one percent had both vaccine and RIG, 32% had only vaccine, 5% had only RIG and 21% had neither vaccine nor RIG available [24]. Considering the results of this telephonic survey it is quite conceivable that administration of rabies vaccine is area dependent across the country. The juxtaposition of Esikhawini Township to the Port of Richard's Bay could explain why this area, which had the lowest recorded number of dog bite cases, had the highest amount of rabies vaccine administered. The Richard's Bay area may have more rabies vaccine dispensed that are related to aspects about the constituents the medical community serves, or because the medical staff could be indiscriminately dispensing supplies regardless of exposure risk.
In Francophone African countries accurate rabies data is scarce [5]. This may be true in other African countries as has been previously eluded from Tanzania [1]. This survey showed a large respondent willingness to participate in community based surveillance at the village level. Community based surveillance activities should be considered in countries which lack central political will or local municipal finances. However, it has been stated that passive systems in developing countries are ineffective; therefore, an economic community based active surveillance system is recommended [25]. Unfortunately, community based systems have been shown to fail, particularly when there is a discrepancy in the interpretation of needs between the community and the donor organization [26]. Therefore, the methodology to be employed would have to be developed from a grassroots level rather than at a higher administrative level, which would take a commitment not previously demonstrated from this rank of society.
Persons living in communities at high risk for canine rabies are interested in animal control laws and regulations. However, there is an indication of concern in the rural areas that these laws would also limit the number of livestock owned. This result may be due to rural areas owning more livestock. An indirect association between the limitations on number of dogs allowed with restrictions on livestock ownership may be behind these results. In rural Texas, USA, a survey regarding cattle ownership conducted from Texas A&M University indicated that ranchers were reluctant to comply with trace back ear-tagging measures, as the procedure would identify to officials how many cattle were owned by each producer at any point in time (Dominguez unpublished data). Responsiveness and dedication to upholding animal control laws in this cultural environment by obliged parties will have to be instilled in a generation of officers committed to uplifting the community.
Crucially, since this simple intervention may be particularly effective in preventing infection, only 21 of the 253 people in our survey bitten by a dog washed their bite wound as a response to treatment. It has been established that washing the bite wound for 15 minutes with soap and water can help reduce the incidence of disease by eliminating or inactivating the virus [5]. As only 44% of households were reported as having indoor plumbing (indicated by flush toilets) lack of available fresh water may explain the low percentage of wound washing as a response to post-bite treatment.
This study shows that greater than 86% of the population has at least heard of the disease called rabies, but the response to dog bites indicates that both the general public and health sectors of the population do not understand the possible consequences related to dog bites in rabies enzootic environments. Availability of vaccine is an important factor in determining if bite victims receive rabies vaccine during clinic visits in KwaZulu-Natal and other parts of Africa; however, factors within the clinic setting such as staff knowledge need to be considered as well. Consideration of the offending dog in the bite incident has not been shown to play a role in victim response to dog bites. Therefore, the wasting of PEP could be as a big a problem as people at risk not receiving necessary vaccine. Our results also indicate that schools and rabies education of schoolchildren can be much improved. Not only are children most at risk of rabies exposure, but schools may present appropriate structures for dissemination of this kind of information and should be utilized to a greater extent. Questions in this survey regarding response to dog bites could have been more detailed. An example would be to include the age of the bite victim as a variable. Regardless, these results lend credence to the statement that an in-depth study regarding the treatment people are receiving and the public knowledge of rabies needs to be conducted.
|
10.1371/journal.ppat.1000061 | Suppression of Plant Resistance Gene-Based Immunity by a Fungal Effector | The innate immune system of plants consists of two layers. The first layer, called basal resistance, governs recognition of conserved microbial molecules and fends off most attempted invasions. The second layer is based on Resistance (R) genes that mediate recognition of effectors, proteins secreted by pathogens to suppress or evade basal resistance. Here, we show that a plant-pathogenic fungus secretes an effector that can both trigger and suppress R gene-based immunity. This effector, Avr1, is secreted by the xylem-invading fungus Fusarium oxysporum f.sp. lycopersici (Fol) and triggers disease resistance when the host plant, tomato, carries a matching R gene (I or I-1). At the same time, Avr1 suppresses the protective effect of two other R genes, I-2 and I-3. Based on these observations, we tentatively reconstruct the evolutionary arms race that has taken place between tomato R genes and effectors of Fol. This molecular analysis has revealed a hitherto unpredicted strategy for durable disease control based on resistance gene combinations.
| In agriculture, the most environmentally friendly way to combat plant diseases is to make use of the innate immune system of plants, for instance by crossing into crop varieties polymorphic resistance genes that occur in natural populations of the crop plant or its close relatives. Plant pathogens, however, have co-evolved with their host plants and have developed ways to overcome the immune system. To effectively make use of components of the plant immune system, it is therefore important to understand the co-evolution of plants and their pathogens at the molecular level. For the interaction between a fungal pathogen and tomato, this paper presents a breakthrough in this respect. A small protein secreted by some strains of the fungus Fusarium oxysporum was found to suppress the activity of two disease resistance genes of tomato. However, a third resistance gene specifically targets this suppressor protein and renders the plant fully resistant against fungal strains that produce it. With this insight, together with knowledge of the genetic variation in the pathogen population, a combination of resistance genes is suggested that is expected to confer durable resistance in tomato against Fusarium wilt disease.
| Long periods of co-evolution of plants and microorganisms have led to complex mechanisms of attack and defence, involving the innate immune system of plants and virulence factors of pathogens [1]. The first layer of plant defence, called basal immunity, is based on recognition of conserved microbial molecules but can be suppressed by microbial virulence factors known as “effectors”. Plants respond to this suppression by employing a second layer of defence, Resistance (R) gene-based immunity, which relies on recognition of effectors [2]. In turn, at least bacterial pathogens have found ways to manipulate or evade this second layer of defence [3]. It is unclear to what extent this capacity exists in eukaryotic plant pathogens like oomycetes and fungi.
Like bacteria, many plant-pathogenic fungi secrete proteins that are recognized by R-genes [4],[5]. One of these fungi is Fusarium oxysporum, a common soil inhabitant. It propagates asexually and is mostly harmless. However, pathogenic and host-specific clonal lines have evolved that cause severe diseases in crops, such as banana, cotton, cucumber, melon and tomato [6],[7]. Many of these diseases are caused by colonisation of the water-conducting xylem system of the roots followed by upward growth through xylem vessels, with wilting and death as a dramatic result. Strains of F. oxysporum that cause wilt of tomato plants are grouped in forma specialis (f.sp.) lycopersici. Several polymorphic resistance (R) genes have been identified in the tomato gene pool that each confer resistance against a subset of F. oxysporum f.sp. lycopersici (Fol) strains. These are I (for Immunity), I-1, I-2 and I-3 [8]. Races of Fol are named historically according to the R gene that is effective against them: the I gene and the (unlinked) I-1 gene are effective against race 1, race 2 overcomes I and I-1, but is stopped by I-2, while race 3 overcomes I, I-1 and I-2 but is blocked by I-3 [9]. Race 1 strains have been further divided into subgroups based on whether or not they are able to (partially) overcome I-2 or I-3 [9],[10].
Based on the gene-for-gene hypothesis [11], it is assumed that disease resistance conferred by R genes in tomato requires ‘matching’ avirulence (AVR) genes in Fol. The I gene originates from Solanum [Lycopersicon] pimpinellifolium and resides on chromosome 11 [12],[13], while the I-1 gene is located on chromosome 7 in another wild relative of tomato, Solanum [Lycopersicon] pennellii [14]. The I-2 gene has been cloned and encodes an R protein of the common NB-LRR class [15]. The I-3 gene has not yet been cloned [16], but the matching AVR gene has: it encodes a small protein, Six1 (“Secreted in xylem 1”), which is secreted by Fol during colonization of the xylem system [17] and contributes to fungal virulence [9]. Six1 is now called Avr3 to indicate its gene-for-gene relationship with the I-3 resistance gene.
We describe here the identification and analysis of a second avirulence factor of Fol, Avr1. Surprisingly, this protein does not only act as an avirulence factor in conjunction with the I gene, but also suppresses disease resistance mediated by I-2 and I-3.
In an initial analysis of the xylem sap proteome of tomato plants infected with Fol race 1 using 2-D gel electrophoresis and mass spectrometry, three small secreted proteins of Fol were identified in addition to Avr3 (Six1), named Six2, Six3 and Six4, and their genes cloned [18]. We now find that one of these, Six4, is not secreted by Fol race 2 (Fig. 1). For reasons detailed below, we now call this protein Avr1. Like the AVR3 (SIX1) gene, AVR1 is surrounded by repetitive elements (Fig. 2A). In all of the race 1 strains we examined, PCR experiments detected the presence of AVR1 and no sequence polymorphism was detected in the coding regions of seven isolates from different clonal lines (see [9] for the list of strains; 17 of these are race 1, 23 are race 2 or 3). AVR1 was not detected in race 2 or 3 strains by PCR nor is AVR1 present in the genome sequence of the race 2 strain 4287 (Fusarium oxysporum Sequencing Project; Broad Institute of Harvard and MIT (http://www.broad.mit.edu)). Absence of AVR1 or closely related genes in the race 2 and race 3 strains used in this study was confirmed by DNA gel blot analysis (Fig. 2B, lanes 4 and 7, respectively).
To test whether AVR1 is indeed responsible for avirulence of Fol on plants carrying the I gene, we created an AVR1 gene knock-out in a race 1 strain (Fol004) through Agrobacterium-mediated transformation (Fig. 2). For the AVR1 gene, the frequency of homologous recombination leading to gene knock-out turned out to be extremely low, with only a single knock-out mutant obtained out of ∼200 transformants (Fig. 2B, lane 2). A disease assay with this mutant (avr1Δ) confirmed that indeed deletion of AVR1 leads to breaking of I-mediated disease resistance (Fig. 3A, panel A, quantified in Fig. 3B). Re-introduction of AVR1 in the avr1Δ strain (Fig. 2B, lane 3) restored the original avirulence phenotype (results not shown). In addition, we found that disease resistance conferred by the unlinked I-1 gene in tomato also depends on recognition of Avr1, since the avr1Δ strain (but not its parental strain) is virulent on a plant line carrying I-1 (line 90E402F, results not shown). This suggests that I and I-1 express the same resistance specificity.
To confirm that the AVR1 gene is sufficient to trigger recognition by the I gene, we transformed AVR1 to a race 2 strain (Fol007) and a race 3 strain (Fol029) that do not contain AVR1 (Fig. 2B, lanes 4–9) and are virulent on I-containing tomato lines. Ten independent transformants (six of race 2 and four of race 3) containing AVR1 were unable to cause disease on I-containing plants (Fig. 3A, panels B and C, quantified in Fig. 3B), confirming the avirulence character of AVR1. In contrast to Avr3 [9], Avr1 is dispensable for full virulence towards plants that do not contain R genes against Fol (results not shown).
Although all Fol strains possess an intact AVR3 gene, most race 1 strains nevertheless cause disease on plants carrying only the I-3 gene [9]. One explanation for this is that Avr1 itself is involved in suppression of I-3 mediated disease resistance. To test this, we inoculated a plant line containing only the I-3 gene with the set of Fol strains described above. The results clearly show that Avr1 indeed has this suppressive activity: deletion of AVR1 in race 1 leads to loss of virulence towards I-3 plants (Fig. 3A, panel D, quantified in Fig. 3B), while introduction of AVR1 in race 2 or race 3 leads to gain of virulence towards I-3 plants (Fig. 3A, panels E and F, quantified in Fig. 3B). Furthermore, we discovered that Avr1 also suppresses I-2-mediated disease resistance (Fig. 3A, panels D and E, quantified in Fig. 3B). This means that the ability of some race 1 strains to cause disease on I-2 plants, as observed earlier [10], is likely to be caused by suppression of I-2 rather than loss of AVR2. In accordance with earlier observations using I-3 plants [9], we found that virulence due to suppression of I-2 and I-3 is partial compared to strains lacking the corresponding AVR gene (Fig. S1). It should be noted that not all race 1 strains are virulent on I-2 and/or I-3 plants [9],[10], even though all contain AVR1 with identical sequences (results not shown). Apparently, suppression of R gene-based immunity by Avr1 is dependent on unknown factors in the genetic background of the fungus. Since suppression works in Fol007 (race 2) and Fol029 (race 3), the genetic background in which AVR1 is effective is not restricted to race 1 strains.
Our observation that Avr1 is not required for virulence to plants without I genes may be due to the existence of other effectors that are redundant for such an activity. Alternatively, the role of Avr1 is restricted to the suppression of I-2 and I-3-mediated disease resistance. A mechanistic explanation for the latter role could be that Avr1 interferes directly with Avr2 and Avr3. However, at least Avr3 accumulates in xylem sap and remains unaltered in the presence of Avr1 [9],[18]. A direct interaction between the two proteins could also not be demonstrated in vitro by pull down experiments (results not shown). Unlike bacteria, pathogenic fungi are not known to inject proteins directly into plant cells, but many are known to secrete small, frequently cysteine-rich, but otherwise unrelated proteins during colonization of plants [5]. Avr1, like Avr3, falls within this group, the predicted mature protein having 184 residues including 6 cysteines and lacking homology to other proteins [18]. The mode of action of most of these small secreted proteins has remained unclear. Molecular targets have been described for Avr2 and Avr4 from the leaf mold Cladosporium fulvum: Avr2 is a protease inhibitor [19] while Avr4 binds chitin in the fungal cell wall and protects it against attack by plant chitinases [20]. These two proteins act in the apoplast to enhance fungal virulence, but others act inside plant cells [4]. Uptake from the apoplast by plant cells has been shown directly for ToxA, a small secreted protein that acts as a host-selective toxin [21]. This may also occur with Avr2, since I-2 is a cytoplasmic protein [15]. Avr1, then, may interfere with the uptake of Avr2 and Avr3. Alternatively, it may be taken up itself and interfere with I-2 and I-3 or with signal transduction processes downstream of these R proteins (Fig. 4).
Suppression of effector-triggered (R gene-mediated) immunity has been observed in bacteria [3],[22],[23]. In plant pathogenic fungi, suppression of avirulence by unlinked loci has been demonstrated by genetics in rust fungi [24]. In the flax rust fungus, two dominant alleles or tightly linked genes at the I (“inhibitor”) locus suppress – sometimes partially – either one (M1) or several (M1, L1,7,8,10) R genes out of 30 against flax rust [24],[25]. The flax rust inhibitor locus is not itself linked to avirulence. Here, we report the identification of a fungal avirulence factor that suppresses disease resistance conferred by two R genes.
Interpreting this phenomenon in terms of molecular arms races between plants and their pathogens [1], we envisage the following scenario. During evolution of the tomato-Fol pathosystem, I-2 and I-3 have evolved to recognize, respectively, Avr2 and Avr3. Since Avr3 is required for full virulence of Fol, evasion of I-3 recognition through loss of the AVR3 gene would entail a serious fitness penalty. This explains why all Fol strains analysed so far retained AVR3 [9],[26]. Point mutations in AVR3 preventing recognition have not been found either [9]. A possible explanation for this is that the I-3 protein operates in accordance with the guard model, in which not the Avr3 protein itself but the effect it has on its virulence target is recognized [27]. In any case, Fol has (partially) regained virulence towards I-3-containing plants by acquisition of AVR1, which, as shown here, suppresses the function of I-3. Subsequently, tomato responded to this ‘invention’ with the employment of the I gene, or the unlinked I-1 gene, to specifically recognize and respond to Avr1. Apparently, I and I-1 are themselves insensitive to the suppressive effect of Avr1 (Fig. 4).
The agricultural ‘arms race’ between Fol and tomato is different from the natural one because it is dictated by successive R gene deployment in commercial cultivars [8]. The I gene from the wild tomato relative Solanum [Lycopersicon] pimpinellifolium was the first R gene to be introgressed into tomato cultivars to resist Fusarium wilt in the 1940s [12]. At that time, Fol strains without Avr1 may already have been present in some locations, since I-breaking race 2 strains were quickly discovered [28] even though major outbreaks did not occur before 1960 [29]. The I-2 gene, also from S. pimpinellifolium and directed against Avr2, was introduced in commercial cultivars in the 1960s to protect tomato against Fol race 2 [29],[30]. The combination of I and I-2 was effective for about two decades until the appearance of race 3 in both Australia and North America [31], which probably emerged from a race 2 background through selection for loss or mutation of AVR2. To combat race 3, the I-3 gene was introgressed from S. pennellii [31]. From the results presented here, we deduce that the combination of I (or I-1) and I-3 may yield durable resistance of tomato to Fusarium wilt disease of tomato, since I-3 is directed against a virulence factor (Avr3) and I (and I-1) against the suppressor of I-3 (Avr1).
The molecular toolbox that is now gradually filling up (Avr1, Avr3, I-2) will help us to define host targets and evolutionary bottlenecks that govern the arms race in the Fol-tomato pathosystem. It also may allow development of new strategies for breeding plants with durable resistance against fungal pathogens.
The following tomato lines were used (Fol resistance genes between brackets): GCR161 (I) [32], 90E402F (I-1) [31],[33]; 90E341F (I-2) [29] and E779 (I-3) [31], C32 (no I gene) [32]. The following Fol strains were used: Fol004 (race 1), Fol002 (race 2), Fol007 (race 2), Fol029 (race 3), Fol004avr1Δ (Fol004 with AVR1 deleted by gene replacement), Fol004avr1Δ+AVR1 (Fol004avr1Δ transformed with AVR1), Fol007+AVR1 (Fol007 transformed with AVR1), Fol029+AVR1 (Fol029 transformed with AVR1). See Rep et al. (2005) [9] for a more detailed description of the wild type Fol strains.
Proteins present in xylem sap of tomato plants infected with Fol were isolated and separated with 2-dimensional gel electrophoresis as described earlier [18], using for the first dimension an Immobiline DryStrip of 13 cm, pH 6–11 NL (Amersham Biosciences).
Ten day old seedlings of tomato were inoculated with a fungal spore suspension and disease was scored after three weeks as described earlier [17]. The outcome of the disease assays was quantified in two ways: 1) average plant weight above the cotyledons and 2) phenotype scoring according to a disease index ranging from zero (no disease) to four (heavily diseased or dead) [17].
The AVR1 disruption construct was made by PCR amplification of AVR1 upstream and downstream sequences for homologous recombination, and their insertion in front of and behind the hygromycin resistance gene in the vector pRW2h (see below): an upstream fragment, from 714 bp to 1 bp upstream of the start codon, was cloned into pRW2h between the PacI and KpnI sites, and a downstream fragment, from 375 bp after the start codon to 537 bp downstream of the stop codon, was cloned into pRW2h between the XbaI and BssHII sites (see Fig. 2A for location of the primers). The construct for complementation was made by amplification of a AVR1 expression cassette from 714 bp upstream of the start codon to 537 bp downstream of the stop codon (Fig. 2A), which was inserted between the XbaI and StuI sites of pRW1p (see below). Transformation of these constructs to Fol was done with Agrobacterium as described earlier [34].
pRW2h is a binary vector for Agrobacterium-mediated transformation of fungi. It was made through insertion of a NheI-XbaI fragment from pAN7.1, carrying the hygromycin resistance gene hph under control of the Aspergillus (Emericella) nidulans gpd promoter and trpC terminator [35], into the unique XbaI site of pPZP-201BK [36]. Similarly, pRW1p was derived from pPZP-201BK through insertion of a NheI-XbaI fragment from pAN8.1 [35] carrying the phleomycin resistance gene ble under control of the same gpd promoter and trpC terminator.
Genomic DNA of F. oxysporum was isolated according to Raeder and Broda [37], digested with HindIII and BamHI, separated in a 1% agarose gel and blotted to Hybond N+ according to Sambrook et al. [38]. The probe containing the AVR1 ORF and 3′ sequences (1402 bp, Fig. 2A) was generated by PCR and contains sequences from 72 bp upstream to 537 bp downstream of the ORF. The probe was radioactively labelled with α32P dATP using the DecaLabel™ DNA labeling kit from MBI Fermentas (Vilnius, Lithuania). Hybridization was done overnight at 65°C in 0.5M phosphate buffer pH 7.2 containing 7% SDS and 1 mM EDTA. Blots were washed at 65°C with 0.2 X SSC, 0.1% SDS. The position of sequences hybridizing to the probe were visualized by phosphoimaging (Molecular Dynamics).
The AVR1 (SIX4) locus: AM234064
The Avr1 (Six4) protein: CAJ84000 |
10.1371/journal.pgen.1004236 | CYP6 P450 Enzymes and ACE-1 Duplication Produce Extreme and Multiple Insecticide Resistance in the Malaria Mosquito Anopheles gambiae | Malaria control relies heavily on pyrethroid insecticides, to which susceptibility is declining in Anopheles mosquitoes. To combat pyrethroid resistance, application of alternative insecticides is advocated for indoor residual spraying (IRS), and carbamates are increasingly important. Emergence of a very strong carbamate resistance phenotype in Anopheles gambiae from Tiassalé, Côte d'Ivoire, West Africa, is therefore a potentially major operational challenge, particularly because these malaria vectors now exhibit resistance to multiple insecticide classes. We investigated the genetic basis of resistance to the most commonly-applied carbamate, bendiocarb, in An. gambiae from Tiassalé. Geographically-replicated whole genome microarray experiments identified elevated P450 enzyme expression as associated with bendiocarb resistance, most notably genes from the CYP6 subfamily. P450s were further implicated in resistance phenotypes by induction of significantly elevated mortality to bendiocarb by the synergist piperonyl butoxide (PBO), which also enhanced the action of pyrethroids and an organophosphate. CYP6P3 and especially CYP6M2 produced bendiocarb resistance via transgenic expression in Drosophila in addition to pyrethroid resistance for both genes, and DDT resistance for CYP6M2 expression. CYP6M2 can thus cause resistance to three distinct classes of insecticide although the biochemical mechanism for carbamates is unclear because, in contrast to CYP6P3, recombinant CYP6M2 did not metabolise bendiocarb in vitro. Strongly bendiocarb resistant mosquitoes also displayed elevated expression of the acetylcholinesterase ACE-1 gene, arising at least in part from gene duplication, which confers a survival advantage to carriers of additional copies of resistant ACE-1 G119S alleles. Our results are alarming for vector-based malaria control. Extreme carbamate resistance in Tiassalé An. gambiae results from coupling of over-expressed target site allelic variants with heightened CYP6 P450 expression, which also provides resistance across contrasting insecticides. Mosquito populations displaying such a diverse basis of extreme and cross-resistance are likely to be unresponsive to standard insecticide resistance management practices.
| Malaria control depends heavily on only four classes of insecticide to which Anopheles mosquitoes are increasingly resistant. It is important to manage insecticide application carefully to minimise increases in resistance, for example by using different compounds in combination or rotation. Recently, mosquitoes resistant to all available insecticides have been found in Tiassalé, West Africa, which could be problematic for resistance management, particularly if common genetic mechanisms are responsible (‘cross-resistance’). Tiassalé mosquitoes also exhibit extreme levels of resistance to the two most important classes, pyrethroids and carbamates. We investigated the genetic basis of extreme carbamate resistance and cross-resistance in Tiassalé, and the applicability of results in an additional population from Togo. We find that specific P450 enzymes are involved in both extreme and cross-resistance, including one, CYP6M2, which can cause resistance to three insecticide classes. However, amplification of a mutated version of the gene which codes for acetycholinesterase, the target site of both the carbamate and organophosphate insecticides, also plays an important role. Mechanisms involved in both extreme resistance and cross resistance are likely to be very resilient to insecticide management practices, and represent an alarming scenario for mosquito-targeted malaria control.
| Malaria mortality has decreased substantially in sub-Saharan Africa over the last decade, attributed in part to a massive scale-up in insecticide-based vector control interventions [1]. As the only insecticide class approved for treatment of bednets (ITNs) and the most widely used for indoor residual spraying (IRS), pyrethroids are by far the most important class of insecticides for control of malaria vectors [2]. Unfortunately pyrethroid resistance is now widespread and increasing in the most important malaria-transmitting Anopheles species [3]–[5] and catastrophic consequences are predicted for disease control if major pyrethroid failure occurs [6]. With no entirely new insecticide classes for public health anticipated for several years [5], [6] preservation of pyrethroid efficacy is critically dependent upon strategies such as rotation or combination of pyrethroids with just three other insecticide classes, organochlorines, carbamates and organophosphates [6], [7]. In addition to logistical and financial issues, insecticide resistance management suffers from knowledge-gaps concerning mechanisms causing cross-resistance between available alternative insecticides, and more, generally how high-level resistance arises [8]. With strongly- and multiply-resistant phenotypes documented increasingly in populations of the major malaria vector Anopheles gambiae in West Africa [9]–[13] such information is urgently required.
Of the four classes of conventional insecticide licensed by the World Health Organisation (WHO), pyrethroids and DDT (the only organochlorine) both target the same para-type voltage-gated sodium channel (VGSC). This creates an inherent vulnerability to cross-resistance via mutations in the VGSC target site gene [14]–[16], which are now widespread in An. gambiae [5]. In contrast, carbamates and organophosphates cause insect death by blocking synaptic neurotransmission via inhibition of acetylcholinesterase (AChE), encoded by the ACE-1 gene in An. gambiae. Consequently, target site mutations in the VGSC gene producing resistance to pyrethroids and DDT will not cause cross-resistance to carbamates and organophosphates. The carbamate bendiocarb is being used increasingly for IRS [17], [18], and has proved effective in malaria control programs across Africa targeting pyrethroid- or DDT-resistant An. gambiae [18]–[20]. A single nucleotide substitution of glycine to serine at codon position 119 (Torpedo nomenclature; G119S) in the ACE-1 gene, which causes a major conformational change in AChE, has arisen multiple times in culicid mosquitoes [21], [22], and is found in An. gambiae throughout West Africa [23]–[25]. The G119S mutation can produce carbamate or organophosphate resistance [26] but typically entails considerable fitness costs [27]–[30]. This is beneficial for resistance management because in the absence of carbamates or organophosphates, serine frequencies should fall rapidly [29], [31]. In Culex pipiens, duplications of ACE-1 create linked serine and glycine alleles, which, when combined with an unduplicated serine allele, creates highly insecticide resistant genotypes with near-full wild-type functionality, thus providing a mechanism that can compensate for fitness costs [28], [31]. Worryingly, duplication has also been found in An. gambiae [23] though the consequences of copy number variation for fitness in the presence or absence of insecticide are not yet known in Anopheles. Though far from complete, information is available for metabolic resistance mechanisms to pyrethroids and DDT in wild populations of An. gambiae [5], [6], [32]–[34]. Indeed, a specific P450 enzyme, CYP6M2, has been demonstrated to metabolize both of these insecticide classes, suggesting the potential to cause cross-resistance in An. gambiae [32], [35]. By contrast little is known about metabolic mechanisms of carbamate resistance in mosquitoes and, as a consequence, potential for mechanisms of cross-resistance are unknown.
A particularly striking and potentially problematic example of insecticide resistance has been found in one of the two morphologically identical, but ecologically and genetically divergent molecular forms comprising the An. gambiae s.s. species pair (M molecular form, recently renamed as An. coluzzii [36]) in Tiassalé, southern Côte d'Ivoire. The Tiassalé population is resistant to all available insecticide classes, and displays extreme levels of resistance to pyrethroids and carbamates [11]. The VGSC 1014F (‘kdr’) and ACE-1 G119S mutations are both found in Tiassalé [11], [25]. Yet kdr shows little association with pyrethroid resistance in adult females in this population [11]. ACE-1 G119S is associated with both carbamate and organophosphate survivorship [11], but this mutation alone cannot fully explain the range of resistant phenotypes, suggesting that additional mechanisms must be involved. Here we apply whole genome microarrays, transgenic functional validation of candidates, insecticide synergist bioassays, target-site genotyping and copy number variant analysis to investigate the genetic basis of (1) extreme bendiocarb resistance and (2) cross-insecticide resistance in An. gambiae from Tiassalé. Our results indicate that bendiocarb resistance in Tiassalé is caused by a combination of target site gene mutation and duplication, and by specific P450 enzymes which produce resistance across other insecticide classes.
Our study involved two microarray experiments (hereafter referred to as Exp1 and Exp2), involving solely M molecular form An. gambiae (Table S1), to identify candidate genes involved in bendiocarb resistance (full microarray results for Exp1 and Exp2 are given in Table S2A). In Exp1 gene expression profiles of female mosquitoes from bendiocarb-susceptible laboratory strains (NGousso and Mali-NIH) and a bendiocarb-susceptible field population (Okyereko, Ghana), none of which were exposed to insecticide, were compared to those of Tiassalé females. Two Tiassalé groups were used: either without insecticide exposure (Figure 1A), or the survivors of bendiocarb exposure selecting for the 20% most resistant females in the population [11] (Figure 1B). We used a stringent filtering process to determine significant differential expression (detailed in the legend to Figure 1), which included criteria on both the probability and consistency of direction of differential expression, and also required a more extreme level of differential expression in the Tiassalé-selected than Tiassalé (unexposed) vs. susceptible comparisons. Inclusion of this third criterion enhanced the likelihood that genes exhibiting differential expression are associated with bendiocarb resistance, rather than implicated via indirect association with another insecticide. Moreover, the requirement for significance in comparisons involving both bendiocarb-exposed and unexposed Tiassalé samples (Figure 1A, B) negates the possibility that any differential expression identified was a result solely of induction of gene expression by insecticide exposure.
In Exp1 145 probes were significant, out of a total of 14 914 non-control probes, with almost all (143/145) expressed at a higher level in the resistant samples (Table S2B). Functional annotation clustering analysis detected two significant clusters within the significantly over-expressed genes (Table S2C). The larger cluster was enriched for several P450s and the functionally-related genes cytochrome b5 and cytochrome P450 reductase. Of these, CYP6P3, CYP6P4, CYP6M2 and cytochrome b5 are evident amongst the most significant and/or over-expressed probes in Figure 2A. Of the five physically-adjacent CYP6P subfamily genes in An. gambiae, CYP6P1 and CYP6P2 were also significant (Table S2B), and CYP6P5 only marginally non-significant according to our strict criteria (five out of the six comparisons q<0.05). The four probes for the ACE-1 target site gene exhibited the strongest statistical support (lowest q-values) for resistance-associated overexpression in the Exp1 dataset (Figure 2A).
Experiment 2 employed a simpler design in which bendiocarb resistant samples from Kovié (Togo) were compared to the same Okyereko field samples used in Exp1 and to a second field population from Malanville (Benin). Significant differential expression was determined according to the first two criteria employed for analysis of Exp1 (Figure 1). The likelihood of specificity of results to the bendiocarb resistance phenotype was enhanced because all three populations used in Exp2 exhibit resistance to pyrethroids and DDT, all are susceptible to organophosphates, but only the Kovié population is resistant to bendiocarb. In Exp2 2453 probes were significantly differentially expressed (Table S2D); likely reflecting the lower number of pairwise comparisons available for stringent filtering than in Exp1. Consequently we do not consider results from Exp2 alone in detail. Nevertheless it is interesting to note that the lowest q-values and highest fold-changes were both for alcohol dehydrogenase genes (Figure S1), and the latter is the physical neighbour and closest paralogue of the highly overexpressed alcohol dehydrogenase in Exp1 (Figure 2A). Sixteen probes, representing only seven genes, were significant in both Exp1 and Exp2 (Figure 2B), including all replicate probes for three of the CYP6 P450 genes highlighted previously. Of these, CYP6M2 was most highly over-expressed, second only to Ribonuclease t2. However, results for Ribonuclease t2 were much more variable, with differential expression dramatically high compared to lab strains, but moderate or low compared to wild populations (Table S2E). Evidence for specific involvement in bendiocarb resistance is suggested by significance of two of the CYP6M2 probes in the (relatively low-powered) direct comparison of bendiocarb selected vs. unselected samples within Exp1; the other two CYP6M2 probes and two of those for ACE-1 were marginally non-significant (0.05<q<0.10; Figure S2).
Five genes were chosen for further analysis: ACE-1 and CYP6P3 from Exp1; CYP6M2 and CYP6P4 from Exp1+Exp2; and CYP6P5, which we included because of a suspected type II error in the microarray analysis (see above). qRT-PCR estimates of expression, relative to the susceptible Okyereko population, showed reasonable agreement with microarray estimates albeit with some lower estimates (Figure S3). CYP6M2 and CYP6P4 exhibited up to eight and nine-fold overexpression, and ACE-1 six-fold compared to Okyereko, though high variability among biological replicates for the P450 genes resulted in relatively few significant pairwise comparisons (Figure 3). Nevertheless the hypothesis that fold-changes should follow the rank order predicted by the level of bendiocarb resistance in each comparison (i.e. Tiassalé selected>Tiassalé unexposed>Kovié) was met qualitatively for all genes (Figure 3).
For functional validation via transgenic expression in D. melanogaster, we chose CYP6P3 and CYP6M2; both of which have been shown to metabolize pyrethroids [34], [35], and CYP6M2 also DDT [32]. The capacity of each gene to confer resistance to bendiocarb, to the class I and II pyrethroids permethrin and deltamethrin, respectively, and to DDT and was assessed by comparing survival of transgenic D. melanogaster, exhibiting ubiquitous expression of CYP6M2 or CYP6P3 (e.g. UAS-CYP6M2/ACT5C-GAL4 experimental class flies), to that of flies carrying the UAS-CYP6M2 or CYP6P3 responder, but lacking the ACT5C-GAL4 driver (e.g. UAS-CYP6M2/CyO control class flies). For CYP6M2 the relative expression level of the experimental flies was 4.0 and for CYP6P3 4.3 (Table S3). As indicated by elevated LC50 values (Figure S4), expression of either CYP6M2 or CYP6P3 produced pyrethroid resistant phenotypes, and CYP6M2 expression also induced significant DDT resistance (Table 1). Assays for CYP6P3 with DDT did not produce reproducible results (data not shown). Flies expressing the candidate genes exhibited greater survival across a narrow range of bendiocarb concentrations (Figure S4). However, at a discriminating dosage of 0.1 µg/vial [37] a resistance ratio of approximately seven was exhibited for CYP6M2/ACT5C: CYP6M2/CyO flies (Mann-Whitney, P = 0.0002; Figure 4) with a much smaller, but still significant, ratio of approximately 1.4 (Mann-Whitney, P = 0.019) for CYP6P3/ACT5C: CYP6P3/CyO flies. Caution is required in quantitative interpretation of the resistance levels generated, both because of the non-native genetic background and also ubiquitous expression of genes that may be expressed in a tissue-specific manner [38]. Nevertheless, the bioassays on transgenic Drosophila show that each P450s can confer resistance to more than one insecticide class.
Recombinant CYP6M2 and CYP6P3 were expressed in E. coli with An. gambiae NADPH P450 reductase and cytochrome b5. An initial experiment, using 0.1 µM P450 and 2 hour incubation with bendiocarb, demonstrated metabolism of bendiocarb by CYP6P3 (64.2% mean depletion ±4.0% st.dev) but no metabolic activity of CYP6M2 (0±11.0%). Further investigation of CYP6P3 activity across a range of incubation times (Figure 5a) and enzyme concentrations (Figure 5b) supported the initial observation, with metabolism plateauing at a maximum of 50%.
An. gambiae from Tiassalé are classified as resistant to all classes of WHO-approved insecticides (<90% bioassay mortality 24 hours after a 60 min exposure), with resistance phenotypes stable across wet and dry seasons (Figure 6, Table S4). Nevertheless, resistance varies markedly among insecticides (Table S4), with notably higher prevalence for bendiocarb and DDT than the organophosphate fenitrothion. The synergist PBO, which is primarily considered an inhibitor of P450 enzymes, exerted a significant influence on bioassay mortality (Table S4) for four of the five insecticides tested, with only DDT not significantly impacted (Figure 6). The synergising effect of PBO was strongest for bendiocarb, with a near five-fold increase in mortality, equivalent to an odds ratio for PBO-induced insecticidal mortality exceeding ten (Figure 6). However, for all of the insecticides, apart from fenitrothion, over 20% of the population survived even with PBO pre-exposure.
The ACE-1 G119S substitution is the only non-synonymous target site mutation known in An. gambiae [23], and the resistant (serine) allele is common in Tiassalé with an estimated frequency of 0.46 (N = 306). All occurrences of serine are in heterozygotes (95% confidence limits for heterozygote frequency: 0.87–0.94), which underlies a dramatic deviation of genotype frequencies from Hardy-Weinberg equilibrium (÷2 = 135.5, P≈0). To examine the independence of putatively P450-mediated resistance and AChE target site insensitivity, we typed the G119S locus in females from the diagnostic (60 min) bendiocarb assays with and without pre-exposure to PBO. In either case absence of the 119 serine allele appears to almost guarantee mortality to bendiocarb (Table S5), as previously observed for fenitrothion bioassays in Tiassalé [11]. However, the strong bendiocarb resistance association of G119S was reduced significantly by PBO pre-exposure (homogeneity ÷2 = 8.3, P = 0.004) with the probability of survival for heterozygotes reduced to approximately 50% (Table S5). To investigate whether heterozygote survivorship might be linked to copy number variation, via a difference in numbers of serine and glycine alleles, we examined the qPCR dye balance ratio for live and dead individuals within the heterozygote genotype call cluster (Figure 7A). In many individuals called as heterozygotes, a markedly higher ratio of 119S: 119G dye label than the 1∶1 expected for a true heterozygote is evident (Figure 7A), and surviving heterozygotes exhibited a significantly higher serine: glycine dye signal ratio than those killed (t-test, P = 1.5×10−5). We designed an additional qRT-PCR diagnostic to investigate copy number more directly in a portion of the surviving and dead individuals typed as G119S heterozygotes. The difference in copy number was highly significant between survivors and dead (Figure 7B), with 15/16 survivors but only 5/16 dead females exhibiting a copy number ratio in excess of 1.5 (Table S5), consistent with possession of an additional allele. These results show that independent of the enzymes inhibited by PBO survival, females heterozygous for the G119S mutation (i.e. most individuals in Tiassalé) depends upon Ace-1 copy number variation and possession of additional resistant serine alleles.
Bendiocarb is an increasingly important alternative to pyrethroids for IRS, but with carbamate resistant malaria vectors now established in West Africa [9]–[13] detailed understanding of the underlying mechanisms is urgently required to combat resistance and avoid cross-resistance [6]. Exhibiting the most extreme carbamate resistance and multiple insecticide resistance phenotypes documented to date in An. gambiae [11], the Tiassalé population represents an especially suitable model to address this question. Our results show how P450s contribute to multiple resistance in Tiassalé, and couple with overexpression of ACE-1 resistant alleles to produce extreme bendiocarb resistance.
The major biochemical mechanisms of carbamate resistance in mosquitoes have previously been identified as modified AChE (via point substitutions, most notably G119S) and less frequently esterase-mediated metabolism [7]. PBO-induced increases in carbamate mortality have been reported in wild mosquito populations exhibiting low to moderate resistance levels, including M form An. gambiae from West Africa [12], [39],[40]. The significant synergizing effect of PBO in the present work and these previous studies is consistent with a role of P450s in carbamate resistance, but should not be taken alone as direct proof [41] because PBO exposure can also inhibit some esterases [42], [43]. However, our microarray data clearly identified over-expression of multiple CYP6 P450 genes, whereas only a single carboxylesterase gene (COEAE6G) was significant, and expressed at a lower level (Table S2B). Taken together, the synergist data and transcriptional profiles indicate that a substantial proportion of the Tiassalé population is dependent upon the action of P450s for resistance to bendiocarb. Near-equivalent synergism of permethrin and deltamethrin, coupled with identification and functional validation of shared candidate genes, suggests the same conclusion for pyrethroids. For fenitrothion, the effect of PBO is also consistent with P450 involvement, but in the absence of specific candidate genes, additional supporting evidence will be required to confirm this hypothesis.
Genes from the CYP6P cluster emerged as strong candidates for involvement in P450-mediated detoxification. CYP6P3 overexpression has been linked repeatedly with pyrethroid resistance in An. gambiae [33], [34], as has its orthologue in An. funestus CYP6P9 [44], [45] and both enzymes can metabolise class I and II pyrethroids [34], [35], [45]. We demonstrate that CYP6P3 can produce significant resistance to both classes of pyrethroid and, to a lesser extent bendiocarb, in D. melanogaster. We also show that recombinant CYP6P3 can metabolise bendiocarb in vitro; the third mosquito P450 to metabolise a carbamates, after An. gambiae CYP6Z1 and CYP6Z2 which have been demonstrated to metabolise the insecticide carbaryl [46]. Interestingly CYP6P4, which, in contrast to CYP6P3, was also significantly overexpressed in the Togolese Kovié population, is the orthologue of the resistance-associated CYP6P4 gene in An. funestus [44], and along with CYP6P3 was recently found to be overexpressed in DDT-resistant samples of both M and S molecular forms of An. gambiae from Cameroon [47]. Although we were unable to obtain data for the impact of CYP6P3 expression on survival with DDT exposure in D. melanogaster, the potential of CYP6P genes to act on DDT merits further investigation. It is also interesting to note that both cytochrome b5 and cytochrome P450 reductase, both important for P450-mediated insecticidal detoxification [48] are overexpressed in Tiassale, suggesting a possible role in resistance for co-expression of these genes with the CYP6 P450s.
CYP6M2 was overexpressed in Tiassalé, Kovié, and also in the Tiassalé bendiocarb-selected vs. control comparison. CYP6M2 expression generated Drosophila phenotypes significantly resistant to bendiocarb, DDT, and class I and II pyrethroids. Overexpression of CYP6M2 has been linked repeatedly to pyrethroid [33], [34] and DDT resistance [32], [47] in An. gambiae, and is known to metabolise both these classes of insecticide [32], [35]. Our data now suggest a role in bendiocarb resistance, and overall provide strong evidence for involvement in resistance to three classes of insecticide. The biochemical mechanism of involvement remains unclear however because CYP6M2 did not metabolise bendiocarb in vitro, though we cannot rule out the possibility that some unknown, and thus currently, absent co-factor might be required. Sequestration also seems unlikely since CYP6M2 does not appear to bind bendiocarb. A role in breakdown of secondary bendiocarb metabolites certainly remains plausible, though at present knowledge of such mechanisms for any insecticide in mosquitoes is very limited [49], [50]. High variability in CYP6M2 expression among biological replicates, especially evident in qRT-PCR, suggests that the regulatory mechanism(s) generating overexpression is far from fixation in Tiassalé. Further work is required to determine whether the cause of overexpression might be gene amplification, as seen for insecticide-linked CYP6P genes in An. funestus [44] and CYP6Y3 in the aphid Myzus persicae [51] or a cis regulatory variant, or both, as documented for CYP6G1 in D. melanogaster [52]. In either case, the actual level of expression in individuals possessing causal regulatory variant(s) may be much higher than we detected from pooled biological replicates. As a consequence, it is possible that CYP6M2 (and other key P450s) might be expressed at too high a level for PBO to fully inhibit at the dosage applied, resulting in only partial synergy. Indeed it is interesting that CYP6M2 generated significant DDT resistance in transformed Drosophila in our study and has been shown metabolise DDT [32] yet PBO provided only very slight and non-significant synergy for DDT-exposed Tiassalé females. An inadequate concentration of PBO might be important, but it is worth noting that levels of DDT resistance in West African An. gambiae can be extreme and are likely to be underpinned by additional mechanisms [32] such as the significantly resistance-associated kdr L1014F target site mutation in Tiassalé [11]. Whilst incomplete synergy of highly expressed P450 enzymes might be a partial explanation, our results point to target site mechanisms as a key factor underpinning survival following PBO and bendiocarb exposure.
Possession of the ACE-1 119 serine variant appears to be a near-prerequisite for bendiocarb-survival in Tiassalé, as documented previously for fenitrothion [11]. This is apparently not the case in all An. gambiae populations, with some individuals lacking the serine mutation surviving a standard 60 min exposure [12], [39]. Over 90% of Tiassalé mosquitoes are heterozygous for G119S, which could be consistent with fitness costs for individuals lacking a fully-functional wild-type allele since the serine allele exhibits lowered activity [28]. It is apparent though that possession of the ACE-1 G119S mutation represents only a portion of the target site mediated resistance mechanism. Tiassalé females generally showed much higher expression of ACE-1 than all other populations in our experiments, reaching approximately six-fold in the highly resistant bendiocarb-selected group compared to the Okyereko susceptible group. Following PBO-mediated P450 inhibition, survival of G119S heterozygotes was reduced to approximately 50% and our results show that individuals exhibiting a higher ACE-1 copy number and more copies of the serine allele had a significant survival advantage. Together these results indicate that the primary explanation for the ubiquitous heterozygosity found in Tiassalé is an elevated copy number of expressed ACE-1 alleles. At least in individuals possessing additional serine alleles, this enhances carbamate resistance, and can apparently generate resistance independently of P450 activity.
Extra copies of ACE-1 alleles have been found in West African An. gambiae, and lack of sequence variation suggests that duplication is a very recent event [23]. Consequences of ACE-1 duplication have not been documented previously in Anopheles but Cx. pipiens possessing two G119S resistant alleles and a wild type susceptible allele can exhibit near maximal fitness in the presence and absence of organophosphate treatment [30]. If this fitness scenario is similar in An. gambiae ACE-1 duplicates could spread rapidly, or may have already done so but have been largely undetected by available diagnostics. The estimated copy numbers we detected in some individuals suggests that more ACE-1 copies may be present in An. gambiae than are known in Cx. pipiens, perhaps more akin to the high level of amplification found in spider mites Tetranychus evansi [53]. This raises the possibility of a potentially multifarious set of resistant phenotypes dependent upon the number and G119S genotype of the copies possessed by an individual, understanding of which will benefit from further application of the DNA-based qPCR diagnostic we have developed.
Extreme levels of resistance to single insecticides, and multiple resistance across different insecticidal classes represent major problems for control of disease vectors, and pest insects generally. Tiassalé An. gambiae show exceptionally high-level carbamate resistance and the broadest insecticide resistance profile documented to date. Our results indicate that overexpression of specific CYP6 enzymes and duplicated resistant ACE-1 alleles are major factors contributing to this resistance profile. Results from the less resistant Kovié population show that at least some of the mechanisms are not restricted to Tiassalé and could be quite widespread in West Africa. The involvement of CYP6P3 and CYP6M2 in resistance to multiple insecticide classes parallels the cross resistance engendered by CYP6 genes in other insect taxa [54], [55] and is extremely concerning because resilience to standard resistance management strategies is likely to be increased greatly. Further work is now required to understand the biochemical role of CYP6M2 in detoxification of bendiocarb and also to better understand any associated fitness costs of elevated CYP6P gene expression. In addition, whilst we have demonstrated involvement of elevated expression of the CYP6 P450s in insecticide resistance, the impact of structural variants within these genes remains to be investigated and is very poorly understood for P450-mediated insecticide resistance in mosquitoes. In spite of a major impact of PBO on three distinct insecticide classes, too many females remained alive to suggest that PBO provides a resistance-breaking solution. Nevertheless, we suggest that this preliminary conclusion may be worth further testing: (i) using higher PBO concentrations; (ii) in females old enough to transmit malaria, which are usually less insecticide resistant [56]–[58]; or (iii) in less resistant populations. Monitoring the spread of ACE-1 duplications should be an immediate priority, whereas modification of AChE-targeting insecticides to reduce sensitivity to the G119S substitution [59], [60] represents an important longer-term goal.
Our study involved Anopheles gambiae samples for bioassays coupled with target site genotyping and copy number analysis, and two microarray experiments. The first (Exp1; see Figure 1A, B) compared samples from laboratory strains or field populations entirely susceptible to carbamates, with bendiocarb-resistant females from Tiassalé, which were also the subject of bioassays. Exp2 (see Figure 1C) involved a comparison of a population moderately resistant to bendiocarb (Kovié) with two fully carbamate susceptible field populations. Sample site details and resistance profiles for each population or strain used in the microarrays are given in Table S1. For field populations, larvae were collected and provided with ground TetraMin fish food. Emerged adults were provided 10% sugar solution. All 3–5 day old females for subsequent gene expression analysis were preserved in RNALater (Sigma). With the exception of a selected group from the Tiassalé population (below), all samples were preserved without exposure to insecticide. The Tiassalé selected group were survivors of exposure to 0.1% bendiocarb (using WHO tubes and papers) for 360 min which induces approximately 80% mortality after 24 h (11); unexposed controls were held for 360 min with control paper, which did not induce mortality. All mosquitoes used in the study were identified as An. gambiae s.s. M molecular form using the SINE-PCR method [61].
The effect of the insecticide synergist piperonyl butoxide (PBO), a primary action of which is to inhibit P450 monooxygenase enzymes [41], was evaluated using WHO bioassays. Eight replicates of 25 adult female An. gambiae emerging from larvae obtained from an irrigated rice field in Tiassalé were exposed to five insecticides (permethrin, deltamethrin, DDT, bendiocarb and fenitrothion). Immediately prior to each 60 min insecticide exposure, mosquitoes were exposed to 4% PBO paper for 60 min. 100 females were exposed to PBO alone as control. Chi-squared tests were used to compare the mortality with and without PBO. A TaqMan qPCR assay [62] run on an Agilent Stratagene real-time thermal cycler was used to genotype PBO-exposed samples for the ACE-1 G119S polymorphism, with qualitative calling of genotypes based on clustering in endpoint scatterplots. G119S genotype call data for samples not exposed to PBO was taken from a prior publication [11]. Following qualitative genotype calling, endpoint dR values for each dye were exported, and the data from individuals called as heterozygotes was analyzed quantitatively to investigate the possibility of sub-grouping within this genotype cluster. Specifically we tested whether surviving and dead mosquitoes, heterozygous for G119S, might possess different numbers of serine and alleles by comparing FAM (serine label)/VIC (glycine label) dye ratios using an unequal variance t-test. To further quantify the copy number variation suggested by the TaqMan genotyping results we designed a qRT-PCR to amplify fragments from three different exons of the ACE-1 gene, with normalisation (for varying gDNA concentration among samples) provided via comparison with amplification of a fragment from each of two single-copy genes CYP4G16 and Elongation Factor. Primer details are given in Table S6 and qRT-PCR conditions are the same as listed below for gene expression analysis. Relative copy number levels for Ace-1 were estimated relative to two pools of samples (N = 4 each) from the Kisumu laboratory strain by the ΔΔCT method [63]. ΔΔCT values for each test sample are the mean for the three ACE-1 amplicons following normalisation to both single copy genes and subtraction of the average normalised Kisumu values. Test samples were 16 ACE-1 G119S heterozygote survivors and 16 dead, chosen at random from those genotyped by the TaqMan assay. ΔΔCT values were compared between survivors and dead using an unequal variance t-test.
Total RNA was extracted from batches of 10 mosquitoes using the Ambion RNAqueous-4PCR Kit. RNA quantity and quality was assessed using a NanoDrop spectrophotometer (Thermo Fisher Scientific) and a 2100 Bioanalyzer (Agilent Technologies) before further use. Three biological replicate extractions of total RNA from batches of 10 mosquitoes for each sample population or colony (except Ngousso where there were N = 2 replicates) were labelled and hybridised to Anopheles gambiae 8×15 k whole genome microarrays using previously described protocols [32]. Exp 2 employed a fully-interwoven loop design (Figure S6), optimal for study power [64] whilst, owing to the large number of comparisons and unbalanced replication, a pairwise full dye-swap design was used for Exp1 with indirect connection through the (resistant) Tiassalé groups (Fig. 1 A, B). Exp1 was analysed using GeneSpring GX v9.0 software (Agilent), which is readily applied to dye swap experiments, while the R program MAANOVA [65], with LIMMA [66] for normalisation prior to ANOVA, was used to analyse the interwoven loop in Exp2, using previously-described custom R-scripts [32]. For both experiments, the basic significance threshold for any single pairwise comparison was a q-value with false discovery rate (FDR) set at 0.05 (i.e. an FDR-corrected threshold for multiple testing). Full details of the criteria applied to determine overall significance within and across Exp1 and 2 are given in Figure 1. Within Exp1, the direct comparison of Tiassalé bendiocarb-selected vs. Tiassalé control comparison was analysed separately and not used to determine overall significance, owing to the lower power expected for a within-population experiment involving the same level of replication as the cross-population comparisons [34]. Significantly over-expressed genes emerging from Exp1 were studied at functional level using the software DAVID Bioinformatics resources 6.7 [67]. Microarray data are deposited with ArrayExpress under accession numbers E-MTAB-1903 (Exp1) and E-MTAB-1889 (Exp2).
Quantitative real-time PCR was used to provide technical replication of results from the microarray experiments for a subset of significantly over-expressed genes. Samples were converted to cDNA using oligo(dT)20 (Invitrogen) and Superscript III (Invitrogen) according to the manufacturer's instructions and purified with the QIAquick PCR Purification Kit. Three pairs of exon-spanning primers were designed for each gene of interest and from each triplicate a pair was chosen that produced a single peak from melt cure analysis, and PCR efficiency closest to 100%, determined using a cDNA dilution series obtained from a single sample. Primers details are listed in Table S7. All qRT-PCR reactions were run on an Agilent Stratagene real-time thermal cycler and analysed using Agilent's MXPro software (Mx3005P). The PCR conditions used throughout were 10 min for 95°C, 40 cycles of 10 s at 95°C and 60°C respectively, with melting curves run after each end point amplification at 1 min for 95°C, followed by 30 s increments of 1°C from 55°C to 95°C. The same RNA samples used for microarrays from Tiassalé (selected and unexposed), Kovié and Okyereko plus an additional two replicates (N = 5 for all but the Tiassalé selected group where N = 3) were used. Expression levels for each gene of interest were estimated relative to the Okyereko population (chosen as the reference bendiocarb susceptible group because it was present in both microarray experiments) by the ΔΔCT method following correction for variable PCR efficiency [63], and normalisation using two stably-expressed genes (Rsp7 and Elongation Factor); primers and efficiencies are listed in Table S7. Statistical significance of over-expression of each group relative to Okyereko was assessed using equal or unequal variance t-tests as appropriate, depending on results of F-tests for homoscedasticity.
cDNA clones containing the open reading frames for CYP6M2 and CYP6P3 (sequences from the An. gambiae Kisumu laboratory strain) were PCR-amplified using high fidelity AccuPrime Pfx polymerase (Invitrogen). PCR primers contained EcoRI and NotI restriction sites within the forward and reverse primers, respectively. PCR products were gel-purified using the GenElute Gel Extraction Kit (Sigma) and subsequently digested with the aforementioned restriction enzymes (New England Biolabs). The pUAST-attB plasmid (obtained from Dr. Konrad Basler, University of Zurich) digested with EcoRI and NotI was gel purified, as noted above, and incubated with PCR-amplified, restriction enzyme-digested products of the CYP6M2 or CYP6P3 clone and T4 DNA ligase (New England Biolabs). Ligation mixtures were transformed into competent DH5α cells, and individual colonies were verified using PCR. The EndoFree Plasmid Maxi Kit (Qiagen) was utilized to obtain large amounts of plasmids for subsequent steps. pUAST-attB clones containing the CYP6M2 or CYP6P3 insertion were sent to Rainbow Transgenic Flies, Inc. (Camarillo, CA, USA) for injection into Bloomington Stock #9750 (y1 w1118; PBac{y+-attP-3B}VK00033) embryos. The PhiC31 integration system in this stock enables site-specific recombination between the integration vector (pUAST-attB) and a landing platform in the fly stock (attP) [68]. Upon receiving the injected embryos, survivors were kept at 25°C, and Go flies that eclosed were sorted by sex prior to mating. To establish families of homozygous transgenic flies, Go flies were crossed with w1118 flies, and G1 flies were sorted based on w+ eye color (as a marker for insertion events). G1 w+ flies were crossed inter se to obtain homozygous insertion lines. The following D. melanogaster stocks were obtained from the Bloomington Drosophila Stock Center (Bloomington, IN, USA): y1 w1; P{Act5C-GAL4}25FO1/CyO, y+, w* (BL4414); P{GawB}Aph-4c232 (BL30828), and w1118 (BL3605). Virgin females from CYP6M2 or CYP6P3 insertion stocks were crossed with Act5C-GAL4/CyO (ubiquitous Actin5C driver) flies for expression studies.
For each class within a cross (control and experimental), 8–10 two-day-old flies were obtained and flash-frozen in liquid nitrogen, and then stored at −80°C in triplicate. Total RNA was extracted using TRI Reagent (Sigma), and 1 µg of RNA was treated with RNase-Free DNaseI (Fisher Scientific). For each synthesis, a 10 µL reaction was created using 1 µL DNase-treated RNA; three technical replicates were performed for each biological replicate. Primers for amplification of cDNA product, used at a concentration of 0.75 µM, were: Cyp6M2_Forward: 5′-ACGAGTTCGAGCTGAAGGAT-3′, Cyp6M2_Reverse: 5′-GTTACACTCAATGCCGAACG-3′, Cyp6P3_Forward: 5′-TATTGCAGAGAACGGTGGAG-3′, Cyp6P3_Reverse: 5′TACTTCCGAAGGGTTTCGTC-3′. Relative expression was compared using Actin primers [69] at a concentration of 0.50 µM. qRT-PCR reactions were performed using USB VeriQuest SYBR Green One-Step qRT-PCR Master Mix (2X) on a 7500 Fast Real-Time PCR System (Applied Biosystems). Cycling conditions used were 50°C for 10 minutes and 95°C for 10 minutes, followed by 40 cycles of 90°C for 15 seconds and 56°C for 30 seconds, with the fluorescence measured at the end of each cycle.
Recombinant CYP6M2 and CYP6P3 were commercially co-expressed with An. gambiae NADPH P450 reductase and cytochrome b5 in an E. coli system by Cypex (Dundee, UK). Using previously described methodologies [35] a first experiment showed that CYP6M2 was unable to metabolise bendiocarb (10 µM) after a 2 hour incubation and thus only CYP6P3 was investigated in subsequent experiments. For time course measurements, reactions were performed in 200 µL with 10 µM insecticide, 0.1 µM CYP6P3 membrane in 200 mM Tris-HCl pH 7.4 and started by adding the NADPH regenerating system (1 mM glucose-6-phosphate (G6P), 0.25 mM MgCl2, 0.1 mM NADP+, and 1 U/mL glucose-6-phosphate dehydrogenase (G6PDH)). Reactions were incubated for a specified time at 30°C with 1200 rpm orbital shaking and stopped by adding 0.2 mL of acetonitrile. Shaking was carried for an additional 10 min before centrifuging the reactions at 20000 g for 20 min. 200 µl of supernatant was used for HPLC analysis. Reactions were performed in triplicate and compared against a negative control with no NADPH regenerating system to calculate substrate depletion. An additional experiment with different enzyme concentrations was performed, using the methods above, for 20 mins with P450 concentrations of: 0.2, 0.1, 0.075, 0.05, 0.025 and 0.0125 µM. The reactions were performed in parallel against a negative control (−NADPH).
In each experiment the supernatants were analyzed by reverse-phase HPLC with a 250 mm C18 column (Acclaim 120, Dionex) and a mobile phase consisting of 35% acetonitrile and 65% water. The system was run at a controlled temperature of 42°C with 1 ml/min flow rate. Bendiocarb insecticide was monitored at 205 nm and quantified by measuring peak areas using OpenLab CDS (Agilent Technologies). Retention time was around 14.9 minutes.
An appropriate amount of insecticide was added to 100 µl of acetone and placed into individual 16×200 mm glass disposable culture tubes (VWR Scientific). Tubes were then placed on their sides and rotated continuously, coating the entire interior of the tube, until all acetone was evaporated. A total of 8–12 control and 8–12 experimental transgenic flies, aged 3–5 days post-eclosion, were added to each tube. Flies from experimental and control classes were mixed in single insecticide-coated vials for assays, to ensure equivalent exposure to insecticide. The tubes were capped with cotton balls saturated with a 10% (w/v) glucose/water solution. Tubes were then incubated at 25°C for 24 h, after which mortality was assessed. Linear regression models were used to fit dose-response curves, from which LC50 values (and confidence intervals) were estimated using Prism v5.0. However, for bendiocarb this was not possible owing to a very sharp inflection in the dose-response profile. Instead differences between lines were assessed at a diagnostic dose of 0.1 µg bendiocarb/vial, applied previously to Apis mellifera [37], [70], using Mann-Whitney U tests.
|
10.1371/journal.pgen.1006976 | Northeast African genomic variation shaped by the continuity of indigenous groups and Eurasian migrations | Northeast Africa has a long history of human habitation, with fossil-finds from the earliest anatomically modern humans, and housing ancient civilizations. The region is also the gate-way out of Africa, as well as a portal for migration into Africa from Eurasia via the Middle East and the Arabian Peninsula. We investigate the population history of northeast Africa by genotyping ~3.9 million SNPs in 221 individuals from 18 populations sampled in Sudan and South Sudan and combine this data with published genome-wide data from surrounding areas. We find a strong genetic divide between the populations from the northeastern parts of the region (Nubians, central Arab populations, and the Beja) and populations towards the west and south (Nilotes, Darfur and Kordofan populations). This differentiation is mainly caused by a large Eurasian ancestry component of the northeast populations likely driven by migration of Middle Eastern groups followed by admixture that affected the local populations in a north-to-south succession of events. Genetic evidence points to an early admixture event in the Nubians, concurrent with historical contact between North Sudanese and Arab groups. We estimate the admixture in current-day Sudanese Arab populations to about 700 years ago, coinciding with the fall of Dongola in 1315/1316 AD, a wave of admixture that reached the Darfurian/Kordofanian populations some 400–200 years ago. In contrast to the northeastern populations, the current-day Nilotic populations from the south of the region display little or no admixture from Eurasian groups indicating long-term isolation and population continuity in these areas of northeast Africa.
| Northeast Africa has geographic and historical links to Eurasia via the Middle East and the Arabian Peninsula, but the demographic history of the region itself has been more elusive. We investigate genomic diversity of northeast African populations and found a clear bimodal distribution of variation, correlated with geography, and likely driven by Eurasian admixture in the wake of migrations along the Nile. This admixture process largely coincides with the time of the Arab conquest, spreading in a southbound direction along the Nile and the Blue Nile. Nilotic populations occupying the region around the White Nile show long-term continuity, genetic isolation and genetic links to ancestral East African people. Compared to current times, groups that are ancestral to the current-day Nilotes likely inhabited a larger area of northeast Africa prior to the migration from the Middle East as their ancestry component can still be found in a large area. Our findings reveal the genetic history of Sudanese and South Sudanese people, broaden our knowledge on demographic history of humans, and quantify the impact of large-scale historic migration events in northeast Africa.
| The Nile River Valley and northeast Africa have experienced a long history of human habitation. The region harbored some of the most ancient civilizations in the world and contains fossil finds of the earliest anatomically modern humans [1–3]. Agriculture has a long history in the Nile River valley, and crops of potential Near Eastern origin as well as sorghum found in Sudan have been dated to 3000BC [4]. Livestock was introduced into northeast African and Sudan in the 5th millennium BC (likely from the North) and pastoralism spread rapidly across sedentary agriculturalists who lived along the Nile as well as to the nomadic populations inhabiting the drier surrounding regions [4]. Following the introduction of agriculture and pastoralism, settlements started growing, which led to the forming of political units. In Nubia (roughly the northern parts of current-day Sudan), the Kingdom of Kerma emerged around 3000 BC. Nubia has successively been at the center of several ensuing states, and the historical records show interactions with neighboring states through trade and confrontation, possibly reaching back to predynastic times [4–6]. Modern-day Sudan and South Sudan cover parts of the Nile River and the joining of the Blue and the White Nile, areas that link the northern part of the Nile Valley and North Africa with East Africa. Today, these areas display great linguistic diversity, with Sudan and South Sudan housing 137 living languages [7], which belong to three of the four linguistic macro-families found on the African continent: Afro-Asiatic, Nilo-Saharan, and Niger-Congo.
Previous genetic studies focusing on human history in Sudan and South Sudan have used uniparentally inherited markers [8–10], low density polymorphic autosomal markers [11–17], or were only covering a limited number of populations [18]. These studies have found substantial genetic differentiation in northeast Africa and indications of migration and admixture. For instance, Tishkoff, Reed [18] investigated more than one hundred African populations using some 800 microsatellites, including six populations from Sudan and South Sudan and showed that eastern Africa harbors substantial amounts of genetic diversity. However, wide ranges of populations, representative of all the main linguistic groupings, in and around Sudan and South Sudan have not been studied in order to decipher population history using high-resolution genome-wide data.
In this study we genotyped some 3.9 million SNPs in 221 individuals from a total of 18 populations from South Sudan and Sudan to investigate population structure and admixture patterns, which we use to reconstruct the genetic history of this region of northeast Africa. We find a genetic differentiation within the Sudanese and South Sudanese groups that is driven by Eurasian admixture, which may have followed the Nile southward and coincides with the time of the Arab conquest.
We investigated the genetic variation of Sudanese and South Sudanese populations by genotyping 221 individuals sampled from 18 populations (Fig 1A, Table 1) using the Illumina Human Omni5MExome array. The sampled populations cover a range of languages belonging to three major linguistic families that include the sub-groupings; Semitic, Cushitic, Eastern Sudanic, Kordofanian, Ancient Egyptian, and Chadic (Fig 1A, Table 1). Some of the sampled populations have been suggested to be recent migrants to the area (such as the Hausa and Copts), while others are assumed to have a long standing history in Sudan (i.e. Nubians) and South Sudan (i.e. Nilotes) [11, 14, 18] (note that we will use population names and/or ethnic grouping, Table 1, when discussing the genetic results).
Following quality filtering (~3.9 million SNPs remained, see SI), we merged the Sudan and South Sudan genotype dataset to relevant published genotype datasets from neighboring and other relevant populations [19–24] (Fig 1A, S1 Table) in order to bring the genetic variation into a regional and global context (SI, Method Section). This dataset is likely the most comprehensive dataset assembled to date of northeast African populations.
Northeast African individuals and groups displayed marked levels of population structure and differentiation (Figs 1B and 2, S1–S6 Figs), and some groups showed strong affinities to groups from other areas, including Europe, Middle East and western Africa (Fig 2, S1–S6 Figs). Focusing on population structure in Sudan and South Sudan, we found that genetic variation was correlated with geography (r = 0.39, p<0.01, Mantel test), to a greater extent than to linguistic classification (r = 0.28, p<0.01), indicating that geography drives population structure in the area. Several populations, in particular from the North and East of Sudan displayed genetic affinities to non-Africans, which is consistent with recent admixture into these groups (Fig 2, S1–S6 Figs). This admixture unifies the Nubian, Arabic and Beja populations from the north, and it is almost completely absent in the western Sudanese and South Sudanese populations.
Among the populations from Sudan and South Sudan, the four Nilotic populations formed a notable population cluster based on the genome-wide data. They were genetically uniform with little genetic differentiation among themselves (pairwise FST values ≤ 0.0028, Fig 1B, S7A Fig). In the ADMIXTURE analyses, the Nilotic populations retained a specific ancestry component (blue), which is shared with other northeast African groups at low values of K, where most of the Sudanese populations have a substantial fraction of this ancestry (Figs 2 and S1–S6). Even at higher values of K, the Nilotes formed their own ancestry component, a component found in modest proportions in populations from Sudan and South Sudan. The Nilotes also appeared as one of the most common source populations for other Sudanese and South Sudanese populations (Figs 2 and 3A). We furthermore compare the affinity between the Nilotes and Neolithic European farmers (represented by an individual from the Linearbandkeramik (LBK)), using the 4,500 year old Mota individual from Ethiopia to represent an East African group that has not been affected by Eurasian admixture in the last 4,500 years [25]. Testing the population tree D(Ju|’hoansi,LBK;Mota,Nilote) shows no support for an affinity between Neolithic European farmers and Nilotes (S8A Fig), as can also be seen from the f4-ratio estimates of Eurasian ancestry in Nilotes (Fig 3B, S9A Fig). Previous studies of uniparental or few markers also found little support for incoming gene-flow to the Nilotic populations [9, 11, 15, 25], and, taken together with our results, Nilotic populations appear to have remained relatively isolated over time.
The Nilotes are predominantly pastoralist populations, they live in Uganda, Ethiopia, Kenya, Tanzania, and are the most prominent ethnicity in South Sudan. They are traditionally strongly endogamic which could account for low levels of admixture. In terms of specific Nilotic populations, the f3 test showed no significant signal of gene flow with external populations for the Nuer and Baria (Fig 3A), however, we detected indications of external gene flow from West Africa (YRI) into Dinka (f3 = -0.001038, Z = -5.283) and TSI to Shilluk (f3 = -0.002565, Z = -7.951, S2 Table). These observations taken together, suggest long term isolation and continuity between the current-day Nilotic populations and the ancestral populations of northeast Africa.
All the investigated Sudanese and South Sudanese populations, except the Hausa, showed almost no West African (orange in Fig 2) component or, at a higher K, Bantu component (Fig 2, yellow in S3 Fig) in the ADMIXTURE analysis. The Bantu migration that swept over most of sub-Saharan Africa 3–4 thousand years ago (kya) [26] did not cause massive admixture in northeast Africa, contrary to what has been found in many other sub-Saharan African regions, e.g. East Africa and southern Africa [18, 27, 28]. This expansion seems to have passed south of the Sudanese Nilotic populations in an eastward direction from West-Africa. The strongly endogamic Nilotic populations could have acted as a migration barrier for northeast Africa preventing admixture with Bantu-speaking groups of West African origin during the migrations of the Bantu expansion, potentially in addition to climatic barriers connected to the agriculture of the Bantu-speakers. Although there are a few Bantu speaking populations in South Sudan [29] that likely migrated during the Bantu expansion, they do not appear to have mixed much with local Nilotic groups.
The Afro-Asiatic speaking Hausa population from northeastern Sudan was the exception to the observation of little West African affinity in Sudan and South Sudan (Fig 1). The Hausa, originally of western Africa, comprises the largest West African population that have migrated to Sudan during the past 300 years, traditionally employed mainly in agricultural activities [30, 31]. In S11 Fig they cluster in between the West African Yoruba and Nzime, and the Darfurian/Kordofanian and Nilotic populations. This finding is consistent with previous analyses [18, 30, 32, 33]. Even though the ADMIXTURE analysis showed some level of local Nilotic genetic material (~30% at K11 and higher, Fig 2, S3 Fig), the f3 statistics did not provide significant evidence for admixture with Darfurian/Kordofanian and Nilotic populations. Using LD decay patterns [34], we estimate an admixture event in the Hausa to 31.2 ± 9.3 generations ago (Z = 3.34683) from a Eurasian source. This is before the historically documented settlement of the Hausa in the Sudan and it is still unknown if the Hausa populations of West Africa also show this admixture signal. These observations point to that the Hausa originated in West Africa and migrated recently to Sudan, where they have stayed relatively isolated from neighboring populations.
The Nubians inhabit the Nile valley in the arid desert of northern Sudan and speak Eastern Sudanic languages of the Nilo-Saharan linguistic family that are close to the languages spoken by Nilotic populations (Table 1, Fig 1A). The Nubian populations have a long history in the region, dating back to dynastic Egypt [5]. They showed little genetic differentiation among individuals and groups, with a maximum (across all pairwise comparisons) pairwise FST (Weir and Cockerham’s estimator) of 0.004513 between the Mahas and the Halfawieen (Fig 1B, S7A Fig). The FST values to the surrounding Arabic and Beja populations were also low, which hints at gene-flow or shared ancestry with the neighboring populations. Even though the Nubians and the Nilotes are linguistically closer to each other than to the Afro-Asiatic groups, the Nubians showed the greatest genetic differentiation (FST between 0.02 and 0.04) to the Nilotes (Fig 1, S7A Fig). To investigate whether this signal of genetic differentiation is driven by the Eurasian admixture into the Nubians (as seen in Fig 2), we created pseudo-‘unadmixed’ (in terms of not having Eurasian admixture) allele frequencies (see SI) and calculated Wright’s FST, which showed that an ‘unadmixed’ Nubian gene-pool is genetically similar to Nilotes (S7B Fig). The strongest signal of admixture into Nubian populations came from Eurasian populations (S10 Fig, S2 Table) and was likely quite extensive: 39.41%-47.73% (f4-ratio, Z-scores between 22.8 and 26.7 Fig 3B, S9 Fig). Interestingly, the Nubians showed the highest level of allelic richness, number of private alleles and shared private alleles (ADZE, between Danagla and Halfawieen, S12 Fig) among all Sudanese and South Sudanese groups. This observation together with a smaller total length of runs of homozygosity, between lengths of 0.5–1 kilobases, points to substantial admixture in Nubians (Fig 4). Hence, the Nubians can be seen as a group with substantial genetic material relating to Nilotes that later have received much gene-flow from Eurasians (likely Middle Eastern) and from East Africans (Fig 2).
All the populations that inhabit the Northeast of Sudan today, including the Nubian, Arab, and Beja groups showed admixture with Eurasian sources and the admixture fractions were very similar. The admixture component in the northeastern groups cluster with the greater European and Middle Eastern group assuming few clusters, and for greater number of assumed clusters, when a predominantly Middle Eastern cluster emerged, the admixture in northeastern Sudan connected to the Middle East (ADMIXTURE, Fig 2, f3, S10 Fig). According to historical and linguistic studies, and recent Y-chromosome data it has been suggested that the northeastern Sudanese populations especially Nubians and Beja were strongly affected by Eurasian migrations since the introduction of Islam from the Arabian Peninsula through Egypt and the Red Sea starting around 651 A.D [9, 35].
Assuming that the Nubian population is a mixture of an incoming Eurasian (TSI is used as a proxy) group and a resident group that is genetically similar to the current day Nilotes (Nuer is used as a proxy), first contact is dated using patterns of LD-decay [34] to roughly 56 generations ago for the Danagla (54.45 ± 10.34, Z = 5.26437) and the Mahas (58.35 ± 12.2, Z = 4.78402); the Halfawieen have received Eurasian admixture later, around 19 generations ago (19.31 ± 3.81, Z = 5.05949, S7 Table, Fig 3C). Assuming a generation time of 30 years, the admixture dates for Danagla and Mahas predate the Arab expansion in the 7th century, and may suggest that the migrations and admixture predate Islamic conquest. However, the confidence intervals overlap with the 7th century, and these admixture estimates largely coincide with the Arab expansion into the northeast of Sudan. It is known from historic sources that Arabic groups encountered the Nubians first in the 7th century, and were held back from advancing further into the Sahel until the fall of Dongola in 1315/1316AD [36] and the collapse of the Kingdom of Makuria. This is consistent with the later date for the admixture into Halfawieen and the Arabic populations of Sudan. Previous studies [37, 38] have found a similar pattern for populations of Maghreb, where admixture times coincide with the time of the historically documented Arab conquest.
The Eurasian migrations also appear to have expanded and migrated into northeast Africa where they admixed with local populations giving rise to Arabic-speaking groups (Shaigia, Gaalien and Bataheen) that today inhabit areas of central Sudan (Fig 2). We further tested the source of admixture into the central Sudan Semitic speaking Arab groups (Shaigia, Gaalien and Bataheen) using ancient samples from Europe (LBK) and East Africa (Mota) and the population history of D(Ju|’hoansi,LBK;Mota,X), (where the Ju|’hoansi is an outgroup Khoe-San population from Namibia), which suggested Eurasian admixture into central Sudan Arab groups (see SI, S8A Fig). This migration and admixture occurred later than the events that brought Eurasian gene-flow into the Nubians (S3 Table, Fig 3C). Interestingly, when we overlay the Eurasian genetic component onto a geographic map, it appears as if the expansion could have spread along the Blue Nile (Fig 3B and 3C), showing a gradient of higher to lower admixture proportion and older to younger admixture dates from northern Sudan to South Sudan. The Eurasian admixture proportion in the Arab populations is high, ranging between ~40%–48% (SI, Fig 3B and S9A Fig). The presence of a northeast African genetic signature similar to Nilotic populations and the recent admixture signal from Eurasia indicates that the populations in central Sudan that self-identify as Arab were originally a local northeast African population (similar to the Nubians and the Beja) that mixed with a Eurasian population during the Arab expansion, or possibly earlier. However, the mixed groups kept the language and culture of the incoming migrants.
Beja groups, who generally reside in eastern areas of Sudan close to the sea, show high non-African admixture in all tests (Figs 2 and 3B, S1–S6 and S8–S10 Figs). The Beni Amer also showed a strong admixture signal with a Eurasian population as well as a shared ancestry component with the Somali population (pink component in Fig 2), which suggest admixture with the East African Cushitic-speaking populations, perhaps as a result of migration along the coast. We dated the admixture of the Beja populations with the Cushitic-speaking Somalian population [39], and the admixture dates go far back in time, about 59 generations ago for the Hadendowa and about 68–75 generations for the Beni Amer (S3 and S4 Tables). The large proportion of the East African (pink in Fig 2) component is therefore not a result of recent admixture of East Africans into the Beni Amer. Admixture of non-Africans into the Beni Amer was also dated to an early event about 107.7 ± 24.4 generations ago (Z = 4.41711) and a younger event, 34.2 generations ago (± 9.6, Z-score = 3.55532 Fig 3C, S7 Table) suggesting an early migration from Eurasian into these coastal African populations, possibly across the sea. However, these old admixture events into the Beni Amer could be driven by admixture from the Cushitic-speaking populations of the Horn of Africa, which themselves have received 30–50% non-African ancestry about 100 generations ago, or 3kya [22, 40].
The Copts represent a well-known ethnic group, generally practicing Christianity, which migrated from Egypt to Sudan around 200 years ago, settling in a predominately Muslim region. The ADMIXTURE analyses and the PCA displayed the genetic affinity of the Copts to the Egyptian population (Fig 2, S1–S6, S11 and S13–S16 Figs). Assuming few clusters, the Copts appeared admixed between Near Eastern/European populations and northeastern Sudanese and look similar in their genetic profile to the Egyptians. Assuming greater number of clusters (K≥18), the Copts formed their own separate ancestry component that was shared with Egyptians but can also be found in Arab populations (Fig 2). This behavior in the admixture analyses is consistent with shared ancestry between Copts and Egyptians and/or additional genetic drift in the Copts [41, 42].
The Copts and the Egyptians have a historically documented shared history. We further investigate the relationships of the Copts and the Egyptians to other groups. All population histories tested in every possible combination of either Copts or Egyptians, and Bedouin and Nuer, with Ju|’hoansi as outgroup to the others were rejected (D-statistic, |Z|>5.5), which points to a non-tree-like history of the Copts and Egyptians. Our results instead indicate that they are an admixed population of at least one sub-Saharan population and one Eurasian population, but had subsequent admixture with additional groups. The population tree that has the most support finds the Nuer (Nilotic) as an outgroup to the Bedouin and Copts (D(Ju|’hoansi,Nuer;Bedouin,Copts) = 0.0103, Z = 5.550). The Copts were estimated to be of 69.54% ± 2.57 European ancestry and the Egyptians of 70.65% ± 2.47 European ancestry (f4-ratio, Fig 3B, S9A Fig).
The Egyptians and Copts showed low levels of genetic differentiation (FST = 0.00236, Fig 1B), lower levels of genetic diversity (S17 Fig) and greater levels of RoH (Fig 4) compared to other northeast African groups, including Arab and Middle Eastern groups that share ancestry with the Copts and Egyptians (Fig 2) [41]. A formal test (D(Ju|’hoansi,X;Egypt,Copt)), did not find significant admixture into the Egyptians from other tested groups (X) as the explanation of the (admittedly low level of) differentiation between the two groups, and the Copts and Egyptians displayed similar levels of European or Middle Eastern ancestry (S8A and S8B Fig). Taken together, these results point to that the Copts and the Egyptians have a common history linked to smaller population sizes, and that the Copts have remained relatively isolated since the arrival to Sudan with only low levels of admixture with local northeastern Sudanese groups (S8B Fig).
The Messiria, a Semitic speaking Arab population, are nomads who inhabit a wide area in the Darfur and Kordofan regions. They were genetically closer to other Darfurian/Kordofanian populations than to the Arab populations of central Sudan (Fig 2, S3 Fig). The Messiria were clearly genetically differentiated from the Arab populations of northeastern Sudan (FST values of 0.0083–0.0229, compared to 0.0–0.0056 to Darfurian/Kordofanian populations, Fig 1B) while the other Arab populations of central Sudan were genetically closer to each other (FST 0–0.0052, Fig 1B). The Messiria showed a significant signal of admixture between Nilotes (Nuer) and Eurasians (TSI), but the signal was stronger for other Arabs (S8 and S10 Figs). The Eurasian fraction in the Messiria was about 15% compared to the (40%-48%) in the northeastern Arabic populations (Fig 3B). The admixture was dated to about 7 generations ago (S3 Table, Fig 3C). This points to the Messiria being a local Kordofanian population that has acquired the language and culture from an incoming Semitic population that they mixed with some 200 years ago (190–244 years ago assuming a generation time of 30 years, Z = 3.19695).
The Gemar, a Nilo-Saharan speaking population of Darfur and Kordofan also showed signals of Eurasian admixture (f3, S10 Fig) estimated to ~13% (Fig 3B, S9A Fig). This admixture event was dated at 13.36 ± 2.99 generations ago (Malder, S7 Table, Fig 3C). However, a proposed population tree of LBK as an outgroup to Mota and Gemar was supported (S8 Fig), suggesting that the Gemar traces much of their ancestry back to ancestral groups of east Africa. The Zaghawa and the Nuba showed very little Eurasian admixture (Figs 1, 2, S8 and S10) and they showed low genetic differentiation to the Gemar and the Messiria as well as to the Nilotic populations suggesting common ancestry of Nilotic, Darfurian and Kordofanian populations (Figs 1B and 2, S7 Fig).
We have shown that there has been long-term migration into Sudan, moving in a southward direction possibly along the Nile and the Blue Nile. From historic documents, we know that the ancient Egyptians were in contact with the ancient Nubians that inhabited the Nile area in the north of modern-day Sudan. Our study suggests that the later migration followed along the Nile, likely being held up by the Nubians until the fall of the Kingdom of Makuria in the 14th Century [4]. Following that historic event, the Arab expansion spread further southward, which can be seen in a succession of admixture events that occur more recent in time as one travels south. Many populations in Sudan that self-identity as Arab, displayed a population history of local Sudanese populations that have admixed with incoming Eurasian populations, and adopted the language and culture of the incoming migrants. In fact most populations from northeast Sudan (Nubian, Arab and Beja groups) seem to be a mixture of Middle Eastern and local northeast African genetic components, although only the Arab groups shifted to the Semitic languages. Cultural and linguistic replacement following the Arab conquest has been described previously in populations of the Maghreb [37, 38, 43].
The Eurasian admixture had less impact on the populations of western Sudan and South Sudan. The Darfurian and Kordofanian populations showed overall less admixture from non-African groups than the northeastern populations (and the limited admixture that does exist is more recent in time). The Nilotic populations have stayed largely un-admixed, which appears to be the case in Ethiopia too, where a similar observation has been made for the Gumuz [23, 44], an Ethiopian Nilotic population that is genetically similar to South Sudan Nilotes. Northeast African Nilotes showed some distinction from an ancient Ethiopian individual (Mota, found in the Mota Cave in the southern Ethiopian highlands), which suggests population structure between northeast and eastern Africa already 4,500 years ago. The modern-day Nilotic groups are likely direct descendants of past populations living in northeast Africa many thousands of years ago.
The DNA samples were chosen from a set of individuals that had been typed with 15 forensic microsatellites [11]. Blood samples were collected by Dr. H. Babiker with a permission from the Forensic DNA lab in Khartoum, Sudan, in 2009. The research purpose of population genomic investigations was described to each participant, and an informed written and oral consent was obtained from all participants. The samples were prepared for analysis using Whatman FTA Protocol BD09 and slightly adjusted Whatman FTA Protocol BD01 (SI). The samples were amplified using Illustra Genomiphi V2 DNA Amplification Kit following the protocol from Pinard, de Winter [45]. Genotyping was performed on an Illumina Human Omni5MExome SNP-array. Data filtering was performed using PLINK v1.07 and custom scripts (S18 and S19 Figs).
Datasets of different sizes were created to include neighboring and other relevant populations, weighing the amount of SNPs against the number of reference populations. Dataset 1 contains the novel populations and the Nzime [24] (~3.5 Million SNPs), dataset 2 contains the populations of dataset 1 and populations from [19, 20, 23] (1.4 Million SNPs), and dataset 3 containing dataset 2 and populations from [22, 46] (~220 thousand SNPS) (S17 Fig). Due to the risk of allelic drop-out (for some individuals) caused by imperfect whole genome amplification, which can result in the appearance of hemizygous stretches (SI), we also created a ‘haploidized’ dataset by randomly picking one allele at each position (if variable). This ‘haploidized’ dataset will avoid underestimating diversity in population samples even in the presence of some level of allelic drop-out (SI-Summary statistics). All results performed on diploid datasets were verified by repeating the analyses with the ‘haploidized datasets’ (S1–S6, S13–S17 and S20–S22 Figs). The datasets were furthermore merged with the Ju|’hoansi population from Namibia (to act as an outgroup), and two ancient individuals, an ancient Ethiopian (Mota), to provide an African sample with no European admixture [25], and a European Linearbandkeramik individual (LBK) as a European reference of Neolithic times [47] (S9, S23 and S24 Figs).
We computed genetic diversity within populations (Heterozygosity, runs of homozygosity) and between populations (Weir and Cockerham’s estimator of FST, Wright’s FST), using plink v1.07, v1.9 [48, 49] and in-house scripts. A Mantel test was performed to calculate the correlation of genetic to linguistic and geographic distances (S25 Fig). Measurements of allelic richness, number of private alleles and uniquely shared alleles were computed using ADZE [50] on allelic and haplotype-based data. S27 Fig shows that the pattern is not driven by ascertainment bias.
Patterns of population structure was investigated using ADMIXTURE [51], CLUMPP (v. 1.1.2, [52] and distruct v. 1.1 [53]. Formal tests of admixture (f3 test, D-statistic) were performed using admixtools [39]. f3(Nuer,TSI;X) was used to estimate non-African admixture and f3(X,Mota;Ju|’hoansi) was used to estimate ancestral East African affinity. D-statistics were calculated as D(Ju|’hoansi,LBK; Mota, X).
The time in generations of admixture was calculated using a haploidized version of the data (see SI) with Malder [34] and Rolloff [39] and converted to calendar years assuming 30 years/generation. An ancient individual has shown widespread back admixture into East Africa [25] from Eurasia. To formally quantify the extend of the Eurasian admixture proportion we performed f4-ratios on dat2a, calculated as f4(CHB,GBR;X,LBK)/f4(CHB,GBR;Mota,LBK) similar to Gallego Llorente, Jones [25]. The ancient Ethiopian (Mota) [25] was used as an ancestral unadmixed (in terms of no Eurasian admixture) East African sample and the LBK individual [47] to substitute for an ancient Eurasian population.
|
10.1371/journal.pgen.1005491 | The Saccharomyces cerevisiae AMPK, Snf1, Negatively Regulates the Hog1 MAPK Pathway in ER Stress Response | Accumulation of unfolded proteins in the lumen of the endoplasmic reticulum (ER) causes ER stress. Snf1, the Saccharomyces cerevisiae ortholog of AMP–activated protein kinase (AMPK), plays a crucial role in the response to various environmental stresses. However, the role of Snf1 in ER stress response remains poorly understood. In this study, we characterize Snf1 as a negative regulator of Hog1 MAPK in ER stress response. The snf1 mutant cells showed the ER stress resistant phenotype. In contrast, Snf1-hyperactivated cells were sensitive to ER stress. Activated Hog1 levels were increased by snf1 mutation, although Snf1 hyperactivation interfered with Hog1 activation. Ssk1, a specific activator of MAPKKK functioning upstream of Hog1, was induced by ER stress, and its induction was inhibited in a manner dependent on Snf1 activity. Furthermore, we show that the SSK1 promoter is important not only for Snf1-modulated regulation of Ssk1 expression, but also for Ssk1 function in conferring ER stress tolerance. Our data suggest that Snf1 downregulates ER stress response signal mediated by Hog1 through negatively regulating expression of its specific activator Ssk1 at the transcriptional level. We also find that snf1 mutation upregulates the unfolded protein response (UPR) pathway, whereas Snf1 hyperactivation downregulates the UPR activity. Thus, Snf1 plays pleiotropic roles in ER stress response by negatively regulating the Hog1 MAPK pathway and the UPR pathway.
| All organisms are always exposed to several environmental stresses, including ultraviolet, heat, and chemical compounds. Therefore, every cell possesses defense mechanisms to maintain their survival under stressed conditions. Numerous studies have shown that a family of protein kinases plays a principal role in adaptive response to environmental stresses and perturbation of their regulation is implicated in a variety of human pathologies, such as cancer and neurodegenerative diseases. Elucidation of molecular mechanisms controlling their activities is still important not only for understanding how the organism acquires stress tolerance, but also for development of therapies for various diseases. In Saccharomyces cerevisiae, the Hog1 stress-responsive MAP kinase is activated by ER stress and coordinates a pleiotropic response to ER stress. However, the mechanisms for regulating Hog1 activity during ER stress response remain poorly understood. In this paper, we demonstrate that a Saccharomyces cerevisiae ortholog of mammalian AMP–activated protein kinase (AMPK), Snf1, negatively regulates Hog1 in ER stress response. ER stress induces expression of Ssk1, a specific activator of the Hog1 MAPK cascade. Snf1 lowers the expression level of Ssk1, thereby downregulating the signaling from upstream components to the Hog1 MAPK cascade. The activity of Snf1 is also enhanced by ER stress. Thus, our data suggest that Snf1 plays an important role in regulation of ER stress response signal mediated by Hog1.
| The endoplasmic reticulum (ER) is the cellular organelle responsible for the folding and modification of newly synthesized secretory or membrane proteins. Environmental or developmental changes which perturb ER homeostasis, or genetic alterations causing production of irreversibly misfolded proteins lead to an accumulation of unfolded and misfolded proteins within the ER. This condition, which is collectively termed ER stress, is toxic to cells and has been implicated in a variety of human pathologies, such as diabetes, cancer and neurodegeneration, including Alzheimer, Parkinson and Huntington disease [1, 2]. Therefore, when ER stress is sensed, cells actuate adaptive signaling pathways to alleviate ER stress [1, 3]. In the budding yeast Saccharomyces cerevisiae, the unfolded protein response (UPR) signaling pathway, composed of an ER transmembrane protein Ire1 and a transcriptional activator Hac1, plays a principal role in ER stress response [1, 3]. When activated by ER stress, Ire1 excises the translation-inhibitory intron from HAC1 mRNA, initiating splicing of HAC1 mRNA and consequent production of Hac1 protein. Hac1 induces expression of target genes, such as genes encoding chaperones and proteins functioning ER-associated degradation, thus increasing the protein folding capacity of the ER. Although the UPR is undoubtedly essential for yeast cells to alleviate ER stress, a previous genome-wide study [4] has predicted that not less than 100 genes are involved in response to ER stress. Therefore, it remains to be fully elucidated how ER stress response is precisely controlled.
AMPK is evolutionarily conserved in eukaryotic cells and a key sensor of cellular energy status [5–7]. In Saccharomyces cerevisiae, a catalytic subunit of AMPK is encoded by the SNF1 gene (S1 Fig). Similar to other members of the AMPK family, Snf1 forms a heterotrimeric complex with two regulatory subunits, the γ subunit Snf4 and one of the three alternative β subunits, Sip1, Sip2, or Gal83 [5]. The catalytic activity of Snf1 is regulated by phosphorylation at Thr-210 that is located in the activation loop of its kinase domain [8, 9]. Three upstream kinases, Sak1, Tos3, and Elm1, have been identified as kinases responsible for Snf1 activation [10–12]. Oppositely, Snf1 is inactivated by the Reg1-Glc7 protein phosphatase 1 complex; the catalytic subunit Glc7 is directed to Snf1 through the regulatory subunit Reg1 [13, 14]. Besides critical roles in adaptation to glucose deprivation and utilization of alternative carbon sources to glucose, the Snf1 complex is involved in the response to environmental stresses, such as heat and oxidative stresses [5, 15]. However, the role of Snf1 in ER stress response is as yet poorly understood.
The budding yeast Hog1, which is structurally highly similar to the mammalian p38 MAPK, was originally identified as a key protein kinase required for the adaptation of yeast cells to osmotic stress [16, 17]. In osmotic stress response, the Sln1-Ypd1-Ssk1 multistep phosphorelay system, which is homologous to bacterial two-component systems, regulates the Hog1 MAPK cascade (S2 Fig) [16, 17]. Under normal osmotic conditions, the membrane-associated histidine kinase Sln1 phosphorylates itself. The phosphate group is transferred to Ssk1 through the Ypd1 phosphotransmitter. Hyperosmotic stress inactivates Sln1, resulting in downregulation of the phosphorylation level of Ssk1. Dephosphorylated Ssk1 directly binds to and activates the Ssk2 and Ssk22 MAPKKKs, and consequently, leads to sequential activation of Pbs2 MAPKK and Hog1 MAPK. In addition to a pivotal role in osmotic stress response, Hog1 has been revealed as a regulator of a wide array of stress responses, including cold, heat and ER stresses [16–19]. In ER stress response, Hog1 is activated in an Ssk1-dependent manner [18]. However, the mechanisms that control Hog1 activity in ER stress response are still poorly understood.
In this study, we identified Snf1 as a negative regulator of Hog1 in ER stress response. Cells lacking Snf1 have elevated levels of active Hog1, whereas upregulation of Snf1 activity reduces Hog1 activation. ER stress induces expression of Ssk1, but this induction is counteracted by Snf1. These results indicate that Snf1 modulates Hog1 activation by controlling the expression level of its activator Ssk1. We also demonstrated that loss of Snf1 leads to upregulation of the UPR pathway, whereas the UPR activity is downregulated in Snf1-activated cells. Thus, Snf1 negatively regulates the Hog1 MAPK pathway and the UPR pathway in ER stress response.
In order to test whether the Snf1 protein kinase regulates ER stress response, cells carrying snf1 deletion were plated on medium containing tunicamycin, a natural inhibitor of N-linked glycosylation that is widely employed as an inducer of ER stress. We unexpectedly found that compared to wild-type cells, the snf1 mutant was resistant to tunicamycin (Fig 1A). To confirm that snf1 mutation caused tunicamycin resistance, we transformed the snf1 mutant with the plasmid that expresses SNF1 and tested the transformants for growth on medium containing tunicamycin. Expression of SNF1 significantly rescued the tunicamycin-resistant phenotype associated with the snf1 mutation (Fig 1A and S3A Fig). To address the biological importance of Snf1 kinase activity, we generated a catalytically inactive form of Snf1 [Snf1(K84M)], in which Lys-84 in the ATP-binding motif was mutated to methionine [8]. When Snf1(K84M) was expressed in snf1 mutants, the tunicamycin resistance was not rescued (Fig 1A). To examine the effect of Snf1 hyperactivation on ER stress response, we generated Snf1(G53R), a catalytically active form in which Gly-53 in the kinase domain has been mutated to arginine [8]. Expression of Snf1(G53R) resulted in hypersensitivity to tunicamycin (Fig 1B). These results indicate that Snf1 negatively regulates the response to ER stress in a manner dependent on its kinase activity.
We next asked if the regulatory subunits of Snf1 are involved in ER stress response. We found that deletion of the SNF4 gene, which encodes the γ subunit of the Snf1 complex, caused increased resistance to tunicamycin (Fig 1C). Furthermore, cells harboring both snf1 and snf4 mutations displayed ER stress tolerance indistinguishable from that observed in the snf1 single mutants (Fig 1C), indicating that Snf1 and Snf4 act in the same pathway. We next examined whether the β subunits, Sip1, Sip2, and Gal83, regulate ER stress response. We found that the sip1 sip2 gal83 triple mutant was resistant to tunicamycin, although neither of their single mutants exhibited the obvious tunicamycin-resistant phenotype (Fig 1D). These results indicate that the Snf1 complex negatively regulates ER stress response.
Snf1 is phosphorylated at Thr-210 and activated by exposure of cells to alkaline pH and oxidative stresses [15]. Therefore, we investigated whether Snf1 is activated by treatment with ER stress. Anti-phospho AMPK antibodies that recognize the phosphorylated, activated form of AMPK were used to monitor phosphorylation of Snf1 at Thr-210 in the budding yeast [12]. In wild-type cells, we could detect Snf1 phosphorylation under unstressed conditions (Fig 2A). Treatment of cells with tunicamycin stimulated Snf1 phosphorylation within 1.5–3 hr, and its phosphorylation was persisted for at least 7.5 hr (Fig 2A). Similar observation was seen when cells were exposed to dithiothreitol (DTT), which causes ER stress by blocking disulfide bond formation in the ER (Fig 2B). Thus, ER stress induces activation of Snf1 through phosphorylation at the Thr-210 residue. We next asked whether ER stress-induced Snf1 activation is mediated by the upstream kinases, Sak1, Tos3, and Elm1. We found that sak1 tos3 elm1 triple mutations completely abolished activation of Snf1 both in the presence or absence of ER stress (Fig 2C), although activated Snf1 levels were only slightly decreased in each single mutants (S4A and S4B Fig). This result indicates that ER stress induces Snf1 activation in a manner dependent on the three redundant kinases. Snf1 is inactivated through dephosphorylation mediated by the Reg1-Glc7 phosphatase complex. We next investigated the role of the Reg1-Glc7 complex in ER stress-induced Snf1 activation. Because the glc7 deletion strain is lethal [20, 21], we used reg1 deletion. Both in the presence or absence of tunicamycin treatment, phosphorylated Snf1 levels were clearly upregulated by reg1 deletion (Fig 2D), indicating that the Reg1-Glc7 protein phosphatase 1 acts to inactivate Snf1 in ER stress response.
We next asked whether phosphorylation of Snf1 at Thr-210 is important for its function in ER stress response. We first examined the ability of Snf1(T210A), which contains a mutation of Thr-210 to Ala, to complement ER stress resistance associated with snf1 deletion. The Snf1(T210A) mutant failed to complement the snf1 defect (Fig 2E). We investigated whether the ER stress response involves the three upstream kinases and the phosphatase complex. We found that the sak1 tos3 elm1 triple mutant cells, which were defective in Snf1 activation, were resistant to tunicamycin (Fig 2F). In contrast, reg1 mutant cells in which Snf1 activity is upregulated exhibited hypersensitivity to tunicamycin; however, the stress sensitivity of the reg1 mutant was completely suppressed by snf1 deletion (Fig 2G). Similar results were obtained when DTT was used as an ER stressor (Fig 2H). Taken together, these results demonstrate that Snf1 is activated by ER stress through phosphorylation at Thr-210 and then negatively regulates ER stress response.
Next, we explored the mechanism underlying the effect of Snf1 on ER stress response. Previous analyses in Saccharomyces cerevisiae have revealed that the UPR, composed of Ire1 and Hac1, is at the center of ER stress response [1, 3]. Therefore, we investigated a potential role for Snf1 in regulating the UPR. Upon ER stress, activated Ire1 excises the translation-inhibitory intron from HAC1 mRNA, consequently producing Hac1 protein. Hac1 transcriptionally activates its target genes, including KAR2 and ERO1. We first examined the kinetics of HAC1 mRNA splicing (Fig 3A). In wild-type cells under unstressed conditions, the unspliced form of HAC1 mRNA (HAC1u) was robustly detected, but the spliced form (HAC1s) was rarely detectable. Treatment of cells with ER stress promoted splicing of HAC1 mRNA. The amount of HAC1s peaked 1.5 to 3 hr after DTT addition and gradually decreased thereafter. We next investigated the role of Snf1 in regulation of HAC1 mRNA splicing using the snf1 and reg1 mutant cells. We found that both in snf1 and reg1 mutant cells, HAC1 mRNA splicing was normally promoted in response to ER stress (Fig 3A). We found that downregulation of HAC1 mRNA splicing was unaffected by snf1 mutation. On the other hand, in the reg1 mutant cells, HAC1s was decreased rapidly within 3 hr of DTT addition. Therefore, we compared the protein level of Hac1 between wild-type and the reg1 mutant cells (Fig 3B). In wild-type cells, Hac1 protein was hardly detectable prior to ER stress treatment. Production of Hac1 was induced within 1.5 hr after exposure to DTT and subsequently downregulated. In the reg1 mutant cells, Hac1 production was induced at levels comparable to that of wild-type cells. However, the amount of Hac1 declined rapidly within 3 to 4.5 hr after DTT treatment. A rapid decrease in Hac1 protein was also seen when cells harboring reg1 mutation were exposed to tunicamycin (S5A Fig). Consistent with the protein level of Hac1, expression of the well-known Hac1 target genes, ERO1 and KAR2, was reduced by reg1 mutation (Fig 3E and S5B Fig). These UPR defects observed in the reg1 mutant could be significantly restored by snf1 mutation (Figs 3A, 3C and 3E and S5B). These results suggest that Snf1 participates in downregulation of the UPR pathway.
The observation that the UPR activity was downregulated by reg1 mutation prompted us to perform a detailed analysis of the UPR activity in the snf1 mutant cells. We found that the UPR activity under unstressed conditions was increased in the snf1 mutant cells. In the absence of ER stress, the level of HAC1s in cells harboring snf1 deletion was statistically higher than that in wild-type cells (Fig 3A). Consistent with this, the snf1 mutant cells expressed a small amount of Hac1 protein even prior to treatment with ER stress (Fig 3D). Furthermore, under unstressed conditions, both ERO1 and KAR2 mRNAs were statistically significantly increased in the snf1 mutant compared to wild-type cells (Fig 3E and S5B Fig). Taken together, these results indicate that Snf1 negatively regulates the UPR pathway.
The snf1 mutation significantly enhanced resistance against ER stress, although UPR upregulation caused by snf1 mutation was detected only under unstressed conditions. Therefore, additional mechanisms may contribute to ER stress resistance caused by snf1 mutation. In the budding yeast, Hog1 MAPK is activated by ER stress through phosphorylation at critical threonine and tyrosine residues located in the activation loop [16, 17], and is in fact required for protecting cells against ER stress [18, 19]. Anti-phospho-p38 antibodies that recognize the phosphorylated form of mammalian p38 MAPK can be used to detect activated Hog1 in the budding yeast [22]. As observed previously [18], western blot analysis with anti-phospho-p38 antibody marginally detected the activated form of Hog1 in wild-type cells and its abundance was increased by treatment of cells with DTT (Fig 4A). To investigate the role of Snf1 in regulation of Hog1 activity, we monitored the activated form of Hog1 in the snf1 mutant following induction of ER stress. We found that activated Hog1 levels were increased in snf1 mutant cells both in the presence or absence of DTT treatment (Fig 4A). Similar results were obtained when cells were exposed to tunicamycin (S6A Fig). These results suggest that Snf1 has the inhibitory effect on Hog1 activation. The observation that Snf1 is activated by ER stress prompted us to test whether Snf1 acts to downregulate Hog1 activity during recovery from ER stress. We observed that Snf1 remained active even after removal of DTT from the medium (Fig 4C). In contrast, DTT removal allowed reduction of Hog1 activity (Fig 4B). However, Hog1 activation was prolonged in cells lacking Snf1 (Fig 4B). These results suggest that ER stress-activated Snf1 participates in the process that Hog1 activity returns to the basal level. Furthermore, it is suggested that additional mechanisms function in Hog1 inactivation during recovery from ER stress, since Hog1 dephosphorylation after DTT removal was delayed, but occurred in snf1 mutant cells. Next, we examined the effect of Snf1 hyperactivation on ER stress-induced Hog1 activation. Strikingly, activation of Hog1 in response to DTT was diminished by reg1 mutation which increases Snf1 activity (Fig 4D), and this defect could be entirely restored by snf1 mutation (Fig 4E). Similar results were obtained when cells were exposed to tunicamycin (Fig 4F and S6B Fig). Our finding that reg1 mutation interferes with Hog1 activation in response to ER stress strongly suggests that Snf1 acts as a negative regulator of Hog1 in ER stress response.
Previous report showed that ER stress-activated Hog1 accumulated in the nucleus [18]. To investigate the effect of Snf1 on nuclear accumulation of Hog1 in response to ER stress, we used the strain which expresses Hog1 carboxyl-terminally tagged with GFP (Fig 4G). As observed previously [18], Hog1 was uniformly distributed in the nucleus and cytosol under normal conditions and became enriched in the nucleus after ER stress treatment. Loss of Snf1 slightly but significantly increased nuclear localization of Hog1 even in the absence of ER stress. In contrast, nuclear accumulation of Hog1 in response to ER stress was obviously decreased in the reg1 mutant cells; however, this defect was clearly suppressed by snf1 deletion. These observations support a role of Snf1 in negative regulation of Hog1 in ER stress response.
It has been well-characterized that yeast cells activate Hog1 when exposed to hyperosmotic extracellular environments [16, 17]. We therefore examines whether Snf1 might be involved in the osmotic stress response mediated by Hog1. In wild-type cells, activated Hog1 is robustly detectable within 3 min of NaCl treatment and then rapidly decreases by 30 min (Fig 4H). Hog1 activation in response to hyperosmotic stress appeared to be enhanced and reduced by snf1 and reg1 mutations, respectively (Fig 4H). These alterations are probably attributed to a potential role of Snf1 in inhibiting Hog1 activation. Indeed, snf1 mutation elevated the basal activity of Hog1 (Fig 4A), and reg1 mutation partially suppressed the lethality of SLN1- and YPD1-deleted cells in which Hog1 is constitutively activated (see below). However, we could not find that reg1 mutation resulted in hypersensitivity to osmotic stress (S6C Fig). Therefore, it remains obscure whether Snf1-mediated Hog1 regulation is physiologically important for osmotic stress response.
We next examined whether enhanced ER stress resistance in the snf1 mutants is caused by Hog1 hyperactivation. We constructed the snf1 hog1 double mutants and test them for growth on medium containing tunicamycin (Fig 5A). The snf1 hog1 double mutant was sensitive to tunicamycin. However, we also found that ER stress sensitive phenotype of the hog1 mutant could be partially suppressed by snf1 mutation. As snf1 mutation leads to upregulation of the UPR, we compared the effect of snf1 deletion in cells having a wild-type, hac1, or hac1 hog1 background. The snf1 mutation modestly restored ER stress sensitivity caused by hac1 mutation (Fig 5B). In contrast, the snf1 hog1 hac1 triple mutants exhibited hypersensitivity to tunicamycin, similar to the hog1 hac1 double mutants (Fig 5C), indicating that Hog1 and UPR are key targets of Snf1 in ER stress response.
As shown above, activities of the UPR and Hog1 pathways are upregulated by snf1 deletion, but downregulated by reg1 mutation which leads to Snf1 hyperactivation. These observations raised the possibility that Snf1 continuously regulates the UPR and Hog1 pathways. If this is true, we can observe the diminished Hog1 activation in hac1 mutant cells or the reduced UPR activity in hog1 mutant cells. First, we measured Hog1 activity in cells lacking Hac1. However, we could not find that loss of Hac1 reduced Hog1 activation (Fig 5D). We next monitored HAC1 mRNA splicing in hog1 mutant cells. In hog1 mutant cells, HAC1 mRNA splicing was normally induced, but retained longer than wild-type cells (Fig 5E). This is consistent with a previous observation [18] and indicates that hog1 mutation does not reduce, but rather upregulates the UPR activity. Thus, the activities of the UPR and Hog1 pathways are independently regulated by Snf1.
We next investigated how Snf1 negatively regulates Hog1 in ER stress response. The dephosphorylation of MAPK by protein phosphatases is well-known as a common mechanism for the negative regulation of the signaling mediated by MAPK [23]. Hog1 is dephosphorylated and inactivated by Ptp2 tyrosine phosphatase [24, 25]. Previously, it has been shown that loss of Ptp2 results in enhanced resistance to ER stress in a HOG1-dependent manner [19]. Therefore, we examined the relationship between Snf1 and Ptp2. In the ptp2 mutant cells, basal activity of Hog1 was modestly increased and ER stress-induced Hog1 activation was significantly upregulated (S7A Fig). We found that Hog1 activation was enhanced in the ptp2 snf1 double mutants compared with the ptp2 mutant cells (S7A Fig), indicating that Snf1 negatively regulates Hog1 in ER stress response independently of Ptp2.
In ER stress response, signaling through the Hog1 pathway is controlled by the Sln1-Ypd1-Ssk1 phosphorelay system [18, 19]. Disruption of the SLN1 gene results in lethality due to constitutive activation of Hog1 and, indeed, mutations in any of the four downstream genes, SSK1, SSK2, PBS2, and HOG1, suppress the sln1 lethality by blocking activation of Hog1 [26, 27]. As shown above, Hog1 activity is considerably decreased in reg1 mutant cells in which Snf1 is hyperactivated. Therefore, we tested whether deletion of the REG1 gene suppresses the sln1 lethality. We found that reg1 mutation modestly suppressed the growth defect associated with sln1 deletion (Fig 6A). Similarly, the lethality caused by ypd1 deletion was partially suppressed by reg1 mutation (Fig 6B). However, loss of Snf1 interfered with the ability of reg1 mutation to suppress the ypd1 lethality (S8A and S8B Fig). These results suggest that Snf1 regulates the component functioning downstream of Ypd1 in the Hog1 pathway.
In order to identify the molecule that mediates the signaling from Snf1 to Hog1, we examined the expression levels of components that act in the Hog1 pathway. We generated yeast strains carrying the carboxyl-terminally Myc-tagged genes, including SSK1, SSK2, SSK22, and PBS2, and analyzed their expression levels (Figs 6C–6E and S8C–S8E). Among them, we found that Ssk1 expression is changed by treatment with ER stress and genetic modulation of Snf1 signaling. In wild-type cells, the protein abundance of Ssk1 is increased following exposure to DTT and tunicamycin, but not NaCl (Fig 6C–6F), suggesting that ER stress specifically affects Ssk1 expression. The snf1 mutation moderately increased Ssk1 expression (Fig 6D), suggesting that Ssk1 expression is negatively regulated by Snf1. Next, we examined the effect of Snf1 hyperactivation on the expression level of Ssk1. ER stress-mediated Ssk1 induction was effectively inhibited by reg1 mutation that leads to hyperactivation of Snf1 (Fig 6E). This defect could be significantly restored by snf1 mutation (Fig 6E). Similar results were obtained when cells were exposed to tunicamycin (Fig 6F). These results suggest that Ssk1 expression is negatively regulated by Snf1.
We next examined the functional importance of Ssk1 in Snf1-mediated regulation of Hog1 activity. As shown previously [18], activated Hog1 levels were significantly decreased in ssk1 mutant cells (Fig 6G). This defect could not be suppressed by snf1 mutation (Fig 6G), indicating that Ssk1 is important for Hog1 hyperactivation caused by snf1 mutation. We also asked whether Ssk1 is involved in enhanced ER stress resistance of the snf1 mutants. We found that ssk1 mutation rendered cells lacking Snf1 sensitive to tunicamycin (Fig 6H). We also observed that the ssk1 snf1 double mutant cells were more resistant to ER stress than the ssk1 single mutants. ER stress tolerance of the ssk1 snf1 double mutants is probably due to increased UPR activity caused by snf1 mutation. Taken together, these results indicate that Snf1 inhibits Hog1 activation in response to ER stress by negatively regulating the expression level of Ssk1.
To explore the mechanism by which Snf1 regulates the expression level of Ssk1, we measured the amount of SSK1 mRNA by qRT-PCR (Fig 7A and S9A Fig). In wild-type cells, SSK1 mRNA is increased following exposure to ER stress. This induction seemed to be normal in snf1 mutant cells. On the other hand, reg1 deletion significantly inhibited the induction of SSK1 mRNA. This reg1 defect could be restored by snf1 mutation. These results indicate that Snf1 hyperactivation reduce the expression level of SSK1 mRNA.
Numerous studies have revealed that Snf1 regulates the gene expression at the transcriptional level through phosphorylation of transcription factors [5, 6]. This raised the possibility that Snf1 regulates SSK1 promoter activity. To test this possibility, we generated a PSSK1-GFP reporter, consisting of the 5' upstream region of the SSK1 gene to drive GFP expression (Fig 7B). Wild-type cells harboring the PSSK1-GFP reporter displayed GFP expression in the absence of ER stress (Fig 7C). GFP expression from the PSSK1-GFP reporter was increased following incubation with DTT (Fig 7C). On the other hand, we observed that DTT treatment had no obvious effect on expression of GFP derived from the PMCM2-GFP reporter, in which the 5' upstream region of the MCM2 gene is fused to GFP (Figs 7B and 7C and S9B). These results suggest that SSK1 promoter is activated by ER stress. Next, we tested whether PSSK1-GFP induction is regulated by the Snf1 pathway. In contrast to wild-type cells, PSSK1-GFP expression was barely induced by DTT in reg1 mutant cells (Fig 7D). This reg1 defect could be significantly restored by snf1 mutation (Fig 7D). These results strongly support the model in which Snf1 inhibits the activity of SSK1 promoter.
We next examined whether SSK1 induction in response to ER stress is important for resistance to ER stress. To test this, we generated two constructs, PSSK1-SSK1 and PMCM2-SSK1, which express SSK1 under the control of SSK1 and MCM2 promoters, respectively. Introduction of the PSSK1-SSK1 construct significantly rescued ER stress sensitive phenotype associated with ssk1 mutation (Fig 7E). On the other hand, when the PMCM2-SSK1 construct was introduced into ssk1 mutant cells, ER stress sensitivity was less effectively rescued (Fig 7E). This suggests that SSK1 induction via its promoter activation is important for protecting cells against ER stress.
We also attempted to compare the ability of the PSSK1-SSK1 and PMCM2-SSK1 constructs to rescue the osmotic stress sensitivity caused by ssk1 mutation. In osmotic stress response, the Hog1 pathway is activated by the membrane protein Sho1 in addition to Ssk1 [26]; hence, as shown in S9C Fig, the ssk1 sho1 double mutants was sensitive to osmotic stress, although neither of their single mutants exhibited the obvious sensitivity to osmotic stress. Therefore, we transformed the ssk1 sho1 double mutant cells with the PSSK1-SSK1 and PMCM2-SSK1 constructs and tested the transformants for growth under hyperosmotic conditions. We found that the PMCM2-SSK1 construct could rescue the osmotic stress sensitivity caused by ssk1 sho1 mutations to same extent as the PSSK1-SSK1 construct (Fig 7F). Thus, it is unlikely that the SSK1 promoter is involved in regulation of osmotic stress response. Taken together, the mechanisms underlying Hog1 activation mediated by Ssk1 are different between osmotic and ER stresses.
Previous studies have revealed that the snf1 mutant cells exhibited hypersensitivity to a number of environmental stresses, including alkaline pH, heat shock, and genotoxic stress caused by hydroxyurea and methylmethane sulfonate [5]. Therefore, Snf1 was regarded as an essential regulator to confer resistance to environmental stresses. In this study, we unexpectedly found that the snf1 mutants were resistant to ER stress. This finding not only indicates that Snf1 negatively regulates ER stress response, but also reveal a novel inhibitory role for Snf1 in stress response.
Here, we revealed that Snf1 inhibits Hog1 activity by downregulation of the expression level of SSK1 mRNA encoding an upstream activator of the Hog1 MAPK cascade. It is well-known that the dephosphorylation of MAPK by protein phosphatases is crucial for the negative regulation of the signaling mediated by MAPK [23]. The protein phosphatases, such as Ptc1, Ptp2, and Ptp3, play an important role in Hog1 inactivation [16, 17, 24, 25]. Previous report showed that Ptp2 and Ptp3 play the main and minor role, respectively, in negative regulation of Hog1 during ER stress response [19]. Nevertheless, why is Snf1 needed to function in downregulation of Hog1 activity? It is possible that Snf1 coordinates ER stress response with other cellular responses, since Snf1 is activated in the response to various environmental stresses [5, 6, 15]. Indeed, it has been reported that ER stress sensitivity is enhanced under the extracellular environments in which Snf1 activity is known to be elevated [28]. Alternatively, Snf1 may function to inactivate Hog1 in the whole cell level. Previous study showed that upon exposure to ER stress, Hog1 not only translocates into the nucleus and regulates the gene expression, but also functions in activation of autophagy in the cytoplasm [18]. On the other hand, Ptp2 and Ptp3 phosphatases are localized in the nucleus and cytoplasm, respectively [29]. Therefore, it is anticipated that Hog1 activity and its related cellular responses are negatively regulated in a manner different from the nucleus and cytoplasm. In contrast, Snf1 interferes with the signal from Ssk1 to Hog1 MAPK cascade through negative regulation of Ssk1 expression. Therefore, Snf1 is expected to contribute to downregulation of Hog1 in the whole cell level. Our analyses showed that loss of Snf1 moderately increased Hog1 activity, while Snf1 hyperactivation caused by reg1 deletion effectively inhibited Hog1 activation. The existence of protein phosphatases for Hog1 may make apparently difficult to observe the effect of snf1 mutation on Hog1 activity. It is well-known that expression of protein phosphatases for MAPK is induced by environmental stresses [23]. In fact, we found that Ptp2 was induced in response to ER stress (S7C Fig). Consistent with this, Hog1 inactivation after removal of ER stress was modestly delayed, but occurred in snf1 mutant cells. Thus, it is likely that intricate signaling networks regulate Hog1 activity during ER stress response. Therefore, detailed analyses of the relationships between Hog1-mediated ER stress responses and the function of each negative regulator for Hog1 will be important for further understanding how Hog1 activity is controlled during ER stress response.
In this study, we observed that the snf1 mutant cells expressed Hac1 even in the absence of ER stress. On the other hand, the expression level of Hac1 in the presence of ER stress was rapidly decreased in Snf1-hyperactivated cells. These observations suggest that Snf1 acts as a negative regulator of the UPR pathway. Although the activation mechanisms of the UPR pathway has been well-studied, there are only a few reports about how the UPR is inactivated after ER stress. Previously, two groups demonstrated the importance of the phosphorylation state of Ire1 kinase domain in the attenuation of the UPR activity [30, 31]. However, their proposed models are significantly different from each other. Therefore, the mechanisms by which the UPR is finally attenuated have yet to be elucidated. In the course of preparation of this manuscript, Casamayor and colleagues also reported that Snf1 is involved in yeast ER stress response [28]. Consistent with our findings, they showed that reg1 mutation results in increased sensitivity to ER stress in a Snf1-dependent manner. Interestingly, they proposed the model in which Snf1 plays an inhibitory role in attenuation of the UPR by regulating the oligomerization of Ire1. In regard to the UPR activity in the reg1 mutant cells, their results are distinctly different from our observations: they showed that increased UPR activity after ER stress treatment was prolonged in the reg1 mutant cells; we found that in the reg1 mutant cells, the UPR activity declined rapidly during ER stress response. The reason for this discrepancy is not clear now. However, this phenotypic distinction may be attributed to the difference in genetic background: their strains were derivatives of BY4741 and DBY746; we used W303 derivatives. Indeed, we could observe that the snf1 mutant was resistant to tunicamycin, although they found no difference in ER stress sensitivity between wild-type and the snf1 mutant cells. Thus, further analyses should be needed to elucidate the molecular mechanism by which Snf1 regulates the UPR signaling pathway.
We found here that snf1 mutation increases the activities of the Hog1 and UPR pathways and leads to resistance to ER stress. Numerous studies have revealed that improper hyperactivation of stress-responsive signaling pathways is deleterious to cells and organisms [16, 23, 32]. In fact, constitutive activation of Hog1 under unstressed conditions causes cell lethality [26, 33]. We observed that cells deleted for both SNF1 and PTP2 genes showed a high basal activity of Hog1, but remains viable (S7B Fig). Thus, Hog1 activity in the ptp2 snf1 double mutant cells is not high enough to induce lethality. We found that compared with wild-type cells, the snf1 mutant cells possessed an increased Hog1 activity during ER stress response. However, it is noteworthy that a decline in the Hog1 activity occurred after removal of ER stress even in cells lacking Snf1. This indicates that the Hog1 activity in the snf1 mutant cells is upregulated, but still remains under the control of the regulatory mechanisms. Therefore, it is possible that Hog1 upregulation caused by snf1 mutation is preferable for yeast cells to survive in the presence of ER stress. Similar view may be applied to the UPR activity. Previous studies revealed that perturbation of the mechanism for properly attenuating Ire1 activity results in reduction of cell viability in the presence of ER stress [30, 31]. In the snf1 mutant cells, the basal activity of UPR is significantly higher than wild-type cells; however, attenuation of the UPR was nearly unaffected by snf1 mutation. Since the snf1 mutant cells possesses high, but adjustable, UPR activity, loss of Snf1 may be preferable for cells to survive under ER stress. Thus, Snf1 plays pleiotropic roles in negative regulation of ER stress response.
The signaling through the Hog1 pathway is controlled by the Sln1-Ypd1-Ssk1 phosphorelay system [16, 17]. In osmotic stress response, Sln1 inactivation is a key step of Hog1 activation. Under normal osmotic conditions, active Sln1 leads to Ssk1 phosphorylation. Hyperosmotic stress inactivates Sln1, causing an increase of the dephosphorylated form of Ssk1. This promotes the Ssk1-Ssk2/Ssk22 physical interaction and results in activation of the Hog1 MAPK cascade. Previous studies demonstrated that Ssk1 is implicated in regulation of Hog1 activity during ER stress response [18, 19]. Little is understood, however, about the mechanism by which ER stress activates Hog1. In this study, we demonstrated that the expression level of Ssk1 is increased during ER stress response, and that elevation of Ssk1 protein level is important for cells to survive under ER stress conditions. In contrast, Ssk1 expression remained unchanged upon osmotic stress. These findings suggest that the mechanisms for the regulation of Hog1 are different among these different types of stress. Based on our data, we propose the model in which Hog1 activation in response to ER stress involves upregulation of Ssk1 (Fig 8). In unstressed conditions, Ssk1 is phosphorylated and inactivated by the upstream phosphorelay system. In the presence of ER stress, increased expression of Ssk1 overwhelms the phosphorylation activity of upstream phosphorelay system, leading to accumulation of dephosphorylated Ssk1 and consequent activation of the Hog1 MAPK cascade.
It is well-known that many protein kinases of the MAPKKK family can be activated by binding of their activators, similar to the budding yeast Ssk2 and Ssk22 MAPKKKs [23]. In unstimulated cells, MAPKKK is kept catalytically inactive through an autoinhibitory interaction between the regulatory domain and the kinase domain. Upon stimulation, binding of an activator protein leads to the dissociation of the autoinhibitory domain from the kinase domain and a consequent activation of MAPKKK. Mammalian MTK1/MEKK4 is a stress responsive MAPKKK that is structurally highly similar to the yeast Ssk2/Ssk22 and locates upstream in the p38 pathway [34, 35]. Previous studies have identified the GADD45 family proteins (GADD45α, GADD45β and GADD45γ) as MTK1/MEKK4 activators [36]. GADD45 binds to the autoinhibitory domain of MTK1/MEKK4 and relieve autoinhibition. Furthermore, each GADD45 exhibits distinct tissue expression patterns and is induced by a certain subset of environmental stresses, showing functional distinction among the GADD45 isoforms in p38 activation [36]. Our data presented here suggest that ER stress activates the Hog1 MAPK cascade by induction of Ssk1, in a manner similar to activation of p38 MAPK cascade though stress-mediated induction of GADD45.
To date, there is little understanding of the mechanism that controls the expression level of Ssk1. In this study, we show that Snf1 acts as a negative regulator of SSK1 expression in ER stress response. We also demonstrate that the SSK1 promoter is important for Snf1 to negatively regulate the mRNA level of SSK1. Snf1 phosphorylates a large number of transcription factors, and influences the transcription of hundreds of genes, including those involved in the utilization of alternate carbon sources and the metabolism of amino acids [5, 6]. Therefore, Snf1 is the most likely to inhibit the promoter activity of the SSK1 gene through phosphorylation of the transcription factor. We also found that the expression level of Ssk1 protein was slightly different from that of SSK1 mRNA. Ssk1 protein is more abundant in the snf1 mutants than wild-type cells, although there was little difference in the expression level of SSK1 mRNA between wild-type and the snf1 mutant cells. This suggests that Snf1 inhibits Ssk1 expression at the translational or posttranslational levels. Intriguingly, it has been showed that the protein level of Ssk1 is negatively regulated by Ubc7 [37]. Ubc7 is an endoplasmic reticulum-associated ubiquitin-conjugating enzyme responsible for ER-associated degradation (ERAD). Snf1 may modulate Ssk1 degradation mediated by the ubiquitin-proteasome system involving Ubc7. Thus, identification of components downstream of Snf1, for example, which is responsible for induction of Ssk1 in response to ER stress, will provide valuable insights into the evolutionally conserved mechanism for regulation of the p38/Hog1 MAPK cascade.
Strains used in this study are listed in S1 Table. Yeast strains harboring the complete gene deletions and carboxyl-terminally Myc or GFP-tagged genes were generated by a PCR-based method as described previously [38]. All strains constructed by a PCR-based method were verified by PCR to confirm that replacement had occurred at the expected locus. Standard procedures were followed for yeast manipulations [39].
Plasmids used in this study are described in S2 Table. In-Fusion cloning kits (Takara) was used to construct plasmids. The PSSK1-GFP and PMCM2-GFP plasmids were constructed as follows. The DNA fragment encoding GFP followed by the ADH1 terminator (TADH1) was obtained by PCR using the pFA6a-GFP vector [38] as a template. The GFP-TADH1 DNA fragment was fused to 999-bp and 762-bp genomic fragments containing 5' upstream sequences of the SSK1 and MCM2 genes, yielding the PSSK1-GFP and PMCM2-GFP plasmids, respectively. Schemes detailing construction of plasmids and primer sequences are available on request.
Protein extracts were prepared essentially as described previously [40]. Briefly, cells grown to exponential phase were incubated with YPD or SD medium containing 2 μg/ml tunicamycin, 4 mM DTT or 0.4 M sodium chloride, for the indicated times. Cells were transferred into test tubes, mixed 1:1 with boiled medium, submerged in the boiling water for 3 min, and harvested by centrifugation. Cells were then subjected to a mild alkali treatment-based protein extraction method [41]. Western blot analysis was performed using the immunoreaction enhancer solution Can Get Signal (Toyobo) according to the manufacturer's protocol. Anti-HA monoclonal antibody 16B12 (Covance), anti-Myc monoclonal antibody 9E10 (Santa Cruz), anti-GFP monoclonal antibody JL-8 (Clontech), anti-phospho-p38 MAPK monoclonal antibody 28B10 (Cell Signaling), anti-phospho-AMPKα monoclonal antibody 40H9 (Cell Signaling), anti-Hog1 polyclonal antibody y-215 (Santa Cruz), anti-Snf1 polyclonal antibody yk-16 (Santa Cruz), and anti-Mcm2 polyclonal antibody N-19 (Santa Cruz) were used. Detection was carried out by using a LAS-4000 (Fuji Film) with Immobilon Westren (Merck Millipore). Signal intensities were quantified by ImageQuant (GE Healthcare), and statistical analysis was performed with Excel (Microsoft).
Cells grown to exponential phase were incubated with YPD medium containing 2 μg/ml tunicamycin or 4 mM DTT, and harvested at the indicated times. Total RNA was then prepared using ISOGEN reagent (Nippon Gene) and the RNeasy Mini kit (Qiagen). First strands of cDNA were generated using the PrimeScript RT reagent Kit (Takara). The HAC1 cDNA was amplified from first strands of cDNA with Blend Taq (TOYOBO), and then analyzed by agarose gel electrophoresis. Detection, quantification, and statistical analysis was carried out by using a LAS-4000 (Fuji Film), ImageQuant (GE Healthcare), and Excel (Microsoft), respectively. The cDNAs of ERO1, KAR2, and SSK1, were quantitated by a quantitative real-time RT-PCR (qRT-PCR) method using a 7500 fast real-time RT-PCR system (Applied Biosystems) with SYBR Premix Ex Taq (Takara). A standard curve was generated from diluted RNA derived from wild-type cells, and levels of gene expression were normalized to ACT1 expression. HAC1 primers (CTGGCTGACCACGAAGACGC and TTGTCTTCATGAAGTGATGA) were used to monitor splicing of HAC1 mRNA. ERO1 primers (TAACAGCAAATCCGGAACG and ACCAAATTTGACCAGCTTGC), KAR2 primers (AGACTAAGCGCTGGCAAGCT and ACCACGAAAAGGGCGTACAG), SSK1 primers (AGCTGGAAGCAGGGAGAAAG and TGAGTGAGGGTTGGAAGGTG), and ACT1 primers (TGCCGAAAGAATGCAAAAGG and TCTGGAGGAGCAATGATCTTGA) were used to analyze the mRNA level of ERO1, KAR2, and SSK1.
Assays for tunicamycin toxicity were carried out as follows. Cells were grown to exponential phase, and cultures were adjusted to an optical density of 0.5. Cell cultures were then serially diluted 5-fold, spotted onto normal plates or plates containing the indicated concentrations of tunicamycin, followed by incubation at 25°C for 3 days (for plates lacking or containing 0.1 μg/ml tunicamycin), 5 days (for plates containing 0.5 μg/ml tunicamycin) and 7 days (for plates containing above 1 μg/ml tunicamycin). Assays for DTT toxicity were carried out as follows. Cells were grown to exponential phase, and cultures adjusted to an optical density of 0.05 were incubated with YPD medium or YPD medium containing 4 mM DTT for 12 h at 25°C. The sensitivity to DTT was estimated by dividing absorbance units in the presence of DTT by absorbance units in the absence of DTT, and then the ratios of DTT sensitivities of the mutants/wild-type were calculated as the DTT sensitivity index.
To visualize GFP-tagged Hog1 in living cells, cells grown to exponential phase were incubated with SD medium containing 2 μg/ml tunicamycin or 4 mM DTT. Cells were then harvested at the indicated times, suspended in SD medium, and observed immediately using a Keyence BZ-X700 microscope (Keyence Corporation, Japan). Fluorescence intensities were quantified using Hybrid Cell Count BZ-H2C software (Keyence Corporation, Japan). To confirm nuclear localization of Hog1-GFP, cells were fixed for 10 min at 25°C by direct addition of 37% formaldehyde to a final concentration of 3.7%. Cells were then washed with PBS, stained with 4',6-diamidino-2-phenylindole (DAPI) and subjected to microscope observation. Images of Hog1-GFP in fixed cells were similar to those observed in living cells.
|
10.1371/journal.pntd.0005635 | Safety and efficacy of short course combination regimens with AmBisome, miltefosine and paromomycin for the treatment of visceral leishmaniasis (VL) in Bangladesh | AmBisome therapy for VL has an excellent efficacy and safety profile and has been adopted as a first-line regimen in Bangladesh. Second-line treatment options are limited and should preferably be given in short course combinations in order to prevent the development of resistant strains. Combination regimens including AmBisome, paromomycin and miltefosine have proved to be safe and effective in the treatment of VL in India. In the present study, the safety and efficacy of these same combinations were assessed in field conditions in Bangladesh.
The safety and efficacy of three combination regimens: a 5 mg/kg single dose of AmBisome + 7 subsequent days of miltefosine (2.5 mg/kg/day), a 5 mg/kg single dose of AmBisome + 10 subsequent days of paromomycin (15 mg/kg/day) and 10 days of paromomycin (15 mg/kg/day) + miltefosine (2.5 mg/kg/day), were compared with a standard regimen of AmBisome 15 mg/kg given in 5 mg/kg doses on days 1, 3 and 5. This was a phase III open label, individually randomized clinical trial. Patients from 5 to 60 years with uncomplicated primary VL were recruited from the Community Based Medical College Bangladesh (CBMC,B) and the Upazila Health Complexes of Trishal, Bhaluka and Fulbaria (all located in Mymensingh district), and randomly assigned to one of the treatments. The objective was to assess safety and definitive cure at 6 months after treatment.
601 patients recruited between July 2010 and September 2013 received either AmBisome monotherapy (n = 158), AmBisome + paromomycin (n = 159), AmBisome + miltefosine (n = 142) or paromomycin + miltefosine (n = 142). At 6 months post- treatment, final cure rates for the intention-to-treat population were 98.1% (95%CI 96.0–100) for AmBisome monotherapy, 99.4% (95%CI 98.2–100) for the AmBisome + paromomycin arm, 94.4% (95%CI 90.6–98.2) for the AmBisome + miltefosine arm, and 97.9% (95%CI 95.5–100) for paromomycin + miltefosine arm. There were 12 serious adverse events in the study in 11 patients that included 3 non-study drug related deaths. There were no relapses or PKDL up to 6 months follow-up. All treatments were well tolerated with no unexpected side effects. Adverse events were most frequent during treatment with miltefosine + paromomycin, three serious adverse events related to the treatment occurred in this arm, all of which resolved.
None of the combinations were inferior to AmBisome in both the intention-to-treat and per-protocol populations. All the combinations demonstrated excellent overall efficacy, were well tolerated and safe, and could be deployed under field conditions in Bangladesh. The trial was conducted by the International Centre for Diarrhoeal Disease Research (ICDDR,B) and the Shaheed Suhrawardy Medical College (ShSMC), Dhaka, in collaboration with the trial sites and sponsored by the Drugs for Neglected Diseases initiative (DNDi).
ClinicalTrials.gov NCT01122771
| Treatment is one of the key strategies for visceral leishmaniasis control and elimination. Historically a number of monotherapy drugs for VL treatment were used in Bangladesh, including pentavalent antimonials, amphotericin B deoxycholate (AmB), and miltefosine (MF). With the limited number of drugs available, it was necessary to preserve existing drugs and also to develop shorter and safer treatment regimens. At the time the study was initiated, miltefosine monotherapy was a recommended first-line treatment in Bangladesh. The present study aimed to provide safety and efficacy data for three short-course combination regimens including AmBisome, miltefosine and paromomycin when rolled out in field conditions in Bangladesh, and to compare these to AmBisome monotherapy. All combinations proved non-inferior to AmBisome monotherapy and were safe and well tolerated. This study was implemented in field conditions at Upazila level with treatment provided by government doctors, providing further evidence for scaling up new regimens in national program contexts within the public health sector.
| World-wide, 200,000–400,000 new cases of visceral leishmaniasis (VL) occur annually [1]. The majority of these cases occur in South Asia; mainly in Bihar, India and neighbouring regions of Nepal, and in the highly endemic Mymensingh province of Bangladesh.
Before the introduction of single-dose AmBisome, miltefosine was included in the National Guidelines of India, Nepal and Bangladesh as a first line treatment for VL following a phase III trial in India that showed a final cure rate of 94% [2]. However, poor compliance due to its long treatment course (28 days) [3] and possible teratogenic effects have limited its successful roll-out. Moreover, a decrease in susceptibility to miltefosine was found in clinical isolates of relapse patients [4] and the efficacy of miltefosine declined to 90% within the last decade of use in South Asia [3]. A phase III clinical trial in India demonstrated that paromomycin at a dose of 15 mg/kg for 21 days provided a final cure rate of 94.6% [5], but this regimen was never implemented in Asia. As with miltefosine, there were indications that resistance to paromomycin might easily develop when used in monotherapy [6].
Due to the challenges in implementing clinical trials to assess treatment safety and efficacy, very few studies have been conducted in government sector at Upazila Health Complexes under real life conditions within national program settings. The present study aimed to assess the efficacy and safety of alternative combination treatments for primary visceral leishmaniasis and its implementation at different levels of the health system.
To reduce pressure on the drugs and prolong their therapeutic life-span, miltefosine and paromomycin should preferably be used in combination; short course combination regimens will lead to better compliance and are more readily implemented at health facility level. Three short-course combination regimens including AmBisome, miltefosine and paromomycin have been evaluated in a phase III clinical trial conducted in India (2008–2010). All showed an excellent safety profile and an efficacy of at least 97% in controlled conditions [7].
The present study aimed to compare the safety and efficacy of the following combination regimens with AmBisome alone for the treatment of VL in Bangladesh:
Ethical approval was obtained from the Ethical Review Committee of the International Centre for Diarrhoeal Disease Research, Bangladesh (ICDDR,B) and the National Research Ethics Committee (NREC) of the Bangladesh Medical Research Council (BMRC) prior to starting the study. The study was conducted in accordance with the ICH Harmonized Tripartite Guideline–Guideline for Good Clinical Practice (GCP), and ethical principles enshrined in the Declaration of Helsinki.
This was a randomized, controlled, open-label, parallel group phase III clinical trial conducted in Mymensingh province, Bangladesh. Initially, 120 patients were recruited and treated in a hospital setting (the Community Based Medical College, CBMC) primarily for the purpose of confirming the safety of the combination treatments in Bangladesh. After review of safety data up to day 45 from the 120 patients treated at CBMC, and based on good safety profile of the treatments, the Data Safety Monitoring Board (DSMB) recommended to initiate recruitment at Upazila Health Complexes (UZHC). Additional 482 patients were recruited for the trial and treated in the Upazilas of Trishal, Gaffargoan and Bhaluka.
HIV negative primary VL patients between 5 and 60 years screened with positive rK39 rapid immunochromatographic tests (InBios, Seattle, USA) and parasitologically confirmed via bone marrow or spleen aspirates (only at CBMC) were enrolled into the study after giving informed consent. Because of the potential teratogenic effects of miltefosine, women of child-bearing age who were not using an assured method of contraception for the duration of treatment and three months afterwards were excluded, unless they agreed to receive an injection of medroxyprogesterone acetate (DepoProvera, Pfizer, NY, USA). One injection (which is effective for 3 months) was needed to ensure adequate coverage, taking into account miltefosine’s long half-life of approximately 7 days. Other exclusion criteria were: known hepatitis B, hepatitis C, or HIV infection, Hb concentrations less than 5 g/dl, platelet count of less than 40,000/mm3 (at CBMC only), a prothrombin time of more than 5 seconds longer than the control (at CBMC only), severe malnutrition [for adults (> 18 years) defined as BMI <14; for children (< 18 years) defined as BMI for age z score < -3 in children measuring > 121.5cm; and weight for height less than 60% in children measuring <121.5 cm], known alcohol or drug abuse, use of any investigational (unlicensed) drug within the last 3 months, and severe concurrent illnesses (TB, malaria) or chronic conditions (diabetes, hypertension). Pregnant and breast-feeding women, and patients with known hypersensitivity to the study drugs were also excluded.
Patients were randomized to four treatment arms. In the reference treatment group, patients were given the standard regimen of 15 mg/kg AmBisome, given in doses of 5 mg/kg on days 1, 3 and 5, infused over 2 hours in 5% dextrose solution. Patients allocated to the AmBisome + miltefosine arm received 5 mg/kg AmBisome on day 1, followed by 7 days of miltefosine (Impavido, Paladin, Canada). Miltefosine is provided as foil-wrapped blister packs of 10 and 50 mg capsules and was given to adults (>12 years) at a dose of 50 mg once daily if < 25 kg bodyweight, 50 mg twice daily if > 25 kg bodyweight; or 2.5 mg/kg/day divided into 2 doses for children younger than 12 years. Patients assigned to the AmBisome + paromomycin arm received 5 mg/kg AmBisome on day 1, followed by 10 days of paromomycin (Gland Pharma, India) at a dose of 11 mg/kg/day (base) equivalent to 15 mg/kg/day (sulphate salt), given by deep IM injection. Patients allocated to the miltefosine + paromomycin arm received the above described doses daily for 10 days. Patients treated with miltefosine on an outpatient basis in UZHC’s were given detailed instructions on the number of capsules to be taken each day, and asked to return to the clinic in case of vomiting. Miltefosine was re-dosed within one hour of vomiting. Patients were asked to return the empty blister packs for drug accountability.
The main outcome was final cure, defined as initial cure at day 45 and absence of VL signs and symptoms during the follow-up period of 6 months.
The secondary outcome was initial cure, defined as clinical improvement at day 45. In the CBMC hospital setting, initial cure was confirmed by the absence of parasites in splenic/bone marrow aspirates at day 15. In cases of 1+ parasite at day 15, patients were retested at day 45. Patients who presented positive parasitology at day 45, or cases of relapse following combination treatment, were to receive rescue treatment with AmBisome 15 mg/kg. Those who failed to respond/relapsed following the reference treatment, or where AmBisome was contra-indicated, were to be rescued with sodium stibogluconate (SSG) 20mg/kg/day for 30 days, or miltefosine 2.5mg/kg/day for 28 days as recommended by the National Treatment Guidelines of Bangladesh and the investigator’s judgement. Patients with initial cure presenting with VL symptoms during follow-up were suspected of relapse, and were referred for parasitological confirmation at the CBMC hospital. Safety outcomes were adverse events and serious adverse events recorded during treatment and up to 6 months afterwards. Assessments were done on the basis of clinical adverse events systematically at all study sites. At the CBMC hospital, laboratory investigations included liver enzymes alanine aminotransferase (ALAT), aspartate aminotransferase (ASAT), bilirubin, prothrombin time, platelets, RBC count, WBC count, random blood glucose (RBS), urea, creatinine, serum sodium, potassium, magnesium. At the UZHC settings, laboratory parameters were limited to haemoglobin and random blood glucose (RBS). All patients were assessed for dermal manifestations of leishmaniasis (Para-KDL at baseline, and PKDL at day 15, 45 and at 6 months).
A computer generated randomization code was used for patient treatment allocation. Individual, opaque, sealed and sequentially numbered envelopes were provided to each study site (one envelope per patient), indicating the individual patient allocation to treatment. Eligible patients who fulfilled all the inclusion criteria, met none of the exclusion criteria and from whom informed consent had been obtained were randomized to treatment regimens using the sealed envelopes. Randomization was stratified by treatment centre and was done using an equal allocation ratio for married women, men, and children firstly; and further with an equal ratio in single women of childbearing potential. Single women of childbearing age were randomized to receive either AmBisome alone or AmBisome + paromomycin. The allocation ratio was adjusted to account for this, in order to achieve approximately equal numbers of patients in each arm. Thus the allocation ratio was 1:1:1:1 for four treatments in married women, men, and children; and 1:1 (for AmBisome alone or AmBisome + paromomycin) in single women of childbearing potential. This was an open-label study; miltefosine is an orally administered medication and AmBisome and paromomycin are administered IV and IM respectively.
The sample size was calculated assuming a treatment success of 97% in the reference arm (AmBisome) and a margin of non-inferiority of any tested treatment of 7%, leading to a minimally acceptable cure rate of 90% for each treatment. With a power of 90%, the sample size per group in the sample of married women, men and children would be 140. Based on the possible teratogenic effect of miltefosine and the assumption that 20% of the patients would be single women of child bearing potential, 26 extra patients were to be recruited among these women for both non-miltefosine groups. Assuming a drop-out rate of 10%, 154 patients were needed in both miltefosine groups and 183 in both non-miltefosine groups. The total sample size was calculated to be 674 patients.
The primary efficacy analysis was performed using the standard approach of non-inferiority for the comparison AmBisome vs AmBisome + miltefosine and for AmBisome vs. paromomycin + miltefosine on populations ITT1 and PP1. For AmBisome vs AmBisome + paromomycin comparisons, a logistic model was carried out using populations ITT and PP (Table 1).
The Intention to Treat 1 (ITT1) population included married women, men and children randomized to the trial and who received at least one dose of the study medication; but it did not include the single women of childbearing potential who were randomized either to AmBisome or Ambisome + paromomycin arms. The Intention to Treat population (ITT) included all patients randomised to the treatment groups who gave informed consent and who took at least one dose of study medication.
The Per Protocol 1 population (PP1) included patients (married women, men and children) enrolled in the Intent to Treat 1 population who were randomized to the trial, who had no major protocol deviations and who completed the 6 month follow-up visit or were classified as a treatment failure and received rescue medication. The Per Protocol population (PP) included all patients in the ITT population with no major protocol deviations and who completed the 6 month follow-up visit or were classified as a treatment failure and received rescue medication.
The ITT population was 601 and the PP population was 587. The reasons for exclusion in the PP population were 12 cases of withdrawal due to AE/SAE, which required rescue treatment, and 2 deaths.
Of the 673 patients screened, 71 did not meet the inclusion criteria. A total of 24 screened patients did not have VL diagnosis confirmed by rK39 or microscopy. Among the patients with proven disease, the most common reasons for exclusion were chronic underlying disease (n = 14) and simultaneous participation in another study (n = 9). Five patients refused consent, four had abnormal laboratory parameters, 3 were pregnant or lactating women, and 12 were excluded for other reasons. The decision to stop recruitment was made after enrolling 602 patients (120 in the CBMC hospital setting, 482 in UZHC’s). Only 2% rather than the estimated 10% of patients were lost to follow-up, so that the sample size could be reduced by 72 patients.
The patient flow through the study is shown in Fig 1. Patients were recruited from July 6, 2010 to September 2013 and asked to return to the clinic at day 45 and at 6 months after treatment onset, or sooner if any symptoms of VL reoccurred. Follow-up of all patients was completed by the end of March 2014. Randomisation produced groups with no significant differences in the main baseline characteristics across treatment groups (e.g. age, sex, weight, nutritional status, spleen size, haemoglobin) (Tables 2–4).
673 patients screened in the study, 602 randomised in the study, 71 patients did not meet inclusion criteria. 587 patients in the PP population, with: 3 deaths; 4 SAEs and 09 AEs that lead to treatment discontinuation, n = 13; 10 protocol deviations; 2 miltefosine redosing when vomiting occurred > 1h after administration; 3 patients missed a miltefosine dose; 2 received wrong Miltefosine capsule strength; 1 received additional miltefosine medication after finishing treatment; 1 mistake in randomization: randomized to PM+Milt administered AmBisome; 1 patient did not receive single dose of study medication after randomisation to AmBisome arm.
In the safety population (ITT) of 601 patients, 374 (62%) were males and 227 (38%) were females. 85% of patients had haemoglobin < 10 mg/dl at the time of screening. There were 594 (99%) patients who complained of weight-loss and feelings of weakness. There were 588 (98%) patients who had pallor at baseline.
The primary analysis was intention-to-treat; data were available for 158 patients randomized to AmBisome (1 withdrew consent before treatment started), 159 to AmBisome + paromomycin, 142 patients to AmBisome + miltefosine and 142 patients to paromomycin + miltefosine, with a total of 601 patients included in the ITT analysis.
Four patients were withdrawn from the study due to serious adverse events (SAEs), and another nine patients due to adverse events (AEs) that required treatment discontinuation, making a total of 13 early withdrawals due to adverse events. Three deaths occurred in the study: one in the AmBisome arm and two in AmBisome + miltefosine arm.
There were 10 cases of protocol deviations. Protocol deviations were due to miltefosine re-dosing to patients presenting with vomiting more than 1 hour after drug administration (2 patients), missing of a miltefosine dose (3 patients) or inadvertently receiving the wrong miltefosine capsule strength (2 patients). One patient took additional miltefosine after finishing her study medication, and one patient randomized to paromomycin + miltefosine treatment was administered AmBisome. Finally, one patient did not receive even a single dose of study medication after randomization to the AmBisome arm.
There were no patients lost to follow-up.
Cure rates at 45 days (initial cure) and 6 months (final cure) are shown in Tables 5 and 6.
In the ITT population, the initial and final cure rates were >95% in all arms, except in the AmBisome + miltefosine arm (94.4% for both initial and final cure rate). All arms showed a final cure rate of >95% in the PP population.
In children, the initial cure rate is 100% in all groups except the paromomycin + miltefosine group with a cure rate of 96.7%. However, adolescents achieved 100% cure initial cure rate in the paromomycin + miltefosine group (Table 6).
In the ITT population, final cure rates were > 95% in children and adolescents in all groups. The lowest final cure rate was observed in the ITT population in adults treated with AmBisome + miltefosine (91.5%). Treatment failures in this arm were due to two cases of death not related to treatment and AEs or SAEs that lead to treatment discontinuation where patients required rescue treatment. Although not statistically significant, AmBisome + paromomycin was the most effective treatment with initial and final cure rates of 99.4% in the ITT population. Out of the first 120 patients that were included and followed up in the hospital setting (CBMC), parasitological cure measured on day 15 was achieved in > 90% of patients in all groups (patients who did not have cure confirmed were due to ‘no tissue to perform the test’ or ‘test not done’), and in 100% in the AmBisome + paromomycin group (Table 7). There were no relapses and no cases of PKDL up to 6 months follow-up. None of the combination treatments were inferior to AmBisome monotherapy when compared in pairs for both the intention-to-treat and the per-protocol populations (Table 1).
There were 11 patients who experienced a total of 12 SAEs, which included three non-drug related deaths (severe pneumonia, sudden cardiac death and hepatic encephalopathy). In the miltefosine + paromomycin group, three drug-related SAEs occurred; two of which occurred in the same patient. This patient was a 50 year old male who developed drug induced nephropathy and ototoxicity two weeks after treatment, both probably related to paromomycin; mild hearing loss was still present at 6 months after treatment. In a 40 year old male, acute hepatitis developed and worsened during treatment but resolved spontaneously after treatment was finished. In the AmBisome + miltefosine group, a 35 year old female patient presented with high grade fever, rash and swelling of arms and legs after 2 days of Miltefosine, which was possibly drug related. Treatment was interrupted and she was later diagnosed with Rickettsial fever with concomitant nutritional oedema. Rescue treatment was given with AmBisome and she made a full recovery. None of the other non-fatal reported SAEs (encephalitis, internal bleeding due to a peptic ulcer, acute respiratory tract infection, epilepsy and viral encephalitis) were related to treatment.
368 out of 602 patients experienced at least one AE in the study. Approximately 34% of these AEs were related to treatment; these included vomiting in one fifth of patients in miltefosine containing treatment regimens and pyrexia in AmBisome containing treatment regimens. Vomiting directly after administration of miltefosine was common; there were 28 (20%) patients in the paromomycin + miltefosine arm and 16 (11%) in the AmBisome + miltefosine arm that needed re-dosing within the hour. The proportion of patients that experienced any treatment-related side effects was highest in the AmBisome + miltefosine arm (42%), and lowest (27%) in the AmBisome arm (Table 8).
Extensive biochemical testing was only done for patients treated in the hospital setting (120 patients). The clinically significant laboratory adverse events considered related to the study drug are described in Table 9. All of them were classified as mild.
There were no statistically significant changes over the treatment period in any of the biochemical parameters. Haematological parameters (haemoglobin, red and white blood cell counts and platelets) were improved at day 45 as compared to baseline, without a significant difference between the individual treatment arms (S1–S4 Tables, supporting information).
All combinations proved non-inferior to the standard treatment with AmBisome, with definitive cure rate differences in relation to AmBisome compared to: AmBisome + paromomycin ITT 1.3% (95%CI -1.73, 4.27); AmBisome + miltefosine ITT1–3.7% (95%CI -9.20,1.85) and paromomycin + miltefosine ITT1–0.1% (95%CI -4.15, 4.03) (Table 1). Treatments were well tolerated and no new safety signal was identified. Overall adverse events were of mild intensity. The internationally accepted parameters for efficacy of VL treatment (≥95%) [8] were also met for all combinations in both the per-protocol (PP) and the intention-to-treat (ITT) populations, except for AmBisome + miltefosine, which showed an efficacy of 94.4% in the ITT population. These cases of failure were not due to lack of response or relapse, but related to two deaths and a higher number of treatment discontinuations in relation to adverse events, requiring rescue treatment. The trial was not powered to detect differences between the treatment arms; however, a small but significant difference was found between the efficacy of AmBisome + miltefosine and AmBisome + paromomycin in both the PP and ITT population. Post-hoc analysis of the data stratified per age group showed that this difference only remained significant in adult patients (Table 5). There was no loss to follow up at 6 months as patients were actively tracked by committed field workers, and there were no relapses or PKDL. However, there is recent evidence that most relapses occur after 6 months [9, 10]. We therefore recommend, in line with other authors [9], to follow VL patients for at least 12 months before determining the final treatment outcome.
The combination regimens described in this paper have been studied earlier in a Phase III clinical trial conducted in India (2008–2010) [7]. The excellent safety and efficacy outcomes in the present study support those found in India. The main difference with the Indian study is that patients have been mostly treated in field conditions at Upazila level, with treatment provided by government doctors. This study provides evidence that it is feasible to scale up the implementation of combination regimens within national program settings and that these are acceptable to patients as well as doctors. However, the patient population was selected to have non-severe disease, and we recommend active pharmacovigilance at sentinel sites in Bangladesh documenting treatment outcome (side effects, early treatment failure, relapse and PKDL) on the full patient population after implementation of combination regimens.
Recent discussions among decision-makers around the Road Map for elimination of VL in South Asia have led to the inclusion of single dose AmBisome as a first-line treatment and miltefosine + paromomycin as an alternative recommended treatment for VL in India and Bangladesh. Earlier, combination regimens had already been recommended by the WHO Expert Committee [8] and the Regional Technical Advisory Group (RTAG) for adoption by policy makers after demonstration of the feasibility of their implementation in field conditions [11]. The evidence generated in the present study supports the use of combination treatments as valid alternatives to single dose AmBisome therapy. The most cost effective combination appears to be 10 days of miltefosine + paromomycin, since this can be given on an outpatient basis [12]. However, ultimately the choice of treatment will depend on the circumstances. Considering the requirement for a cold chain for AmBisome and the availability of trained staff to give intravenous infusions, miltefosine and paromomycin given on outpatient basis may be a suitable treatment in most settings. But as miltefosine cannot be given to women of child bearing age who refuse contraception, AmBisome and paromomycin is a viable alternative. Given the fact that paromomycin is not currently registered in Bangladesh, AmBisome + miltefosine may be considered as an interim solution.
In Bangladesh, evidence on the excellent efficacy and safety profile of a single dose (10 mg/kg) of AmBisome was generated in 2010 [13] and it was soon thereafter adopted as a highly promising tool for regional elimination. Rolled out in a rural public hospital in Bangladesh, single dose AmBisome showed a final cure rate of 97% and this provided sufficient evidence for scaling up the use of AmBisome in the region as a first-line treatment in hospital settings [14]. Single dose AmBisome was adopted as a first-line treatment in Bangladesh in 2013, when the current clinical trial was still ongoing. Data from this clinical trial support the use of combination regimens as 2nd line treatments for VL in Bangladesh.
The validation and use of combination therapy to provide an alternative to AmBisome in the context of a VL elimination program, where AmBisome is used as a first-line treatment for uncomplicated VL, relapse VL, HIV/VL co-infected patients and PKDL, is crucial.
|
10.1371/journal.pmed.1002540 | Comorbidity health pathways in heart failure patients: A sequences-of-regressions analysis using cross-sectional data from 10,575 patients in the Swedish Heart Failure Registry | Optimally treated heart failure (HF) patients often have persisting symptoms and poor health-related quality of life. Comorbidities are common, but little is known about their impact on these factors, and guideline-driven HF care remains focused on cardiovascular status. The following hypotheses were tested: (i) comorbidities are associated with more severe symptoms and functional limitations and subsequently worse patient-rated health in HF, and (ii) these patterns of association differ among selected comorbidities.
The Swedish Heart Failure Registry (SHFR) is a national population-based register of HF patients admitted to >85% of hospitals in Sweden or attending outpatient clinics. This study included 10,575 HF patients with patient-rated health recorded during first registration in the SHFR (1 February 2008 to 1 November 2013). An a priori health model and sequences-of-regressions analysis were used to test associations among comorbidities and patient-reported symptoms, functional limitations, and patient-rated health. Patient-rated health measures included the EuroQol–5 dimension (EQ-5D) questionnaire and the EuroQol visual analogue scale (EQ-VAS). EQ-VAS score ranges from 0 (worst health) to 100 (best health). Patient-rated health declined progressively from patients with no comorbidities (mean EQ-VAS score, 66) to patients with cardiovascular comorbidities (mean EQ-VAS score, 62) to patients with non-cardiovascular comorbidities (mean EQ-VAS score, 59). The relationships among cardiovascular comorbidities and patient-rated health were explained by their associations with anxiety or depression (atrial fibrillation, odds ratio [OR] 1.16, 95% CI 1.06 to 1.27; ischemic heart disease [IHD], OR 1.20, 95% CI 1.09 to 1.32) and with pain (IHD, OR 1.25, 95% CI 1.14 to 1.38). Associations of non-cardiovascular comorbidities with patient-rated health were explained by their associations with shortness of breath (diabetes, OR 1.17, 95% CI 1.03 to 1.32; chronic kidney disease [CKD, OR 1.23, 95% CI 1.10 to 1.38; chronic obstructive pulmonary disease [COPD], OR 95% CI 1.84, 1.62 to 2.10) and with fatigue (diabetes, OR 1.27, 95% CI 1.13 to 1.42; CKD, OR 1.24, 95% CI 1.12 to 1.38; COPD, OR 1.69, 95% CI 1.50 to 1.91). There were direct associations between all symptoms and patient-rated health, and indirect associations via functional limitations. Anxiety or depression had the strongest association with functional limitations (OR 10.03, 95% CI 5.16 to 19.50) and patient-rated health (mean difference in EQ-VAS score, −18.68, 95% CI −23.22 to −14.14). HF optimizing therapies did not influence these associations. Key limitations of the study include the cross-sectional design and unclear generalisability to other populations. Further prospective HF studies are required to test the consistency of the relationships and their implications for health.
Identification of distinct comorbidity health pathways in HF could provide the evidence for individualised person-centred care that targets specific comorbidities and associated symptoms.
| Heart failure is an increasingly common condition, and patients often experience persistent symptoms and poor quality of life, even when they are receiving the best possible treatment for their heart failure.
Most heart failure patients have other conditions that dominate their health experience, yet heart failure treatment focuses on their cardiovascular status.
There is a lack of understanding about the relationships among different comorbidities and quality of life in heart failure, which are important to guide individualised treatment plans for patients.
We used an established health-related quality of life model to develop and test a new heart failure health model that included the most common heart failure comorbidities.
We tested this model by examining the postulated relationships among comorbidities, symptoms and functional limitations reported by patients, and their overall health experience, using a national register of heart failure patients in Sweden.
We found that non-cardiovascular comorbidities were associated with much higher overall symptom burden and more severe symptoms than cardiovascular comorbidities.
Predominant symptoms for cardiovascular comorbidities were pain and anxiety, whereas for non-cardiovascular comorbidities they were shortness of breath and fatigue.
Heart failure optimising therapies did not influence these symptoms, functional limitations, or quality of life.
Current guidelines in heart failure focus on improving cardiovascular status in response to common heart failure symptoms (shortness of breath, fatigue, and leg swelling).
Our study shows that for some patients, these symptoms might be driven by non-cardiovascular conditions such as diabetes and renal disease, rather than their cardiovascular status. We found that cardiovascular comorbidities were more likely to be associated with pain and anxiety than shortness of breath or fatigue.
To improve health-related quality of life, heart failure guideline-driven care needs to include optimal management of the most prevalent non-cardiovascular comorbidities and routine management of pain and anxiety or depression.
To provide individualised patient care, guidelines need to better align symptoms with the cardiovascular and non-cardiovascular status of the patient.
| Heart failure (HF) is a complex clinical syndrome of multiple symptoms, functional impairments, and poor health-related quality of life (HRQoL). With modern therapies, HF patients are now living longer but with a potentially higher symptom burden [1] that can be worse compared to people with other chronic diseases including cancer [2]. Inadequate symptom control and poor HRQoL are significant drivers of hospitalisations, readmissions, and death in HF [3,4].
HF patients are usually older, with a high number of comorbidities, and a third of patients report that other medical conditions dominate their health experience [5], yet guideline-driven symptom management in HF focuses on cardiovascular status [6]. Persisting symptoms and poor HRQoL after optimisation of HF treatments [7] suggest that comorbidities may be an important determinant of health and that non-cardiovascular comorbidities may be associated with HF-related symptoms [8]. However, the evidence is inconsistent, with some HF studies showing associations between comorbidities and HRQoL [9–11] and others showing no such associations [12,13]. Whilst comorbidities are consistent predictors of morbidity and mortality in HF [14], their interrelationships with symptoms, functional limitations, and overall health have not been explored. We used the Swedish Heart Failure Registry (SHFR), which is one of the largest population-based HF registers, to routinely collect patient-reported outcomes [15] to investigate the interplay among HF comorbidities, symptoms, functional limitations, and patient-rated health.
The SHFR is a national population-based register of all HF patients admitted to hospitals or attending outpatient clinics in participating primary care units [15]. The register collects HRQoL data measured using the EuroQol–5 dimension (EQ-5D) questionnaire, including the EuroQol visual analogue scale (EQ-VAS), at baseline registration, as well as patient demographics and clinical and healthcare information. We obtained first entry data for all HF patients included in the SHFR from 1 February 2008 to 1 November 2013 with EQ-5D recorded at baseline.
Establishment of the registry and analysis of data were approved by a multisite ethics committee and conform to the Declaration of Helsinki. Individual patient consent is not required, but patients are informed of entry into national registries and can opt out. Access to the SHFR was granted following project approval (S1 Proposal) by the SwedeHF Research Board. The original statistical analysis plan was based on standard linear and logistic regression analyses. The sequences-of-regressions approach was adopted at the early stages of data acquisition to provide a novel method of analysis to deal effectively with the complexity underlying the analysis of postulated relationships among many variables. This study is reported as per the RECORD guidelines (S1 Checklist).
Previously, an a priori model by Wilson and Cleary [16] hypothesized a potential pathway from chronic disease to patient-reported outcomes and patient-rated health. The model proposes a pathway of linkages between 5 domains: (i) bio-physiological status, (ii) symptoms, (iii) functional status, (iv) general health perception (i.e., patient-rated health), and (v) quality of life (Fig 1). We used this concept to develop a HF health model with a focus on comorbidities to test the hypothesis that the potential mechanisms underlying health in HF differ among selected comorbidities. We included cardiovascular conditions related to the aetiology of HF (defined as cardiovascular comorbidities) to compare with other, non-cardiovascular comorbidities. Wilson and Cleary’s model has since been adapted to take account of patient and environmental factors that might influence any stage of the model, including bio-physiological status [17]. We used this adapted version, given the clear link between these factors and the development of both HF and comorbidities in older age. All variables selected to include in the model were chosen by an expert group including HF specialists (Fig 2). The overall domains within the postulated pathway were (i) patient and environmental factors to (ii) comorbidities to (iii) left ventricular ejection fraction to (iv) symptoms to (v) functional limitations to (vi) patient-rated health. We tested the direct relationships between the individual domains and patient-rated health and the indirect relationships that were explained by intermediary domains. We did not include the final domain in Wilson and Cleary’s model (overall quality of life), as this includes non-health-related factors.
HF patient factors were age, sex, marital status (single or married/partner), body mass index (BMI), smoking status (current or former/none), HF duration (≥6 or <6 months), heart rate, haemoglobin, devices, and prescription of HF optimising drugs (beta blockers, angiotensin converting enzyme inhibitors [ACEis], angiotensin II receptor blockers [ARBs]) and diuretics. Environmental factors were healthcare provider (inpatient or outpatient) and speciality (cardiology or medicine). Of the continuous variables, age and heart rate values were all within the plausible range, and BMI and haemoglobin were truncated at the boundaries of their respective plausible ranges (<10 values truncated in total).
HF biological status was based on left ventricular ejection fraction defined as reduced (<40%), midrange (40%–49%), or preserved (≥50%). For the sequences of regressions analyses, ejection fraction was dichotomised as ≥40% or <40%. Cardiovascular comorbidities were hypertension, dilated cardiomyopathy (DCM), valve disease, atrial fibrillation (AF), and ischemic heart disease (IHD), and non-cardiovascular comorbidities were diabetes, chronic obstructive pulmonary disease (COPD), and chronic kidney disease (CKD). CKD was defined by an estimated glomerular filtration rate of <60 ml/min/1.73 m2.
Previous evidence has shown only weak associations between HF biological status and shortness of breath (SOB) [18–20], but the influence of other conditions or symptoms on this relationship has not been considered. Two of the most frequently reported HF symptoms are SOB and fatigue [21]. However, HF patients also commonly experience pain and low mood [8,21,22], and these symptoms negatively impact functional performance [23]. So, for the symptoms domain, we included pain and anxiety or depression dimensions from the EQ-5D questionnaire, together with patient-reported SOB and fatigue. For functional limitations, the ‘usual activities’ EQ-5D dimension was selected, which is most inclusive of a range of physical, emotional, and social functions. SOB and fatigue were dichotomised as ‘no or slight’ and ‘marked or severe’, and pain, anxiety or depression, and functional limitations were dichotomised as ‘no problems’ or ‘any problems’.
Patient-rated health is a known predictor of mortality in HF [24]. In this study, patient-rated health was based on the EQ-VAS, which is a numerical scale from 0 (worst health imaginable) to 100 (best health imaginable) [25]. The EQ-5D is valid, comparable to other general [26] and specific quality of life tools [27], and widely used in HF clinical trials [9].
There were 10,575 HF patients with EQ-VAS recorded at inclusion in the register, and patient characteristics are presented in Table 1 (comparison to patients in the overall register including those without EQ-VAS recorded can be seen in S1 Table). Median age was 74 years (IQR 65–81), and 33% were women. A quarter (26%) of patients were inpatients, and most (90%) were prescribed HF optimising drug therapies. Prevalence of symptoms and functional limitations was high, with 85% patients experiencing SOB and fatigue and over a third (35%) reporting marked or severe intensity of both symptoms. Approximately half reported pain (52%) and anxiety (45%). Functional limitations were reported by 37% of HF patients, and the mean EQ-VAS score was 63 (SD 20).
The prevalence of cardiovascular comorbidities was high, with half of HF patients having AF (48%), hypertension (52%), and IHD (54%). Non-cardiovascular comorbidities were also frequent: 44% of patients had CKD, 17% had COPD, and 24% had diabetes. There were 415 (3.9%) HF patients without any comorbidities. These patients were on average 10 years younger, mostly outpatients (83%), and much more likely to have shorter duration of HF (<6 months) than HF patients with comorbidities (82% versus 50%). The prevalence of symptoms and functional limitations was higher for HF patients with comorbidities than for those without and differed between those with cardiovascular and non-cardiovascular comorbidities, as shown in Fig 3.
There was a marked increase in the prevalence of symptoms and functional limitations from HF patients with no comorbidities to those with cardiovascular comorbidities to those with non-cardiovascular comorbidities. Respectively, mean prevalence among the 3 comorbidity categories was 26%, 37%, and 46% for marked or severe SOB; 27%, 36%, and 44% for marked or severe fatigue; 45%, 52%, and 59% for pain; 42%, 46%, and 48% for anxiety or depression; and 28%, 38%, and 45% for functional limitations. Patient-rated health showed a similar pattern (Fig 4), where mean EQ-VAS score progressively declined from no comorbidities (mean EQ-VAS score, 66) to cardiovascular comorbidities (mean EQ-VAS score, 62) to non-cardiovascular comorbidities (mean EQ-VAS score, 59).
Summaries of the sequences of regressions fitted for the HF health model are presented in Tables 2 and S2. Patient-rated health (EQ-VAS score, 1 to 100) was directly explained by a combination of patient and environmental factors, symptoms, and functional limitations. These factors collectively explained 32% of the variance in patient-rated health (coefficient of multiple determination R2 = 0.32, 95% CI 0.31 to 0.34). There were direct negative associations with patient-rated health for older age, single status, longer HF duration, higher heart rate, prescribed diuretic, being an inpatient, having diabetes, and having symptoms and functional limitations. Care under cardiology compared to general medicine was associated with better patient-rated health (mean difference in EQ-VAS score, 1.28, 95% CI 0.54 to 2.03). Two of the strongest direct associations with worse patient-rated health were anxiety or depression (mean difference in EQ-VAS score, −18.68, 95% CI −23.2 to 14.1) and functional limitations (mean difference in EQ-VAS score, −7.51, 95% CI −8.37 to −6.65). Whilst most comorbidities had significant crude associations with patient-rated health (S3 Table), they did not have direct associations with patient-rated health once other variables including symptoms and functional limitations were considered, except for diabetes.
For simplicity of presentation, the indirect pathways between comorbidities and patient-rated health are displayed in 2 graphs (cardiovascular comorbidities in Fig 5 and non-cardiovascular comorbidities in Fig 6). Individual and environmental factors are not displayed.
Whilst none of the cardiovascular comorbidities had a direct association with patient-rated health, 2 had indirect associations through pain, anxiety or depression, and functional limitations (Fig 5). There were no associations between any of the cardiovascular comorbidities and SOB or fatigue. The association between AF and patient-rated health was explained by an increased risk of anxiety or depression (OR 1.16, 95% CI 1.06 to 1.27), which in turn was associated with increased risk of functional limitation (OR 10.03, 95% CI 5.2 to 19.5) and subsequently worse patient-rated health (mean difference in EQ-VAS score, −7.51, 95% CI −8.37 to −6.65). IHD was associated with pain (OR 1.25, 95% CI 1.14 to 1.38) and anxiety or depression (OR 1.20, 95% CI 1.09 to 1.32), which were both associated with functional limitations and subsequent worse patient-rated health. There were also direct associations between pain and worse patient-rated health (mean difference in EQ-VAS score, −4.51, 95% CI −5.31 to −3.71) and anxiety or depression and worse patient-rated health (mean difference in EQ-VAS score, −18.68, 95% CI −23.22 to −14.14) that were not explained by functional limitations. DCM, hypertension, and valve disease in HF were not associated directly or indirectly with symptoms, functional limitations, or patient-rated health. Whilst all cardiovascular comorbidities were directly associated with ejection fraction, there was no associations between ejection fraction status and any of the other symptoms or health domains.
Diabetes, COPD, and CKD had indirect associations with patient-rated health through symptoms and functional limitations, but there were variations in symptom associations (Fig 6). Diabetes and CKD were significantly associated with marked or severe SOB (OR 1.17, 95% CI 1.03 to 1.32, and OR 1.23, 95% CI 1.10 to 1.38, respectively) and fatigue (OR 1.27, 95% CI 1.13 to 1.42, and OR 1.24, 95% CI 1.12 to 1.38) compared to HF patients without these comorbidities. However, there were no associations for either comorbidity with pain and anxiety or depression. COPD, in contrast, was associated with all 4 symptoms: SOB (OR 1.84, 95% CI 1.62 to 2.10), fatigue (OR 1.69, 95% CI 1.50 to 1.91), pain (OR 1.34, 95% CI 1.19 to 1.52), and anxiety or depression (OR 1.21, 95% CI 1.08 to 1.36). As with pain and anxiety or depression, SOB and fatigue were associated with functional limitations (SOB, OR 1.98, 95% CI 1.74 to 2.26; fatigue, OR 2.45, 95% CI 2.16 to 2.79) (and subsequent worse patient-rated health) and also had direct associations with patient-rated health (SOB, mean difference in EQ-VAS score, −6.24, 95% CI −7.30 to −5.14; fatigue, mean difference in EQ-VAS score, −5.53, 95% CI −6.51 to −4.45) that were not explained by functional limitations. Again, ejection fraction did not explain any of the relationships between comorbidities and the other patient-rated health domains.
There were indirect pathways through comorbidities to patient-rated health for age, sex, BMI, smoking, speciality, and healthcare provider. With few exceptions, older age, higher BMI, and being an inpatient rather than an outpatient were associated with higher levels of cardiovascular and non-cardiovascular comorbidities. Women were less likely to have cardiovascular comorbidities and more likely to have COPD and CKD than men. Compared to general medicine, cardiology care was associated with the presence of some but not all cardiovascular comorbidities as well as CKD (see S2 Table). ACEi or ARB prescription in HF patients was associated with reduced fatigue compared to patients not prescribed an ACEi or ARB (OR 0.70, 95% CI 0.59 to 0.82), and those with a device had increased fatigue compared to patients without a device (OR 1.26, 95% CI 1.09 to 1.45), but there were no other significant associations between HF optimising therapies (pharmacology or devices) and symptoms, functional limitations, or patient-rated health (Table 2).
Our study used an a priori evidence-informed health model to investigate the potential pathways linking comorbidities with patient-reported symptoms, functional limitations, and patient-rated health in a large population-based registry of over 10,000 patients. The importance of elucidating these pathways for clinical management is the potential to improve HF prognosis by tailoring interventions to an individual patient’s risk, pathology, and health. Uniquely, by using sequences of regressions to separate out direct and indirect associations, we found that the potential pathways to HF patient health are different for cardiovascular compared to non-cardiovascular comorbidities and among specific comorbid diseases. Importantly, with minor exception, HF optimising therapies were not associated with symptoms, functional limitations, or patient-rated health. These key findings provide the evidence for a step-change in understanding and testing mechanisms between HF and overall patient-rated health and for developing precision medicine that targets specific comorbidities and associated symptoms.
There are 3 key findings with important implications for clinical care. First, the main symptoms associated with cardiovascular comorbidities were pain and anxiety or depression and not SOB and fatigue. Notably, despite the high prevalence and severity of SOB and fatigue in HF generally, around half of all patients reported pain and anxiety or depression, which is similar to levels found in hospitalised HF patients [21]. Furthermore, anxiety or depression had the strongest associations, out of all 4 symptoms, with functional limitations and patient-rated health. Chronic depression was poorly reported in the SHFR, so the number of people with chronic or new depression symptoms is likely a lot higher. This means that the association between depression and functional limitation or poor health may be even greater. Yet, it has long been recognised that psychosocial factors such as anxiety and depression are infrequently assessed or treated in HF clinical practice, with only a minority of HF patients with depression prescribed antidepressant drugs or counselling [22] or referred for cognitive behavioural therapy [29]. Pain is also poorly managed in HF patients, with relatively low use of analgesics or opioids [22].
Second, the 3 common non-cardiovascular comorbidities (COPD, diabetes, and CKD) were associated with the highest overall symptom burden and with more severe symptoms than were associated with cardiovascular comorbidities or with no comorbidities. The predominant symptoms associated with COPD, diabetes, and CKD were SOB and fatigue, although COPD was associated with all 4 symptoms. Explanations for the increase in cardiovascular-related symptoms with non-cardiovascular diseases are multiple and varied. It is likely that comorbidities interact with the HF via conflicts in medications, efficacy of interventions, patient self-care, or shared risk factors such as obesity and reduced exercise [30].
The implications of both these findings are important. The symptom focus in HF guidelines and clinical practice is usually on SOB, leg swelling, and fatigue [6]. These symptoms collectively drive clinical management, which focuses on cardiovascular status and associated optimisation of therapies, particularly for HF with reduced ejection fraction [31]. Whilst the management of comorbidities is prioritised for HF patients with preserved ejection fraction, this likely reflects the failure of the numerous Phase II and III trials to show any convincing evidence of benefit from standard HF treatments in this group [32]. So, despite the emphasis on comorbidities for HF patients with preserved ejection fraction, symptoms are not specifically addressed in guidelines, and the comorbidities included are poorly aligned with symptom status.
Our findings show that a potential mismatch exists between these guidelines and patient-rated health, with increased SOB and fatigue being driven by non-cardiovascular status and pain and anxiety or depression being driven by cardiovascular status. This indicates that, for HF patients’ health to improve, new interventions for common HF symptoms need to include the most prevalent non-cardiovascular comorbidities and that management of pain and anxiety or depression needs to become part of routine guideline-driven care and would be an important addition to clinically relevant end-points in clinical trials.
Our third key finding was that although HF optimising therapies were prescribed for most SHFR patients, with minor exception, these therapies were not associated directly or indirectly with comorbidities, symptoms, functional limitations, or patient-rated health. Previous evidence has found only modest benefit from HF pharmacological treatment for quality of life in HF [33], but we also found a poor relationship between therapies and both symptoms and functional limitations. Given the strong associations between patient-rated health and outcomes in HF [3,4], this raises the question of whether HF management needs to be more precise and tailored to the individual patient’s comorbidity and related symptom status. HF patients are older (median age 74 years) and experience multiple conditions and a wide range of symptoms, as indicated by our study, which means that novel multi-condition and multi-disciplinary approaches to care will be required to improve their prognosis and health [34]. However, such structured or multi-disciplinary care is not part of current guideline recommendations [22].
The SHFR includes patients from most (>85%) hospitals in Sweden as well as primary care units and covers the full range of patients with preserved and reduced ejection fraction as well as those with comorbidities, who are often excluded from clinical trials. The availability of such data meant that an a priori HF health model could be hypothesized and tested to provide a detailed concept of the potential pathways between comorbidities and patient-rated health in HF. It is important to note that health measures were only available for a quarter of all patients entered into the register. Whilst the SHFR was created in 2005, EQ-5D has only been included as part of baseline registrations since February 2008. There has been a slow increase since then in the proportion of records completed, particularly in the inpatient setting. Consequently, our sample was predominantly based in the outpatient setting, and patients were slightly younger, more often men, more often married, with a reduced ejected fraction, and more frequently prescribed HF optimising medications. This could imply an underestimation of poor patient-rated health in our study, as older age groups, women, and patients in hospital reported more symptoms than their younger, male, and outpatient counterparts. The impact of such differences on the complex relationships between comorbidities and health is difficult to estimate; we took account of these factors in the model, but we cannot rule out unmeasured confounding.
Whilst patient-rated health in our study was based on EQ-VAS, which is a simple, valid, and reliable measure of patient’s own health, comorbidities and symptoms were based on clinical recording, which can be subject to misclassification, leading to under-ascertainment. That said, any such misclassification is likely to bias the associations towards the null value. We used sequences of regressions for our analyses, meaning that direct and indirect associations between variables could be exposed that are often hidden by conventional statistical analyses. However, as with any method of analysis, the validity of any model depends on having considered the most influential explanatory variables at the outset. The study was retrospective and observational, with few missing data, but an important limitation was the lack of some relevant variables that might impact health such as social status/deprivation, emotional, and spiritual measures. However, our intention was to uncover the indirect relations between comorbidities and patient-rated health through subjective patient-reported health measures as opposed to producing a definitive HRQoL model, and we did use a range of clinical, patient, environmental, and biological variables based on an a priori hypothesized model. Although the investigation used a cross-sectional design, it provides novel and original findings for a HF health model that can be further tested in prospective studies and externally validated in different countries to assess the consistency of relationships postulated by the model.
In HF, distinct health pathways exist among cardiovascular and non-cardiovascular comorbidities that are not influenced by HF optimising therapies. Our HF health model reveals the potential interplay of different factors that underpin health in HF. These findings highlight the need to refocus on person-centred HF care and consider the specific mechanisms that contribute to overall patient health, and to design effective interventions that target the comorbidities and the range of symptoms with the biggest impact on health.
|
10.1371/journal.pgen.1005276 | The DnaA Protein Is Not the Limiting Factor for Initiation of Replication in Escherichia coli | The bacterial replication cycle is driven by the DnaA protein which cycles between the active ATP-bound form and the inactive ADP-bound form. It has been suggested that DnaA also is the main controller of initiation frequency. Initiation is thought to occur when enough ATP-DnaA has accumulated. In this work we have performed cell cycle analysis of cells that contain a surplus of ATP-DnaA and asked whether initiation then occurs earlier. It does not. Cells with more than a 50% increase in the concentration of ATP-DnaA showed no changes in timing of replication. We suggest that although ATP-DnaA is the main actor in initiation of replication, its accumulation does not control the time of initiation. ATP-DnaA is the motor that drives the initiation process, but other factors will be required for the exact timing of initiation in response to the cell’s environment. We also investigated the in vivo roles of datA dependent DnaA inactivation (DDAH) and the DnaA-binding protein DiaA. Loss of DDAH affected the cell cycle machinery only during slow growth and made it sensitive to the concentration of DiaA protein. The result indicates that compromised cell cycle machines perform in a less robust manner.
| Cell cycle regulation of the bacterium Escherichia coli has been studied for many years, and its understanding is complicated by the fact that overlapping replication cycles occur during growth in rich media. Under such conditions cells initiate several copies of the chromosome. The active form of the CDC6-like DnaA protein is required for initiation of synchronous and well-timed replication cycles and is in a sense the motor of the cell cycle machine. It has long been debated whether it is the accumulation of enough ATP-DnaA that triggers initiation and determines the replication frequency. In this work we have constructed a strain where the “accumulation of ATP-DnaA triggers initiation” model could be tested. Our results indicate that this model requires some modification. We suggest that cell cycle regulation in E. coli has similarities to that of eukaryotes in that origins are “licensed” to initiate by a cell cycle motor and that the precise timing depends on other signaling.
| The ORC- and CDC6-like prokaryotic initiator protein DnaA has been studied extensively for many years, but it is still not clear whether the protein contributes to actual regulation of the initiation of replication or whether it works as a cell cycle motor which “licenses” initiation at regular intervals. In E.coli the DnaA protein causes strand opening and recruits the helicase and is thus the key contributor to initiation of replication [1,2]. The DnaA protein, bound to ATP or ADP [3], binds to specific DnaA binding sites within the origin [4–6]. High-affinity binding sites can bind both forms of the DnaA protein [3–6] while low-affinity sites bind only the ATP-bound form [7]. The high-affinity boxes are most likely bound by DnaA throughout the cell cycle [8], while binding to the “last” low-affinity sites has been suggested to trigger the initiation process at a time when the ATP-DnaA level has reached a threshold concentration [9]. Formation of a DnaA oligomer in the origin region causes the unwinding of the DNA in the AT-rich region and formation of the open complex [1,3]. This process is probably facilitated by transcription by RNA polymerase [10–13] and by DiaA, a DnaA-binding protein that has been shown to promote formation of ATP-DnaA complexes at oriC and stimulate oriC unwinding in vitro [14–16].
The DnaA protein also has a role as a transcription factor regulating its own transcription [17–20] and the transcription from several other promoters (see [21] for review) some of which are located close to or within the origin region [22]. More recently it was shown to interact directly with the RNA polymerase and to affect the transcription from the gidA promoter, which is situated right next to the origin [23].
The datA site is a 1 kb DNA sequence with five well conserved DnaA-boxes [24] and several weak DnaA-boxes [25]. The datA region has been thought to bind a large amount of the DnaA protein [24,26], and thereby contribute to titrate the DnaA protein away from the origin. However, recently it was shown that datA, together with the IHF protein, has the ability to stimulate the hydrolysis of the DnaA-bound ATP and thereby inactivate the DnaA protein in a process called datA dependent inactivation of DnaA (DDAH) [27]. The level of ATP-DnaA is also affected by the RIDA (Regulatory Inactivation of DnaA) process, where the Hda protein together with the β-clamp of the polymerase stimulates the hydrolysis of the ATP bound to DnaA [28]. Mutations which block RIDA are lethal because they lead to massive over-initiation [29,30] whereas deletion of datA has minor impact on cell growth [26,31] indicating that RIDA is the more important of the two DnaA inactivation systems. De novo synthesis, DARS (DnaA Reactivating Sequence) sites and possibly acidic phospholipids contribute to the regeneration of the active ATP-bound form of the DnaA protein (see [32] for review).
In several earlier studies with over-expression of the DnaA protein, it was shown that a surplus of DnaA in the cells led to excess initiations and reduced initiation mass. It was therefore concluded that the DnaA protein was the factor limiting the initiation frequency [33–38]. However, in many cases the increase in the DNA/mass after the overproduction of DnaA was low [33,35,38,39] or no increase was observed at all [40], results that contradict this conclusion. It has also been shown that cells grown under different conditions can initiate with widely different amounts of DnaA available per origin [41] and that cells grown in minimal medium supplemented with acetate have more DnaA available per origin than cells grown in richer media [23]. Recently it was shown that cells with a reduced cellular amount of DnaA protein (caused by an elevated level of SeqA, which negatively affects the dnaA transcription) have no problem initiating replication [42]. In addition to this it has been shown that the presence of additional copies of oriC on a high copy number plasmid, which would titrate DnaA away from the chromosomal origin, does not change the timing of initiation [43]. Thus, it is hard to understand that the total amount of DnaA is limiting for initiation of replication. Instead, it is more reasonable to assume that even though enough DnaA must be present at the time of initiation, the accumulation of DnaA alone does not trigger the initiation of replication.
In an attempt to clarify the role of the DnaA protein in the timing of initiation of replication we have performed flow cytometry analysis of cells containing one or several extra copies of the dnaA gene and cells with minor changes in the cell cycle machinery. We found that a two-fold elevation in the DnaA concentration does not affect the timing of initiation of replication in any of the growth conditions tested. This argues that the DnaA protein, although it constitutes the key feature of the cell cycle motor, is not the limiting factor for precise timing of initiation of replication.
In order to investigate the effects of a moderate increase in the DnaA concentration on initiation of DNA replication, cells with one extra copy of the dnaA gene under control of its own promoter, leading to a two-fold increase in the DnaA concentration (Table 1 (IF72) and S1 Fig) were analyzed and compared to the parent cells (Fig 1). The cells were grown in minimal medium supplemented with acetate, glucose or glucose and casamino acids (GluCAA), which led to generation times of about 4 hours, 70 minutes and 30 minutes, respectively (S1 Table).
DNA and mass distributions of the cells were obtained by flow cytometry analysis, and were found to give similar results for cells with wild type and two-fold extra DnaA in all three media (Fig 1 and S1 Table). When all cells in a population grow exponentially (with the same generation time and the same replication pattern), it is possible to use the information in the DNA histograms to calculate the cell-cycle parameters (initiation age and replication period). This calculation was done in an excel based simulation program [44] (Fig 1 and S1 Table).
Wild type cells growing slowly in acetate medium initiated at one origin and had two replication forks in the replication period. The fraction of replicating cells (the population of cells containing between one and two chromosome equivalents in the DNA histogram (Fig 1A, middle panels)) was 25–29%. This yielded a replication period of about 80 min (red arrow in Fig 1A, leftmost panel). Cells growing more rapidly in glucose medium were found to initiate replication at two origins in the mother cell before cell division (Fig 1B, leftmost panel) and cells contained through the cell cycle either two whole chromosomes or one or two partially replicated chromosomes (Fig 1B, middle panels). Cells growing faster still, in GluCAA medium, exhibited overlapping replication cycles and initiated replication in the “grandmother” generation at four origins (Fig 1C, leftmost panel). These cells had DNA contents that ranged from three to six chromosome equivalents (Fig 1C, middle panels). To aid the determination of the cell cycle parameters a part of the cell cultures grown in glucose and GluCAA medium were treated with rifampicin and cephalexin to obtain so-called replication run-out DNA histograms (see Materials and Methods). In the drug treated cells (Figs 1B and 1C, small histograms) all ongoing replication was allowed to finish, but new initiations and cell division were inhibited. The number of fully replicated chromosomes then represents the number of origins present in the cells at the time of drug addiction [45]. All origins in a cell are initiated in synchrony and therefore the cells will end up with numbers of chromosomes which are exponential multiples of two (2n) after drug treatment [46]. If control of initiation is compromised this can often be seen as asynchronous initiations, i.e. a chromosome number different from 2n. Here, the cells with two-fold extra DnaA were found to contain the same number of origins as the wild type cells and initiation occurred in synchrony (Figs 1B and 1C, small histograms). Also the average age at initiation, duration of the replication period, average cell mass and the DNA concentration were found to be essentially the same as in wild type cells in all media tested (Fig 1, rightmost panels, S1 Table and S2 Fig). These results indicate that the amount of DnaA in the cell is not likely to be the factor limiting initiation of replication under the growth conditions tested.
It has previously been shown that a large excess of DnaA causes more frequent initiation (see below), and that some replication forks collapse and fail to extend beyond the immediate area flanking oriC [47]. To check that this does not occur in our situation with two-fold extra DnaA, we performed marker frequency analysis of the oriC and Ter region. If initiation had occurred but a failure of forks to extend beyond the immediate area flanking oriC was a problem, this would have shown up as an increase in the oriC/ter ratio in the constructed strain compared to the wild type. It does not (S6 Table). Thus, we conclude from our results that there are no extra initiations in the cells with two-fold extra DnaA.
It has been shown that only the ATP-form of DnaA is active and capable of performing strand opening in vitro [48]. Therefore, one explanation of the above result could be that the amount of ATP-DnaA is unchanged in the cells with a two-fold DnaA concentration, i.e. that the ratio of ATP- to ADP-DnaA is much lower than normal. We therefore investigated the ratio of ATP-DnaA to ADP-DnaA in the wild type cells and the cells with two-fold concentration of DnaA. We then found that the ratios were about the same in the two strains. This means that there is an increase of both forms of the DnaA protein and that the level of ATP-DnaA is indeed significantly higher in the cells with a two-fold DnaA concentration (Table 2 and S3 Fig). Thus, the result indicates that in cells with a substantial surplus of ATP-DnaA, timing of initiation is controlled in the same way as in wild type cells, and that the controller is not the amount of ATP-DnaA.
Before accepting the above result we found that it would be important to check that other factors were not “compensating”, i.e. that the production of extra DnaA did not lead to two changes in timing of replication which cancelled each other. We therefore checked the expression of genes possibly affected by the change in the concentration of DnaA. The dnaN gene, encoding the β clamp of the DNA polymerase, is located in the same operon as the dnaA gene and its expression is therefore, at least partially, dependent on the dnaA promoter [49]. Because this promoter is auto regulated by the DnaA protein there was a possibility that the extra DnaA in the cells repressed this promoter leading to a decreased level of the β clamp. If so, it could be an explanation of the lack of effect of the two-fold extra DnaA. We therefore quantified the level of β clamp in the wild type cells and cells with two-fold extra DnaA by western blotting. No changes in the level of β clamp were detected (S2 Table). We also checked whether the two-fold extra DnaA would give an effect if accompanied by extra β clamp protein. Wild type cells and cells with two-fold extra DnaA were transformed with a plasmid containing the dnaN gene under control of an inducible promoter. Induction with a low level (30μM) IPTG led to levels of DnaN that were between 40 and 80% above wild type level. When these cells were analyzed using flow cytometry no significant changes in the initiation age or the C-period were observed in either wild type or the cells with two-fold extra DnaA (S3 Table). We therefore conclude that the lack of change in the timing of initiation in the cells with two-fold extra DnaA was not due to a reduced level of β clamp.
Since the DnaA protein has also been shown to regulate several other genes in addition to its own and dnaN (see [21] for review), we also performed RNA sequencing to investigate differences in the transcription between the wild type cells and the cells with two-fold extra DnaA (see Materials and Methods). This generated a list of 4692 transcripts (S7 Table) out of which four genes were differentially expressed in the two strains (Table 3). One of these was the DnaA protein itself confirming the elevated transcription of this gene. This also functions as a positive control of the sequencing experiment. Two of the other genes that were up-regulated in the cells with two-fold extra DnaA are situated right next to the λ att site where the extra copy of the dnaA is inserted on the chromosome (See Materials and Methods). Thus, the increased transcription of these two genes is likely to be an effect of the increased local transcription in their surroundings rather than a result of the extra DnaA in the cell. The fourth gene, pka, which has a slightly elevated transcription level encodes an protein lysine acetyltransferase. The roles of acetylation in prokaryotes are not very well known, but one function is apparently to make the cells more tolerant towards environmental stress [50]. We think that it is unlikely that the slight elevation in the level of this protein affects the timing of initiation of replication in the cells with two-fold extra DnaA. The RNA sequencing also confirmed the unchanged level of the dnaN, encoding the β clamp of the DNA polymerase, as well as other genes that might influence the initiation process such as mioC, gidA, seqA and nrdAB [21].
Previous work concerning the effect of extra DnaA on the timing of initiation of replication was mainly performed with plasmids containing the dnaA gene under control of inducible promoters, presumably leading to quite high DnaA concentrations [33–35,38]. To see the effect of a higher concentration of DnaA, but to avoid complications with a burst of production after induction, cells containing a plasmid (pACYC184) bearing the dnaA gene under control of its own promoter were analyzed. These cells had a DnaA concentration that was on average 35, 10 and 11 times higher than the wild type cells when grown in acetate, glucose and GluCAA medium, respectively (Table 1 (MOR90)). We expect that large overproduction of DnaA leads to heterogeneity in growth parameters in the cell population and it has been shown that a large excess of DnaA leads to replication fork collapse [47]. Therefore the conditions for proper cell cycle analysis were not present and accordingly the cell cycle parameters were not simulated for these cells. The DNA contents per cell were found to be higher than normal in the cells with a large excess of DnaA (Fig 2 and S4 Table). For the cells grown in GluCAA medium the increase in DNA/mass was quite low (5%) (Fig 2D and S4 Table). This and the high degree of similarity of the exponential histograms of the wild type and DnaA overproducing cells (Fig 2C) indicated that the amount of over-replication was quite modest. The difference between over-initiation and over-replication may be explained by the disintegration of some replication forks. It has been shown in cells with a large excess of DnaA that some of the replication forks collapse shortly after initiation [47]. Thus, the low increase in the DNA concentration of the exponentially growing cells after a large overproduction of the DnaA might partly be due to such replication fork collapse. For the replication run-out histograms larger differences were found (Fig 2, rightmost panels). The DnaA overproducing cells yielded peaks at higher chromosome equivalents than what would be expected from the corresponding DNA distributions of the exponentially growing cells. It has previously been shown that if cells contain a surplus of DnaA, initiations might sometimes occur during the rifampicin treatment [23,31,34]. Some of the asynchrony observed here is also probably due to such rifampicin resistant initiations.
The above results indicate that when a massive excess of DnaA is present, the otherwise quite robust cell cycle machine breaks down. A similar situation is seen if the RIDA system is inactivated causing massive over-initiation and lethality. We wished to investigate a situation where a less important part of the DnaA activity control, DDAH, was missing. To do so, we investigated cells where the datA site had been deleted.
We found that a deletion of the datA site did not lead to significant changes in the cellular DnaA concentration (Table 1 (MOR177)). When grown in acetate medium, the ΔdatA cells had about the same doubling time (S5 Table), cell mass and DNA concentration as wild type cells (Fig 3A, rightmost panel and S5 Table). However, the number of cells containing one chromosome was much lower for the ΔdatA cells compared to the wild type (Fig 3A). This means that the ΔdatA cells initiate at a lower age. The reduction in initiation age was found to be about 75% and was accompanied by an increase in the length of the replication period compared to the wild type cells (Fig 3A, rightmost panel and S5 Table). The wild type and the ΔdatA cells had about the same average cell mass. Thus, the initiating ΔdatA cells must also be smaller than the initiating wild type cells.
Also in cells grown in glucose initiation occurred earlier in the cell cycle in the ΔdatA cells compared to the wild type cells, but the effect was not as pronounced as in the cells grown in acetate medium (Fig 3B). Also in this case, early initiation was accompanied by an increase in the duration of the replication period (Fig 3B, rightmost panel and S5 Table).
It has previously been shown that the replication pattern and oriC/ter ratio are not changed by deletion of the datA site in cells grown in GluCAA medium [31]. Because these experiments were performed in medium lacking uridine, which affects the replication pattern of MG1655 [51], we investigated this for cells growing in GluCAA medium containing uridine. Also in this medium the DNA histogram of the exponentially growing cells was the same with and without the datA site (Fig 3C). The result shows that the cells grown in GluCAA medium initiate replication at the same time in the cell cycle and that the length of the replication period is the same irrespective of whether the datA site is present or not (Fig 3C, rightmost panel and S5 Table). However, a difference is seen in the run-out DNA histogram (Fig 3C, small histograms), which shows asynchronous initiations. This phenotype has also been observed previously and represents initiations occurring during rifampicin treatment [31].
DiaA is a DnaA interacting protein which has previously been proposed to have both a positive and negative influence on the initiation process [14–16]. To investigate the effect of DiaA with respect to the timing of initiation we studied cells transformed with plasmids carrying the diaA gene under control of its own promoter in wild type cells and in combination with a large excess of DnaA and deletion of datA. The strains containing the diaA plasmid had a DiaA concentration that was about 5 times higher than the wild type in acetate medium and 4 times higher than the wild type in GluCAA medium (Table 4).
In wild type cells grown in acetate or GluCAA medium an elevated level of DiaA did not lead to any significant changes in the timing of initiation (S5 Fig). However, in the ΔdatA cells grown in acetate, which have a lower initiation age compared to wild type cells, the presence of extra DiaA led to a reversal of the phenotype (Fig 4A). In this situation, we observed an increase in the initiation age in the ΔdatA cells when they in addition expressed extra DiaA (Fig 4A, rightmost panel). This indicates a possible inhibitory role for DiaA. Also in rapidly growing cells, an inhibitory effect of extra DiaA was seen in the ΔdatA cells. These cells exhibit an asynchrony and over-initiation phenotype as a result of rifampicin-resistant initiations (Fig 3C and 4B) [31]. We observed that the number of rifampicin-resistant initiations was reduced when the ΔdatA cells also had a 4 times higher concentration of DiaA (Fig 4B). Both results support the idea of an inhibitory role for the DiaA protein at the origin.
It has previously been shown that a diaA deletion strain had a somewhat increased concentration of DnaA [14], so we checked whether overproduction of DiaA led to a reduction in the DnaA concentration in the cells. However, we found that the DnaA concentration was not reduced in the cells with an extra supply of DiaA (Table 1 (IF97)).
We show here that extra DnaA does not lead to a shift in the timing of initiation of replication in any of the growth media tested. We found that cells were insensitive to a more than 50% increase in the concentration of ATP-DnaA. Thus, the result indicates that in cells with a substantial surplus of ATP-DnaA, timing of initiation is controlled in the same way as in wild type cells, and that the controller is not the amount of ATP-DnaA. The amount of DnaA in the cell could still be an important parameter that, for instance, helps couple replication rate to growth rate [52], but it does not seem to determine the precise timing of initiation during steady-state growth in the media tested here. The idea that DnaA is not the limiting factor for initiation of replication is also supported by several other lines of evidence. Previously we have shown that the amount of DnaA per origin in the cells varies with the growth medium /growth rate and that there is more DnaA per origin in cells grown in acetate medium compared to cells grown in richer media [23]. It has also been shown that for cells growing slowly with about the same doubling time, the amount of DnaA required per origin at the time of initiation in chemostat-grown cells is considerably lower than the amount required in batch-grown cells [41]. In cells with mini-chromosomes, and therefore many extra origins to initiate, initiation occurs at the same time and mass in the cell cycle [53]. Support for this idea also comes from a recent study showing that the timing of initiation was unchanged in cells with a 20% decrease in the level of DnaA compared to the wild type cells [42]. These and our results indicate that DnaA is not the limiting factor for initiation of replication during steady-state growth.
That the DnaA protein does not regulate timing of initiation is also apparent in Bacillus subtilis. B. subtilis mutant cells which are abnormally small because of an aberrancy in the regulation of cell division were found to initiate replication at the same cell age as wild type cells [54]. This means that these cells initiate replication at a smaller size, with less DnaA protein available, compared to the wild type cells, which again means that the wild type cells were not limited by the amount of DnaA. The result instead implies that there was sufficient DnaA for initiation to occur and that a different signal must have decided when initiation occurred. Interestingly, small E. coli cells with a similar aberrancy in regulation of cell division [55] were found to behave differently and did not keep the replication pattern irrespective of cell size, but instead kept the cell size at initiation and not the replication pattern [54]. In other words, initiation of replication occurred in older cells in the abnormally small E. coli cells. The authors induced a transient increase in the amount of DnaA and found a change after run-out of replication. This result might be taken to indicate that cells must accumulate a certain amount of DnaA before initiation could occur. However, with a transient induction, i.e. an unbalanced situation, it is difficult to interpret how the regulation works. We find it likely that the regulatory circuits in B. subtilis and E. coli working on the initiation machinery simply are different, and that in neither case does the point in the cell cycle of reaching enough ATP-DnaA decide the time of initiation.
Taken together, the results indicate that although the DnaA protein is highly conserved through evolution both as an initiator and a transcription factor [53] the precise regulatory circuits that govern its activity may not be conserved. This is perhaps not surprising since B. subtilis and E. coli are very distantly related.
Our results do not fit with the current model for regulation, which assumes that initiation occurs as soon as a certain amount of active DnaA is available [9,56]. This model is based on previous findings where overproduction of the DnaA protein was reported to lead to over-initiation and it was therefore concluded that DnaA was the limiting factor for initiation of replication [33–38]. However, many of these experiments were performed with quite high copy-number plasmids carrying the dnaA gene under control of different inducible promoters. This probably led to levels of DnaA that were quite high compared to the level in wild type cells. Also in our work we found that over-initiation occurred when a large surplus of DnaA was present. Thus, it seems clear both from our results and from previous studies, that a large excess of DnaA is capable of initiating origins that would not otherwise initiate. However, these initiations represent a break-down of the cell cycle machine. They are unregulated and probably represent a forced reaction, which is due to the large excess of DnaA protein. Also, the over-initiation appears to be higher than the over-replication because the increase in the DNA/mass was found to be quite modest, especially in the rapidly growing cells (10%). This is also the case in many of the previous studies where the observed increase was only up to about 20% [33,35,38,39]. It has been shown that many of the replication forks collapse in cells with a large excess of DnaA [47]. Thus, the low increase in the DNA concentration after a large overproduction of the DnaA might be due to replication fork collapse. Also, overproduction of DnaA has previously been reported to lead to initiations occurring during the incubation with rifampicin [31,34], a fact that was not known in some of the earlier studies. Therefore, an increase in the number of origins per cell was in many cases interpreted as over-initiation [35,37,38], when it in fact could be a result of rifampicin-resistant initiations.
Our results are not in accordance with one earlier study in which a substantial increase in the DNA/mass was seen after moderate overproduction of DnaA [34]. The reason for this discrepancy is not known, but one difference between our work and this work is that it was Salmonella typhimurium and not Escherichia coli DnaA protein that was expressed. Another difference is that in the previous work the extra DnaA was expressed form an IPTG inducible promoter which necessarily leads to transient changes in the DnaA level rather than a steady state expression as in our case.
Previously it has been shown that during rapid growth in GluCAA medium ΔdatA cells exhibit rifampicin resistant initiations, i.e. initiations occur during the incubation with rifampicin. No changes are seen in the cell cycle of exponentially growing cells [31]. Our work confirms this result, but also shows that removal of the datA site in cells growing more slowly in glucose or acetate medium has a different effect compared to in the rapidly growing cells. Under these growth conditions, a change in the cell cycle was seen also in the exponentially growing cells; the ΔdatA cells initiated earlier in the cell cycle (as smaller cells) compared to the wild type cells. These results demonstrate that the regulatory influence of DDAH on the cell cycle machinery changes with the growth conditions.
The DDAH system (the binding of DnaA and IHF to datA) causes conversion of ATP-DnaA to ADP-DnaA [27]. This means that the cell has two systems for inactivating the active form of the DnaA protein; the RIDA (Regulatory Inactivation of DnaA) system and the more recently discovered DDAH system. In our cells with two-fold extra DnaA the RIDA and DDAH systems ensure that about 60% of the total DnaA is in the ADP-form so that although there is an elevated level of ATP-DnaA compared to the wild type cells there is still a balance in the cell between the two forms of the DnaA protein. Mutations which block the RIDA system are lethal due to over-initiation [29,30]. This shows that the cell cycle machinery is dependent on conversion of ATP-DnaA to ADP-DnaA in order to work. That the cell cycle can work when datA is deleted (i.e. without DDAH) indicates that RIDA is the more important of the two systems.
A difference between the DDAH and RIDA systems is that the RIDA system is dependent on ongoing replication whereas DDAH is not [57]. In rapidly growing cells with overlapping replication cycles there are always active replication forks in the cell which means that the RIDA system will always be active. Thus, the DDAH system may be of less importance in taking down the level of ATP-form DnaA in rapidly growing compared to slowly growing cells. In accordance with this assumption, major changes in the level of ATP-DnaA after deletion of the datA site was not found in rapidly growing cells [58]. However, in the acetate grown cells only around 25% of the cells contain active replication forks while the rest of the cells in the population are either in the B- or the D-period where no replication occurs. It might therefore be that the DnaA-inactivating activity of DDAH becomes more important under such conditions. It is possible that loss of DDAH in slowly growing cells leads to a shift in the balance between the ATP-form and ADP-form of DnaA that is larger than what we get with two-fold extra DnaA (where still a lot of this is in the ADP-form). This unbalance in the ratio of ATP-DnaA and ADP-DnaA could lead to premature initiations similar to the situation with a large surplus of DnaA.
The DiaA protein has been shown to affect the initiation process both positively and negatively in vitro [14–16], and it has been suggested that DiaA might have a dual role in the initiation process. First, a stimulatory role early in the initiation process where it aids in the recruitment of the DnaA to the origin and in formation of the open complex, and later an inhibitory role where it inhibits too early loading of the DnaB helicase and the rest of the replication machinery [16]. How the transition between the DiaA-bound inactive complex and the DnaB-bound active replicative complex might occur is not known. We show in this work that over-production of DiaA does not have any effect in an otherwise wild type situation. This result indicates that the transition to the DnaB-bound active replicative complex is not simply a question of a competition between DiaA and DnaB for binding to DnaA, as one would then expect initiation to be delayed when a surplus of DiaA is present. No such delay was observed. It might therefore be an active mechanism or a signal that leads to release of DiaA from its binding site, allowing DnaB to bind.
In contrast to the wild type cells, we did see an effect of extra DiaA in cells lacking the DDAH system where the too early and rifampicin resistant initiations were reduced by expression of extra DiaA. This indicates an inhibitory role for DiaA under these circumstances and shows that in cells with a less robust or a compromised cell cycle additional changes in the levels of regulators/components of the cell cycle, such as DiaA, is more likely to have an effect. These results support the previously suggested inhibitory role for the DiaA protein [16] and a possible explanation of the results is that the extra DiaA in the ΔdatA cells inhibits premature and rifampicin resistant initiations by inhibiting the loading of the DnaB helicase.
Regulation of initiation must fulfill two requirements. It must prevent extra initiation events, and it must ensure sufficient initiation so that one initiation event occurs per generation per origin. Several mechanisms have been discovered that ensure that extra initiation events do not occur (origin sequestration, inactivation of DnaA, inhibition of dnaA transcription), but less is known about the timing of replication initiation, i.e. the rate limiting steps, and whether the same factor(s) are required under all conditions. The frequency of replication must match the growth rate, otherwise the cellular DNA concentration will be altered. We propose that i) E. coli has a robust replication cycle driven by the cycling of the levels of ATP-DnaA and ADP-DnaA, ii) initiation cannot occur unless sufficient amounts of ATP-DnaA at oriC has licenced this event and iii) signals depending on the cell’s environment govern the exact timing of initiation.
All strains used are Escherichia coli K-12 and are listed in Table 5. Cells were grown in AB minimal medium [59] supplemented with 10 μg/ml thiamine, 25 μg/ml uridine and either 0.4% sodium acetate, 0.4% glucose or 0.4% glucose and 0.5% casamino acids at 30°C (Acetate and glucose medium) or 37°C (GluCAA medium). For determination of the ATP-DnaA to ADP-DnaA ratio, cells were grown in low phosphate medium (see below). The growth rates were determined by measuring the optical density of the cultures at 450 nm.
IF72 was made by amplification of the dnaA gene including the promoter region with the primers 5`-CGAGGATCCTTACGATGACAATGTTCTG and 5`-CGGAGCTCGGCTTTATTGGATATCCG. This fragment was ligated into the vector pAH150 and the resulting plasmid pAH150_dnaA was integrated into the chromosome at the λ att site as described in [60]. After construction both copies of the dnaA gene in IF72 were sequenced to ensure that no mutation had been introduced.
The pACYC184_dnaA plasmid was constructed in previous work [61]. The plasmid pACYC184_diaA was made by amplifying the diaA gene with the upstream promoter region (derived from [14]) with the primers 5`GCACTGCAGGTTAACCACCAAACAGAC and 5`CGAGGATCCTTAATCATCCTGGTGAGG followed by sub-cloning in the pGEM-T-Easy vector (Promega) and ligation of the gene fragment into the EcoRI site of pACYC184.
The pFH2102_dnaN plasmid was made by amplifying the dnaN gene with the primers 5`GGCGGATCCATGAAATTTACCGTAGAAGCTGAG and 5`GGCGAATTCTTACAGTCTCATTGGCATGACAAC and ligation of the gene fragment into the BamHI and EcoRI sites of pFH2102. The two plasmids were then transformed into MG1655 and IF72 by electroporation.
Exponentially growing cells (OD ~ 0.15) were harvested and fixed in 70% ethanol or treated with 300 μg ml-1 rifampicin and 10 μg ml-1 cephalexin for two or more generations before fixation to inhibit new rounds of initiation and cell division, respectively [46]. Flow cytometry was performed with a LSR II flow cytometer (BD Biosciences). Total protein content in the cells was stained with Fluorescein isothiocyanate (FITC, Sigma-Aldrich) and used to calculate the average mass [62]. The DNA was stained with Hoechst 33258 (Sigma-Aldrich) [41]
The cell cycle parameters (initiation age, replication period) were calculated by combining the data from the flow cytometry analysis, the theoretical age distribution of an exponential culture and the generation time obtained by OD measurements in an excel based simulation program [44]. For cells that have only one round of ongoing replication (cells grown slowly in acetate), the simulation program also provides the percentage of cells found in the respective periods of the cell cycle, i.e. in B-, C- and D-period.
The length of the replication period (C-period) was also calculated with an independent method using the oriC/ter ratios obtained by quantitative PCR and the doubling time (τ) with the formula oriC/ter = 2C/τ (See Supporting information).
Whereas one example histogram is shown in each panel of the figures, the values of cell cycle parameters are the average calculated for samples from several experiments.
Exponentially growing cells (OD ~ 0.15) were harvested by centrifugation and SDS (sodium-dodecyl-sulfate) samples of cell extracts and purified proteins were prepared as previously described [63]. Samples were subjected to 12% SDS-polyacrylamide gel electrophoresis and detection of DnaA was carried out using anti-DnaA-antibody, anti-DnaN-antibody or anti-DiaA-antibody and ECF fluorescence kit (GE Healthcare). Quantification was performed using Image Quant software (Molecular Dynamics).
RNA was isolated using SV Total RNA isolation system (Promega) from MG1655 and IF72 cells grown exponentially (OD ~0.15) in GluCAA medium. The isolated RNA was analyzed on an Agilent Bioanalyzer to confirm quality and depleted for rRNA using Ribo-Zero rRNA Removal Kit for Gram-Negative Bacteria (Illumina).
The rRNA depleted samples were submitted to the Norwegian Sequencing Centre (sequencing.uio.no) where libraries were prepared using TruSeq stranded mRNA reagents (Illumina) according to manufacturer’s instructions, entering the procedure at the RNA fragmentation step with 35 ng rRNA-depleted RNA, and fragmenting for 4 minutes at 94°C. Libraries were sequenced on an Illumina NextSeq-500 instrument with 150 cycle mid-output v1 reagents, according to manufacturer's instructions, employing 75 bp paired-end reads. Image analysis and base calling were performed using Illumina's RTA software version 2.1.3. Reads were filtered to remove those with low base call quality using Illumina's default chastity criteria.
The sequencing was performed on three replicates from each of the two strains.
The sequence reads were received from the Norwegian Sequencing Centre as FASTQ files containing the forward and reverse reads respectively and were analyzed using the tool Rockhopper [64]. The expression level of each transcript is reported using RPKM (Reads Per Kilobase per Million mapped reads), except that instead of dividing by the total number of reads it is divided by the upper quartile of gene expression. To test for differentially expressed genes the software first uses local regression to obtain a smooth estimate of gene expression variances. Then, for each transcript, a statistical test for the null hypothesis, which is that the expression of the transcript is the same in different conditions, is performed. The Negative Binomial distribution is used as the statistical model to compute a p-value indicating the probability of observing a transcript's expression levels in different conditions by chance. Because multiple test are being performed, q-values are reported that control the false discovery rate using the Benjamini-Hochberg procedure [64]. Genes with a q-value ≤0.01 were considered as significantly differentially expressed. To verify differentially expressed genes RT-QPCR was performed (see Supporting information, S8 Table and S9 Table).
Cells grown in TG640-thy-less medium with 0.2% glucose and 40 μg/ml of each amino acid overnight were diluted to OD660 = 0.005 in TG320-thy-less medium and 0.28 mCi of HPO42- (32P) was added to 1 ml of cells (3x1 ml). Subsequently, cells were grown exponentially at 37°C to OD660 = 0.20 and harvested. Cell extracts were made and immuno-precipitated with purified DnaA antiserum as described [65]. To remove unspecific immunoglobulins from the DnaA rabbit antiserum (R22) it was purified by passage through a column containing all cellular proteins except DnaA. The column was made of cyanogen bromide activated Sepharose beads coupled to extract from a dnaA deletion strain.
After immuno-precipitation the nucleotides bound to DnaA (ATP or ADP) were extracted and separated on thin layer chromatography (TLC) in 1M HCOOH containing 0.8M LiCl. Migration of ATP and ADP was determined by mobility of cold ATP and ADP. The amounts of ATP and ADP were quantified in ImageQuant, taking into consideration the different numbers of phosphate groups in ATP and ADP. The percentage ATP from each sample was then calculated.
|
10.1371/journal.pcbi.1003359 | Direct Solution of the Chemical Master Equation Using Quantized Tensor Trains | The Chemical Master Equation (CME) is a cornerstone of stochastic analysis and simulation of models of biochemical reaction networks. Yet direct solutions of the CME have remained elusive. Although several approaches overcome the infinite dimensional nature of the CME through projections or other means, a common feature of proposed approaches is their susceptibility to the curse of dimensionality, i.e. the exponential growth in memory and computational requirements in the number of problem dimensions. We present a novel approach that has the potential to “lift” this curse of dimensionality. The approach is based on the use of the recently proposed Quantized Tensor Train (QTT) formatted numerical linear algebra for the low parametric, numerical representation of tensors. The QTT decomposition admits both, algorithms for basic tensor arithmetics with complexity scaling linearly in the dimension (number of species) and sub-linearly in the mode size (maximum copy number), and a numerical tensor rounding procedure which is stable and quasi-optimal. We show how the CME can be represented in QTT format, then use the exponentially-converging -discontinuous Galerkin discretization in time to reduce the CME evolution problem to a set of QTT-structured linear equations to be solved at each time step using an algorithm based on Density Matrix Renormalization Group (DMRG) methods from quantum chemistry. Our method automatically adapts the “basis” of the solution at every time step guaranteeing that it is large enough to capture the dynamics of interest but no larger than necessary, as this would increase the computational complexity. Our approach is demonstrated by applying it to three different examples from systems biology: independent birth-death process, an example of enzymatic futile cycle, and a stochastic switch model. The numerical results on these examples demonstrate that the proposed QTT method achieves dramatic speedups and several orders of magnitude storage savings over direct approaches.
| Stochastic models of chemical networks are necessary to quantitatively describe random fluctuations and other probabilistic phenomena within living cells. The Chemical Master Equation (CME) describes the time evolution of molecular abundance probabilities in these models, and is a basis for many stochastic simulation and analysis methods. Yet the CME is difficult to solve directly except for very simple structures. Indeed current approaches are susceptible to the curse of dimensionality, that is, the exponential growth of memory and computational requirements in the number of problem dimensions. In this paper, we propose a novel approach that has the potential to overcome the curse of dimensionality. It is based on the use of the recently proposed Quantized Tensor Train (QTT) formatted numerical linear algebra for numerical representation of tensors, using algorithms for basic tensor arithmetics with complexity scaling linearly in the number of reacting species considered, and sub-linearly in the maximum allowed copy number per species. We present this approach and demonstrate its effectiveness by applying it to three problems from systems biology. Numerical experiments are reported which show that several orders of magnitude memory savings is typically afforded by the new approach presented here.
| In spite of the success of continuous-variable deterministic models in describing many biological phenomena, discrete stochastic models are often necessary to describe biological phenomena inside living cells where random motion of reacting species introduces randomness in both the order and timing of biochemical reactions. Such random effects become more pronounced when one factors in the discrete nature of reactants and the fact that they are often found in low copy numbers inside the cell. Manifestations of randomness vary from copy-number fluctuations among genetically identical cells [1] to dramatically different cell fate decisions [2] leading to phenotypic differentiation within a clonal population. Characterizing and quantifying the effect of stochasticity and its role in the function of cells is a central problem in molecular systems biology.
In order to effectively capture this experimentally observed stochasticity, the evolution of the chemical species of interest are commonly modeled using jump Markov processes. Here, each state of the process corresponds to the copy number of one of the constituent species [3]. Within this framework, the evolution of the probability density over the possible configurations of the reaction network is described by a Forward Kolmogorov Equation, frequently referred to as the Chemical Master Equation (CME) within the chemical literature. While analytical solutions can be obtained under specific assumptions about the structure of the chemical network [4], these assumptions prove so restrictive as to exclude the vast majority of biologically relevant systems. In most cases, the CME cannot be solved explicitly and various numerical simulation techniques have been proposed to approximately solve the time-evolution problem.
A first class of methods seeks to compute approximations of the CME solution instead by solving a truncated version of the original Markov process. These methods are advantageous in that they provide explicit error guarantees after simulation. This class includes the finite state projection [5] and sliding window abstraction [6]. In these methods, the truncation is chosen so that both the number of states retained is small enough that it may be computed efficiently but large enough that it retains the majority of the probability mass over the time evolution. Clearly, these two objectives are not complementary. In order to guarantee that the approximation has low error, most biologically relevant reaction networks require truncations with so many states that they are completely intractable on available hardware. The finite buffer method [7], [8] suggests a more sophisticated truncation to the states reachable from a given initial state assuming that only a prespecified finite number of molecules may be spontaneously created. However, its use is limited to explicit time-stepping schemes, in addition to requiring that the finite buffers be large to compute accurate solutions.
A second broad class of methods are the kinetic Monte Carlo approaches which instead seek to produce either exact or approximate realizations of the underlying Markov process [3], [9], [10]. By generating sufficiently many realizations, these methods obtain statistics for events that are biologically important. Unfortunately, in many systems, these important events occur rarely, so that producing enough realizations to estimate these statistics is prohibitive.
A third class of methods use asymptotic approximations to trade accuracy for computational or analytical tractability. This class includes the Moment Closure methods [11], [12], the Linear Noise Approximation (LNA) [13], and Chemical Langevin Equation (CLE) treatments [14], [15]. Each of these methods replaces the discrete description of the population counts with a continuous one and can therefore perform poorly in situations where the discrete dynamics are difficult to capture with continuum models, e.g. when even one of the reacting species exhibits low population count or is constrained to have low population count, for instance, in the presence of conservation laws.
Some of the classes of methods described so far perform well in complementary regimes and recently there has been substantial effort to combine these methods resulting in the so-called hybrid methods. Several methods require a time-scale separation of the dynamics to split the system into fast and slow species and impose a quasi-stationary assumption for the fast reactions. An approximate method which can converge quickly to an accurate approximation of a stationary distribution such as -leaping [16] or the Chemical Langevin Equation [17], [18] is used for the fast species, while the slower but more accurate Gillespie algorithm is used for the slow species. Rather than partitioning the species by time-scales of the associated reactions, other methods separate by average molecule count. The low count species are tracked by kinetic Monte Carlo while an ODE approximation is made for the dynamics of the high count species [19], [20]. While these methods allow faster simulations, speedups come at the cost of accuracy, as modeling errors are introduced by the partial replacement of the CME with cruder descriptions.
In order to provide methods that are both accurate and computationally efficient, several numerical techniques for compressing the dynamics and the solution have been explored in the recent literature. Attempts were made to expand the probability distribution as a linear combination of a small set of so-called “principal”, orthogonal basis functions [21]–[25]. Then, either a Galerkin projection was used to map the dynamics onto the lower dimensional subspace spanned by the basis functions (Method of Lines) or first a time discretization was used and then the basis at each time step was adapted by either adding or subtracting basis elements (Rothe's Method). These methods differ primarily in their choice of orthogonal basis. A common feature of these approaches is that they begin with a basis for probability distributions of a single variable and then use the corresponding tensor product basis for multivariate distributions. This means that they are susceptible to the so-called curse of dimensionality [26], that is, the memory requirements and computational complexity of basic arithmetics grow exponentially in the number of dimensions. In the context of the CME, this means that all of these approaches can exhibit an exponential scaling of the complexity with the number of chemical species in the model.
Recent papers have attempted to address the curse of dimensionality by using a low-parametric representation of tensors known as canonical polyadic decomposition or CANDECOMP/PARAFAC, both notions being subsumed under the acronym CP [27], [28]. CP is a methodology for generalizing the singular value decomposition (SVD) for matrices to tensors of dimension greater than two by representing the solution as sums of rank-one tensors (equivalently, linear combinations of distributions in which species counts are independent at each fixed time point). As long as the tensor rank of the solution to be approximated remains low, these approaches can be very computationally efficient as basic arithmetics for tensors in the CP format scales linearly in the number of tensor dimensions.
A key challenge in applying the CP decomposition to construct approximate CME solvers is to control the tensor rank of the computed solution. Basic algebraic tensor operations such as addition and matrix-vector multiplication generally increase rank and hence computational cost. In [29] it is suggested to recompute a lower rank CP decomposition after every arithmetic operation. This approach turned out to be problematic in practice. One reason is that the problem of tensor approximation (in the Frobenius norm) with a tensor of fixed rank is, in general, ill-posed [30]. Thus, the numerical algorithms for computing an approximate representation may easily fail. Another reason is that the problem is NP-hard [31], [32] and there is no robust algorithm having any affordable complexity.
Another approach [33], related to the present work, attempts to avoid the problem of approximation in the CP format entirely by projecting the dynamics onto a manifold composed of all tensors with a CP decomposition of some predetermined maximal tensor rank. This procedure results in a set of coupled nonlinear differential equations which are then solved using available ODE solvers. While this effectively controls the tensor rank of the approximate solution, still, to the authors' knowledge, there is no way to estimate either theoretically (a priori) or numerically (a posteriori) the CP rank of the full CME solution as a function of given data.
In this paper we propose a new, deterministic computational methodology for the direct numerical solution of the CME, without modelling or asymptotic simplifications. The approach has complexity that scales favorably in terms of the number of different species considered and the maximum allowable copy number of each of these species. It is based on the recently proposed Quantized Tensor Train (QTT) formatted, numerical tensor algebra [34]–[37] which operates on low-parametric, numerical representation of tensors, rather than on their CP representations. This decomposition admits both algorithms for basic tensor arithmetics that scale linearly in the dimension (the species number) and a robust adaptive numerical procedure for the tensor truncation, which is quasi-optimal in the Frobenius norm.
We show in the present paper how the CME can be represented in QTT format, then use -discontinuous Galerkin discretization in time to exploit the time-analyticity of the CME evolution and to reduce the CME evolution problem to a set of QTT structured linear equations that are solved at each time step [38]. We then exploit an algorithm available for solving linear systems in this format that is based on Density Matrix Renormalization Group (DMRG) methods from quantum chemistry.
The numerical experiments reported below (see, in particular, Table 1) show several orders of magnitude memory savings, which is typically afforded by the new approach presented here.
We start our development by formulating the Chemical Master Equation (CME), arising from stochastically reacting chemical species. Then we will devote the remainder of the article for its proposed solution. A “well-stirred” solution of chemically reacting molecules in thermal equilibrium can be described by a jump Markov process, where for each fixed time , is a random vector of nonnegative integers with each component representing the number of molecules of one chemical species present in the system. In [29] and the references therein, it is shown that, given an initial condition , the corresponding probability density function (PDF) of the process solves the Chemical Master Equation (CME):(1)where is the number of reactions in the system, and are the stoichiometric vector and propensity function of the th reaction, respectively. The CME is a system of coupled linear ordinary differential equations with one equation per state .
We briefly outline our proposed methodology for the numerical solution of the CME. Since the state space of solutions is countably infinite, the main challenge to be overcome is the curse of dimensionality. As the state space of the CME is typically countably infinite, there is a countably infinite number of different possible states that could be reached by the chemical system. Our approach consists of employing efficient methods for tensor-structured, rank-adaptive numerical solution of very large but “finite state projection” truncations of the CME. In a nutshell, we are proposing to solve large, coupled systems of linear ODEs with a special, tensor structure inherited from the CME. We now give a general outline of our approach, followed by detailed descriptions of each of these steps.
Munsky and Khammash [5] rewrote the right-hand side of the CME (1) as the action of a linear operator on the probability density at the current time:(2)Throughout this paper, we refer to as the CME operator.
Hegland and Garcke introduced an explicit representation of the CME operator as sums and compositions of a few elementary linear operators [29]: let be the spatial shift of a probability density by a vector and let be multiplication by a real-valued function :Then the CME operator can be written as follows, with denoting the identity operator:(3)To simplify the exposition, we assume that all propensity functions are rank-one separable, i.e. they are of the form(4)for , where each is a nonnegative function in the single variable . Considering rank-one separable propensity functions is sufficient for all elementary reactions which occur as building blocks in more complicated reaction kinetics.
The CME (2) is posed on the (countably) infinite space of states. In this form, the CME (1) is an infinite-dimensional coupled evolution problem which necessitates truncation prior to numerical discretization. In the case of a particular class of monomolecular reactions, Jahnke and Huisinga were able to construct an explicit solution in terms of convolutions of products of Poisson and multinomial distributions [4].
In order to be able to address more complex systems computationally, Munsky and Khammash proposed the Finite State Projection Algorithm (FSP) [5] which seeks to truncate the countably infinite dimensional space of states of the process to its finite subset(5)associated with a multi-index , so that the dynamics over are close to those of the original system; see Theorem 1. In practice, the truncation satisfying a given error tolerance may still require a very large number of states: the dimension of the FSP vector equals rendering a direct numerical solution of even the projected equation (S1.1) infeasible in many cases. The remainder of the paper presents a novel approach for the numerical solution of such FSP truncated systems that retain large numbers of states. For notational convenience, we drop the superscripts and the hat from indicating the FSP since we will only consider systems which have already been truncated. Similarly, we now use the shift and multiplication operators in (3) restricted to the truncated state space without change of notation.
Assuming that a FSP has been performed, we henceforth treat as a -dimensional -vector, i.e. as an array indexed by which we identify with ordered -tuples of indices , where ranges from to . Each dimension (alternatively referred to as a mode or level) has a corresponding mode size , that is, the number of values which the index for that dimension can take. For our chemically reacting system, corresponds to the maximum number of copies of the th species that is considered. For a more detailed introduction to basic tensor operations and terminology see, for example, [39], [40].
For the same ordering of , consider the corresponding d-dimensional -vectors , , containing the values of the propensities on to which we shall refer as propensity vectors:(6)Within the projected CME (S1.1), the operators corresponding to weighting by the propensity functions, involved in (3), are finite matrices: . Then, under the rank one separability assumption (4), with for , there holds(7)
Let us consider the truncated CME (S1.1) with a state space on a finite interval . The Cauchy problem with an initial value reads as find a continuously differentiable function such that(13)The solution to (13) is given theoretically by for , but the straightforward numerical evaluation of the matrix exponential involved is a very challenging task due to the “curse of dimensionality”. Instead, we use the QTT-structured -discontinuous Galerkin (-DG-QTT for short) time-stepping scheme, proposed in [38], to solve (13). The -DG time stepping was proposed earlier in [56] for initial value problems for abstract, possibly non-linear, ODEs. We recapitulate the analysis results from [56] for problems of the particular form (13), which have unique, analytic in time classical solutions. To discuss the tensor structure of the -DG-QTT approach, we revisit [38].
Let us denote by the space of polynomials defined on a finite interval , of degree at most and with coefficients from . Let be a partition of the time interval into subintervals , , and . Consider the spaceof functions, which are polynomials of degree at most on for all . For all let and for all feasible .
Assuming we have a finite state projection of the CME, we summarize our approach to the CME solution by outlining the two main algorithms we propose for its subsequent efficient solution. Given a reaction network and a finite state projection Algorithm 1 (Box 1) approximates the CME operator in QTT format. Algorithm 2 (Box 2) then describes the time-stepping procedure for computing the solution. Note that the integrals in Algorithm 2 may be pre-computed depending on the choice of temporal basis functions. E.g. if one chooses the Legendre polynomials as the basis, then there are explicit solutions of the integrals involved.
The solution at a particular time of a finite state projection of the CME is given analytically by the matrix exponential, but the numerical computation of such solutions for large is often expensive. When is sparse, however, the Krylov subspace method [57], [58] is one approach for performing the computation for the CME as described in [59]. The method uses the Arnoldi iteration to compute the Krylov subspace up to some order of accuracy then computes the matrix exponential in that smaller space (by diagonal Padé approximation). The publicly available Expokit Toolbox by Sidje [60] provides an implementation of the algorithm.
It is important to note that the algorithm steps incrementally in time rather than jumping to the desired time step. In the context of the CME, this means that the faster the support of the pdf fills the set of reachable states, the more expensive this algorithm becomes to compute. When there is reason to believe the support of the pdf remains small, then the algorithm can be expected to compute efficiently over large time intervals. Generically, however,the support of the pdf quickly fills the set of reachable states which may include every state retained in the projection. This renders the Arnoldi iteration computationally expensive at each time step.
The QTT method effectively circumvents this problem by storing the computed solution at each time step in the QTT format and exploiting the fast algorithms for basic tensor arithmetic available in this format. While it is unknown whether a given reaction network and initial probability distribution will produce an evolution that can be represented well by a QTT formatted tensor with low QTT ranks, our numerical experiments find this often is the case and that the savings over using traditional sparse representations of vectors and matrices may be quite substantial.
Below we compare our method to the Krylov subspace approach in the toggle switch example which does not exhibit any pronounced structure favoring either one of the methods (rank-one separability and sparse structure respectively).
We presented a novel, “ab-initio” computational methodology for the direct numerical solution of the CME. The methodology exploits the time-analytic nature of solutions to the CME and the low-rank, tensor structure of the CME operator by combining an -timestepping method that is order and step size adaptive, unconditionally stable and exponentially convergent with respect to the number of time discretization parameters, with novel, tensor-formatted linear algebra techniques for the numerical realization of the method. In particular, after an initial projection on a (sufficiently rich) finite state, the QTT representation allows for the dynamic adaptation of the effective state-space size, as well as of the principal components, or basis elements of the numerical representation of solution vectors in the numerical simulation of the time evolution of the CME solution. We emphasize that, while the performance of our approach is better when the solution can be approximated in the QTT format with a high degree of separability of the “physical” and “virtual” variables (i.e. with low TT ranks), the approach does not require a particular degree of separability, but instead reveals possibly present low TT rank in the solution at runtime. In the course of rank adaptation, the singular vectors, in the span of which the solution is approximated, are also adapted. Hence, the presently proposed approach is superior to fixed basis approaches (even when used with adaptivity), such as those reported in [19], [22], [23], [66]. The precise class of chemical reaction networks that lead to low TT rank in the solution tensor is currently unknown. To the extent that this rank increase during runtime, the effectiveness of the compression will be decreased, which could prove limiting for some problems. However, in this case other methods will be equally challenged. Identifying the architecture of the chemical reaction networks that lead to very low ranks is currently a research problem under investigation.
While the discussion following Theorem 4 relates to the case when the factors of the propensity functions are monomial, the approach presented herein applies equally well to models with propensity functions that are merely smooth enough. For example [67], gives bounds on the QTT ranks of the propensity functions and CME operator in the case of the stochastic mass-action and Michaelis–Menten kinetics with separable propensity functions. Also, the same work proves the bounds on the QTT ranks of product-form stationary distributions [68] of weakly-reversible reaction networks of zero deficiency in the sense of Feinberg [69]. Those bounds explain some of the experimental observations made in the present paper. Furthermore, the approach proposed is suitable for non-separable propensity functions. However, in that case the characterization of the rank structure of the CME operator needs to rely on some extra assumptions ensuring moderate QTT ranks, even though more general than separability, and Algorithm 1 needs to be altered accordingly.
The performance of the approach proposed essentially relies on the efficiency of the numerical solution of TT-structured linear systems of equations. In particular, a globally (or “less strictly locally”) convergent iterative solver would allow us to take larger time steps and to exploit the exponential convergence of the -DG time discretization. We believe that while the presently reported numerical results which were obtained with the DMRG solver are quite encouraging, ongoing research on TT-structured linear system solvers holds the promise for a substantial efficiency increase of the present methodology. We only mention a family of alternating minimal energy methods which was announced very recently in [70].
We also mention that, of course, the choice of the tensor format and, possibly, index ordering, has an essential impact on the performance of the approach. The computational experiments reported in the present paper show that even a straightforward permutation of “virtual” indices produced by quantization may allow to exploit additional structure in the data and the QTT formatted CME solution and, therefore, may improve the performance of the QTT-structured approach dramatically. We point out that the TT format can be considered as a special case of tensor network states: TT formatted tensors belong to the class of simple, rooted tree-type tensor networks. Relating the architecture of the chemical reaction networks and appropriate tensor networks representing its states efficiently, i.e. with low ranks, is currently a research problem under development. The results of [67] mentioned above can be considered as the first step in this direction.
A general discussion of tensor networks and their use in numerical simulations for quantum spin systems can be found in [71], [72]. As for the numerical solution of the CME, particular real-life problems might require more sophisticated tensor networks to be used to efficiently approximate reachable states of the systems in question. The mathematical investigation of the relative merits and drawbacks of tensor formats for particular applications is currently undergoing rather active development; we mention only the recent monograph [40] and the references there.
We finally mention that recently, and independently, TT formatted linear algebra methods for the CME were proposed in [73]; a low order time stepping, and no transposition of tensor trains was used in that work. The CME examples presented in [73] also included a toggle switch, but the authors mostly rely on the intrinsic convergence of their method without analyzing actual accuracies. The latter are reported only for moderate sized examples which are computationally tractable with the direct approach in the full format. However, no attempt is made to analyze the accuracy in comparison to other simulation methods, which are typically applied to larger problems featuring essential difficulties for the direct approach. In the present paper we give comparisons with a state-of-the-art, massively parallel stochastic simulation package. This allows us, on the one hand, to validate the accuracy of the QTT-based solutions obtained here and, on the other hand, to provide evidence of the dramatic increase in efficiency afforded by the new deterministic approach: Monte Carlo simulations on 1500 cores of a high-performance cluster were matched in accuracy and outperformed in the wall-clock time by a MATLAB implementation running on a notebook.
To solve the initial value problem for (2), we exploit the -DG-QTT algorithm proposed in [38] and adapted to the CME as described above, implemented in MATLAB. It uses an implicit, exponentially convergent spectral time discretization of discontinuous Galerkin type. The resulting, time-discrete CME in “species space” is solved in the QTT format. Our implementation relies on the public domain TT Toolbox which provides basic TT-structured operations and solvers for linear systems in the QTT format. The TT toolbox is publicly available at http://spring.inm.ras.ru/osel and http://github.com/oseledets/TT-Toolbox; to be consistent, we use the GitHub version of July 12, 2012 in all examples below. We run the -DG-QTT solver in MATLAB 7.12.0.635 (R2011a) on a laptop with a 2.7 GHz dual-core processor and 4 GB RAM, and report the computational time in seconds.
For the solution of the large, linear systems in the QTT and QT3 formats in each time step, we use the optimization solver, based on the DMRG approach [46]–[48] and elaborated on in the context of the TT format in [74] and available as the function dmrg_solve3 of the TT Toolbox. While the “DMRG” solver still lacks a rigorous theoretical foundation, it proves to be highly efficient in many applications, including our experiments. In [75] a closely related Alternating Least Squares (ALS) approach was mathematically analyzed and shown to converge locally. More on the mathematical ideas behind the ALS and DMRG optimization in the TT format can be found in [76].
The “DMRG” solver, under certain restrictions on the time step, manages to find a parsimonious QTT formatted solution of the linear system (up to a specified tolerance). Moreover, the solver in effect automatically adapts both the QTT rank as well as the QTT “basis” of the solution at every time step guaranteeing that it is sufficiently rich in order to capture the principal dynamics of interest.
In the first numerical example the solution is symmetric and exactly rank-one separable, which allows us to use the standard MATLAB solver ode15 s in the sparse format to obtain the univariate factor of a reference solution. In other examples we used SPSens beta 3.4, a massively parallel package for the stochastic simulation of chemical networks (http://sourceforge.net/projects/spsens/) [77], to construct reference PDFs. The stochastic simulations were carried out on up to cores of Brutus, a high-performance cluster of ETH Zürich (https://www1.ethz.ch/id/services/list/comp_zentral/cluster/index_EN).
|
10.1371/journal.ppat.1003234 | Lymphotoxin Signaling Is Initiated by the Viral Polymerase in HCV-linked Tumorigenesis | Exposure to hepatitis C virus (HCV) typically results in chronic infection that leads to progressive liver disease ranging from mild inflammation to severe fibrosis and cirrhosis as well as primary liver cancer. HCV triggers innate immune signaling within the infected hepatocyte, a first step in mounting of the adaptive response against HCV infection. Persistent inflammation is strongly associated with liver tumorigenesis. The goal of our work was to investigate the initiation of the inflammatory processes triggered by HCV viral proteins in their host cell and their possible link with HCV-related liver cancer. We report a dramatic upregulation of the lymphotoxin signaling pathway and more specifically of lymphotoxin-β in tumors of the FL-N/35 HCV-transgenic mice. Lymphotoxin expression is accompanied by activation of NF-κB, neosynthesis of chemokines and intra-tumoral recruitment of mononuclear cells. Spectacularly, IKKβ inactivation in FL-N/35 mice drastically reduces tumor incidence. Activation of lymphotoxin-β pathway can be reproduced in several cellular models, including the full length replicon and HCV-infected primary human hepatocytes. We have identified NS5B, the HCV RNA dependent RNA polymerase, as the viral protein responsible for this phenotype and shown that pharmacological inhibition of its activity alleviates activation of the pro-inflammatory pathway. These results open new perspectives in understanding the inflammatory mechanisms linked to HCV infection and tumorigenesis.
| Hepatitis C affects nearly 200 million people worldwide. It results from the failure of the immune system to control the hepatitis C virus (HCV) replication and spread, leading to progressive liver disease that can culminate in fibrosis, cirrhosis and cancer. The inflammatory cells that infiltrate the diseased liver functionally contribute to fibrotic disease and cancer development by the release of potent soluble mediators that regulate cell survival and proliferation, angiogenesis, tissue remodelling, metabolism and genomic integrity. The goal of our work was to study the mechanisms of the initiation of the inflammatory process linked to HCV infection. We have shown that the presence of a single viral protein, namely NS5B, the RNA dependent RNA polymerase, promotes pro-inflammatory signaling. Moreover, inhibition of this pathway in HCV transgenic mice fully protects the animals from HCV-linked liver cancer. Our study contributes to a better understanding of the inflammatory mechanisms linked to HCV infection and thereby to tumorigenesis.
| Persistent HCV infection affects about 170 million people worldwide [1] and is one of the most common causes of chronic liver disease [2]. Infected individuals typically suffer from chronic liver inflammation that can last several decades and lead to progressive fibrotic liver that can culminate in hepatic cirrhosis and hepatocellular carcinoma (HCC) (for review see [3]).
Inflammation is the first step of the immune response against HCV infection and as such is beneficial to the host. However, in most cases, the infection is not resolved, fuelling the long-term persistent inflammation, with its many deleterious effects (for review see [4]), including the onset and progression of cancer. Inflammatory cytokines and chemokines are key molecular players in these processes, both by direct signaling, by recruiting further immune cells and by orchestrating production of reactive oxygen species, with their associated risk of inducing DNA mutations (for review see [5], [6].
Although the molecular mechanisms underlying HCV-associated liver cancer remain poorly understood (for review see [7]), there is no doubt that persistent liver inflammation increases the risk of HCC development by providing diverse mediators that perturb tissue homeostasis, including reactive oxygen species [8] and aberrant expression of cytotoxic cytokines [9], [10], [11]. Interestingly, it has been reported that several HCV proteins, namely core, NS3 and NS5A, can induce expression of pro-inflammatory cytokines [12], [13], [14] through yet to be identified mechanisms.
Lymphotoxin-α (LTα) and lymphotoxin-β (LTβ), two members of the tumor necrosis factor (TNF) superfamily, are necessary for organogenesis and maintenance of lymphoid tissues [15], [16]. LTα is soluble whereas LTβ contains a transmembrane domain. In consequence, LT exist both as soluble homotrimers (LTα3) that engage TNF receptor (TNFR) 1 and TNFR2 and the herpes virus entry mediator receptor (HVEM) and as membrane-bound heterotrimers (LTα1β2 or LTα2β1) that activate LTβR [17], [18]. LTβR acts through activation of canonical and alternative NF-κB signaling to induce the expression of a subset of chemokines (for review see [19], [20]. It has been shown that HCV infection is associated with increased hepatic LT expression both in vivo and in vitro [10], [21] and that HCV core protein can interact with the cytoplasmic domain of LTβR, thus stimulating the NF-κB pathways [22], [23]. Moreover, HCV replication in vitro depends on components of the LTβR pathway [24] while an ectopic LT expression in transgenic mice gives rise to liver inflammation and HCC [21]. However, the molecular mechanisms responsible for switching on LT expression in the HCV-infected hepatocytes have not been elucidated.
Here we report that tumors of HCV transgenic mice (FL-N/35 lineage) exhibit constitutively active LTβR and NF-κB signaling. Inhibition of the canonical NF-κB pathway through hepatocyte-specific deletion of Ikkβ [25] fully protects the animals from HCV-linked HCC. We further show that the viral RNA polymerase, NS5B, either alone or in the context of the full complement of viral proteins, is sufficient to induce expression of LT and NF-κB -dependant expression of its downstream target, CXCL10. Our data identify NS5B, recently shown to induce cytokine expression in hepatocytes through an RNA-dependent mechanism [26], as an inducer of the LTβR pathway, and specifically of lymphotoxin beta expression. These findings suggest that inhibitors of lymphotoxin signaling together with viral RNA polymerase inhibitors can be used to reduce HCV induced liver inflammation and HCC risk
FL-N/35 transgenic mice have a hepatocyte-targeted expression of the entire open reading frame (ORF) of the genotype 1b HCV, leading to expression of low levels of the full complement of viral proteins in the liver [27], [28]. In this model, HCV protein expression renders male mice at risk for liver tumorigenesis after one year of age [27]. Despite previous reports of lack of overt inflammation in the FL-N/35 animals, and because a vast majority of human HCV-linked HCC develops in necroinflammatory livers, we decided to reinvestigate a possible more subtle liver inflammatory phenotype of the FL-N/35 mice. In accordance with previously published observations [27], [29], prior to tumor development we detected only rare inflammatory foci, and no significant increase in either the number of inflammatory cells or proinflammatory cytokine expression in FL-N/35 livers compared to wild type mice (Figures S1 and S2). In contrast, multiple cellular infiltrations were present in FL-N/35 tumors (Figure 1A). The infiltrates were polymorphic and more specifically contained macrophages as well as B and T lymphocytes (Figure 1B).
It has been reported that activation of inflammatory signaling triggered by LTβR gives rise to hepatocellular tumors in mice [21]. To investigate whether this pathway is instrumental in HCV-related tumorigenesis in FL-N/35 animals, we studied the expression of several of its key components. Quantitative RT-PCR analysis showed a dramatic increase in LTβ expression in all FL-N/35 tumors analyzed (n = 10). LTα expression was also increased in most tumors, albeit to a lesser extent, while LTβR levels did not differ significantly between tumoral and peritumoral samples (Figure 2A). Tumor-specific augmentation of LTβ expression was confirmed at the protein level (Figure 2B), while immunofluorescence staining showed that hepatocytes were the major source of this cytokine (Figure 2C). Strong LTβ expression was specific to HCV-linked liver tumors, as it was not increased in N-myc driven tumors of WHV/N-myc2 transgenic mice [30] (Figures 3A and 3C). Reinforcing this result, there was no increase in LT expression in rare spontaneous liver tumors arising in animals of the same genetic background as FL-N/35 mice (Figures 3B and 3C). In addition to LTβ, several pro-inflammatory cytokines, notably TNFα, IL6 and Il1β (Figure S3) were mildly, but significantly increased in HCV-related tumors, while changes of interferons α and β expression (Figure S4) did not reach statistical significance. Altogether, these results suggest a specific link between LTβ and HCV-related tumorigenesis.
Increased LT expression has been reported in many human hepatic pathologies, including HCC of different etiologies [10], [21]. We have confirmed these observations by showing significant increase of LTβ in tumoral and peri-tumoral samples of patients carrying HCC of either HCV or alcohol related cirrhosis (Figure S5A). Importantly, hepatocytes are a major source of this cytokine in the diseased liver (Figure S5B).
LTβR signals through canonical and alternative NF-κB pathways to induce expression of several pro-inflammatory chemokines that act to recruit immune cells (for review see [18], [20]). To determine if LTβ upregulation is associated with activation of NF-κB signaling in the FL-N/35 tumors, we first investigated RelA (p65) localization in livers of tumor-bearing animals. Nuclear translocation of p65, indicative of canonical NF-κB activation, was detected in over 60% of tumoral hepatocytes, while less than 5% of peritumoral cells were positive in this assay, suggesting that NF-κB signaling was indeed activated in cells expressing LTβ (Figure 4A). In contrast, NF-κB was not activated in spontaneous liver tumors (Figure 4A). Next we assayed for activation of the alternative NF-κB signaling by visualizing cleavage of p100 into the mature p52 form of NF-κB. In agreement with previous reports of LT mode of action [19], the alternative NF-κB signaling was also activated in the HCV-related mouse tumors (Figure 4B). Moreover, the majority of tested tumors showed a strong increase of expression of CXCL10 (Figure 4C and 4D), an inflammatory chemokine downstream of LTβR (for review see [31]; [32]). Altogether these data suggest that increased LTβ expression in HCV-linked tumors leads to activation LTβR pathway of proinflammatory signaling.
While the role of canonical and alternative NF-κB signaling in liver carcinogenesis is complex (for review see [11]; [25]; [33]), it was suggested that the canonical NF-κB pathway is instrumental in relaying the oncogenic signal provided by LTβR activation [21]. This signal depends on the IKKβ catalytic subunit of the IκB kinase complex [34]. To determine if this scenario is operational in HCV-linked tumors, we crossed FL-N/35 mice with hepatocyte-specific IKKβ-deficient animals (IKKβΔhep) [35]. As previously reported [27], HCV transgenic mice carrying wild type Ikkβ alleles are tumor-prone, with 30% of males developing hepatocellular adenoma and carcinoma after 12 months of age (Figure 5A). In the genetic background compatible with HCV-related liver tumorigenesis ([28] and our unpublished data), we routinely observe spontaneous liver tumors in about 5% of over one year old males. Strikingly, in FL-N/35/IKKβΔhep mice, in which Ikkβ deletion was confirmed by western blot (Figure 5B) and which express similar levels of HCV RNA that the control FL-N/35 animals (Figure 5C), the frequency of tumor formation was indistinguishable from wt non-transgenic males (Figure 5A) and, similarly to spontaneous lesions, the single hepatic tumor that appeared in this cohort was negative for LTβ expression (not shown). Thus, invalidation of IKKβ-dependent canonical NF-κB signaling blocks HCV-related liver tumorigenesis in the FL-N/35 model.
To investigate the mechanism of LTβ induction by HCV proteins, we turned to a full-length HCV replicon propagated in Huh7 human hepatoma cells: the Nneo/C-5B model [36]. The replicon-containing cells expressed significantly more LTα, LTβ and, to a lesser extent, LTβR, compared to the parental Huh7 cells (Figure 6A). As in tumors from HCV transgenic mice, expression of CXCL10 was also induced in the Nneo/C-5B cells, suggesting that pro-inflammatory signaling cascade was activated. Moreover, productive infection of Huh-7.5.1 cells with JFH1-derived Con1/C3 HCV [37], [38] gave rise to a similar pattern of inflammatory signaling (Figure 6B).
While the HCV proteins are organized in an endoplasmic reticulum-associated multiprotein complex [39], isolated viral proteins maintain some activities that may be relevant to the physiopathology of viral infection. To determine if LT pathway activation could be related to a specific viral protein, we established stable polyclonal Huh7 populations in which expression of individual HCV proteins was driven by a heterologous promoter. Out of the five proteins tested (core, NS3, NS4A, NS5A and NS5B), only NS5B, the viral RNA-dependent RNA polymerase, reproduced the increase of LTβ expression (Figure 7A, 7D, Figure S6). This result was not a peculiarity of the cellular model used, since it was confirmed in HepaRG-tetNS5B cells, which are human immature hepatocytes closely resembling primary cells [40] with doxycycline-regulated expression of NS5B (Figure 7B). Interestingly, in contrast to most models used in this study, which are based on HCV proteins of the 1b genotype, the infectious JFH-1-based model and the HepaRG-tetNS5B express the genotype 2a NS5B, demonstrating that the observed phenotype is not restricted to a single viral isolate.
Next we asked if the enzymatic activity of NS5B was required for LTβ upregulation. Huh7 cells constitutively expressing NS5B were treated with 2′-C-Methylcytidine, a pharmacological inhibitor of RNA-dependent RNA polymerase activity [41], [42], [43]. While this treatment had no effect on NS5B expression, it abrogated upregulation of LTβ, LTα and CXCL10 (Figure 7C and 7E). Similarly, expression of a catalytically inactive mutant, NS5B G317V, [44] in HepaRG cells did not activate LTβ synthesis (Figure 7D). Importantly, enzymatic activity of NS5B was also required for activation of both the canonical and the alternative NF-κB signaling (Figure 7 F and G).
Finally, we studied the functional relationship between NF-κB and LT signaling and their downstream effector, the CXCL10 chemokine. We used shRNAs to silence expression of either the p65 NF-κB subunit or LTβ in Huh7-NS5B cells. Silencing of either of these genes fully abrogated CXCL10 induction by NS5B (Figures 7H, Figure S7). Taken together, our results strongly support the notion that NS5B activity, in the absence of viral RNA, gives rise to increased lymphotoxin expression, which in turn activates a NF-κB-dependent pro-inflammatory signaling.
Persistent HCV infection is a major cause of chronic liver disease. In particular chronic inflammation, resulting from continuous immune response against infected hepatocytes, is associated with necro-inflammatory changes, liver fibrosis and cirrhosis and HCC development (for review see [45]). The molecular mechanisms involved in initiation and in fuelling of this process, sometimes over very long periods, are still incompletely understood (for review see [7]). In this report we show an upregulation of a pro-inflammatory cytokine, LTβ, and its downstream targets, NF-κB and CXCL10, in HCV-related tumors and in several cellular models based on expression of HCV proteins. The most spectacular alteration of this inflammatory signaling pathway was a very strong upregulation of LTβ expression in nine out of ten liver tumors of transgenic mice with liver-targeted expression of HCV proteins. The one exception (animal 440 in Fig. 2) had high levels of LTβ transcripts and protein both in the tumoral and peri-tumoral liver samples, suggestive of an ongoing inflammation unrelated to HCV. Augmented LTβ expression was also observed in several hepatocyte cell lines harboring the totality or a subset of HCV proteins or solely NS5B, the RNA dependent RNA polymerase. However, it was not detectable in non-tumoral regions of FL-N/35 transgenic livers despite the presence of detectable viral RNA transcripts. In this context it is noteworthy that while efficient cytokine induction by NS5B requires high levels of the enzyme [26], the expression of HCV proteins is typically over 10–100 fold higher in cellular models compared to the transgenic mouse livers analyzed here [46], probably accounting for lack of LT expression in the livers of the FL-N/35 animals. Interestingly, the level of viral RNA in mouse tumors is comparable to that found in peritumoral liver (data not shown). Although we cannot exclude possible variations of NS5B protein expression between the non-tumoral and the tumoral tissues, as well as within individual cells, our data suggest that LT activation might not initiate tumorigenesis, but rather contributes to tumor progression in this animal model. Indeed, strong LTβ expression in 100% of tumors together with complete abrogation of HCV-linked tumorigenesis in animals invalidated for canonical NF-κB signaling, which acts both as an upstream activator and a downstream effector of LT pathway, prompt us to speculate that an autoregulatory loop involving LT and NF-κB might exist in HCV-linked HCC.
A previous report described strong activation of several additional inflammatory cytokines in mouse livers with orthotopic expression of NS5B [26]. In our experimental set up we detected only a mild, albeit significant, expression of TNFα, Il6 and Il1β and no significant increase in type I interferon in the mouse tumors. This apparent discrepancy between the two studies is once again most likely due to very different levels of expression of NS5B, which in our experimental model is at least an order of magnitude lower and probably closer to the physiological levels present in the majority of chronic hepatitis C patients.
LT exists predominantly as a membrane bound heterotrimer of LTα and LTβ subunits with LTα1-β2 stoichiometry, which binds with high affinity to LTβR [17]. Importantly, increased expression of LTβ was previously described in patients, in the context of chronic hepatitis C-associated cirrhosis and HCC [10], [21], [47], supporting physiopathological relevance of our data.
LTβR activation gives rise to expression of several chemokines through canonical and alternative NF-κB signaling (for review see [19]. Interestingly, in the FL-N/35 HCV transgenic mouse model, where the tumors show strong activation of both LT and NF-κB, abrogation of the canonical NF-κB pathway by hepatocyte-specific IKKβ ablation, led to a dramatic decrease in tumor incidence, arguing for a major role of NF-κB in promoting tumorigenesis in the context of HCV. However, the role of NF-κB in liver carcinogenesis is complex, as it inhibits cell death-promoted tumorigenesis [25], [48], [49], while promoting inflammation-driven tumor-formation in Mdr2-deficient [33] and in LT-transgenic mice [21] and in xenografts of human HCC [50].
It is perhaps not surprising that NF-κB, with its many possible downstream effectors and activities [51] is endowed with both pro- and anti-tumorigenic activities that are dominant under different physiological contexts. However, it is noteworthy that our data, linking HCV with LT and NF-κB signaling in the context of hepatocellular tumorigenesis, are in full agreement with HCC development triggered by ectopic LT expression [21].
We have shown that increased LT expression in hepatocytes expressing viral proteins has functional consequences in that it leads to synthesis of CXCL10. This C-X-C chemokine is expressed by hepatocytes in chronic hepatitis C [21], [48], [52], [53], [54]. It is induced by LTβR via NF-κB [31], [55] and is considered as one of the main chemoattractors for tumor-infiltrating immune cells (for review see [56]). It is thus tempting to speculate that CXCL10, induced by HCV viral proteins via LTβR and NF-κB could initiate liver recruitment of hematopoietic cells as well as intratumoral cellular infiltrates.
Mechanistically, we have shown that NS5B, the viral RNA-dependent RNA polymerase, is sufficient to activate the LT pathway and therefore upregulate chemokine production. Although physiologically NS5B is part of a multiprotein replication complex, the isolated protein also has enzymatic activity [57]. Moreover, NS5B interacts with several cellular proteins, including transcriptional regulators such as Rb [58], [59], RNA cellular helicases such as p68, which modulates RNA structures and is involved in RNA splicing, processing, transcription and translation [60] and eIF4AII, an RNA-helicase translation initiation factor [61]. Furthermore, a recent study described the role of the RNA sequence encoding NS5B as a pathogen associated molecular pattern (PAMP) following RNase L cleavage [62]. While all these interactions might participate in triggering inflammatory signaling downstream of NS5B, our data indicating that the enzymatic activity of NS5B is essential for induction of LT expression suggest that the molecular mechanism of LTβR activation by HCV relies on RNA synthesis, most probably from cellular RNA templates [63]. Further biochemical experiments are needed to formally demonstrate this point.
These uncertainties notwithstanding, the discovery of LT pathway activation by NS5B and the fact that pharmacological inhibition of its enzymatic activity alleviates the pro-inflammatory phenotype, open new perspectives for understanding the inflammatory mechanisms linked to HCV infection. In particular these results suggest that LTβR signaling could be an interesting target for therapies aimed at curbing HCV-related liver inflammation, known to be a major risk factor for severe hepatic pathologies, including HCC.
FL-N/35 transgenic animals [27] and IkkβF/F:Alb-Cre (referred to as IkkβΔhep) [35] were bred and maintained according to the French institutional guidelines. Twelve to twenty month-old males were used in these experiments.
HCC and corresponding nontumoral tissues were obtained from resected specimens from patients treated at the University Hospitals of Bordeaux and Montpellier, France. Small pieces from tumoral and nontumoral livers were snap frozen in liquid nitrogen and stored at −80°C until use. In parallel, samples were fixed and processed for immunohistochemistry. Informed consent was obtained according to the institutional regulations.
Huh7 cells were cultured in DMEM supplemented with 10% fetal bovine serum, 100 µg/ml streptomycin and 100 U/ml penicillin. 400 µg/ml of G418 were added to cells harboring the Nneo/C-5B replicons and 2 µg/ml of puromycin to Huh7-NS5B cells. HepaRG and HepaRG-NS5B tetracycline-inducible cells were grown in William's E medium supplemented with 10% fetal calf serum, 5 µg/ml insulin, 5.10−5 M hydrocortisone hemisuccinate, 100 units/ml penicillin, and 100 µg/ml streptomycin. When appropriate, cells were treated for 24 hours with 6 µg/ml of the NS5B inhibitors 2′-C-Methylcytidine from Santa Cruz Biotechnology (Heidelberg, Germany) or with 0.5 µg/ml of doxycycline from Sigma (St. Louis, MO).
NS5B cDNA sequences from genotype 1b was subcloned in Myc-tagged pMSCV retroviral vectors as previously described [64]. ShRNA coding sequences were cloned in pSIREN-RetroQ (Clontech, Palo Alto, CA). Plasmids were transfected into 293T cells with jetPEI (Polyplus, Illkirch, France), according to the manufacturer's instructions. Supernatants were used to infect Huh 7 cells. Infection efficiencies of 80% were routinely obtained. Puromycin (2 µg/ml) and hygromycin (150 µg/ml) were used as selection agents.
The sense and antisense strands of shRNAs were :
LTβ : 5′- atccgcctctactgtctcgtcggctattcaagagatagccgacgagacagtagaggcttttttctcgagg -3′
3′- gcggagatgacagagcagccgataagttctctatcggctgctctgtcatctccgaaaaaagagctccttaa -5′
P65 (RelA) : 5′- gatccggccttaatagtagggtaagttttcaagagaaacttaccctactattaaggccttttttctcgag -3′
3′- gccggaattatcatcccattcaaaagttctctttgaatgggatgataattccggaaaaaagagctccttaa –5′
ShLuc, the shRNA directed against luciferase, comes from RNAi-Ready pSIRENRetroQ Retroviral Vector kit (Clontech)
The point mutation G317V [45]. was introduced in the GDD motif of the NS5B gene by site-directed mutagenesis (QuikChange II XL, Agilent Technologies), using the following primers :
5′-GCTCGTGAACGTAGACGACCTTGTC-3′, 5′-GACAAGGTCGTCTACGTTCACGAGC-3′.
The specificity of the mutagenesis was verified by DNA sequencing of the entire coding sequence.
Western blots were performed as described previously [65]. Band intensities were quantified with the Gene Tools software (SynGene). Polyclonal rabbit antibodies anti-LTβ (ab 64835) and anti-NS5B (ab 35586) were from Abcam (Cambridge, UK). Polyclonal rabbit antibodies anti-p100/p52 (4882p) was from Ozyme (Saint-Quentin, France). Mouse monoclonal antibodies anti-IKKβ (clone 10AG2, Upstate) and anti-CXCL10 were respectively from Millipore (Temecula, CA, USA) and BD Biosciences (Oxford, UK).
Total RNA was isolated using an RNeasy Mini Kit (Qiagen, Germantown, MD, USA) including DNase treatment to remove possible genomic DNA contamination and used for first strand cDNA synthesis with random hexamers. Analyses were performed as described previously [65].
Mice were sacrificed with an overdose of pentobarbital (Narconen, Basel, Switzerland) and perfused transcardially with 4% paraformaldehyde in phosphate-buffered saline (PBS). The liver was removed, post-fixed and embedded in Tissue-Tek OCT Compound. Sections of 4 µm were stained with haematoxylin and eosin and then mounted in Eukitt.
Four micrometer sections were mounted on glass slides and stained using ABC Vectastain system from Vector laboratory (Burlingame, CA, USA). Monoclonal primary mouse antibodies for mice samples were anti-Mac 2, anti CD3 from eBioscience (San Diego, CA, USA) anti B220 from BD Biosciences (Oxford, UK) and p65 from Santa Cruz Biotechnology (Heidelberg, Germany). For human samples polyclonal rabbit antibodies anti-LTβ (ab 64835) was from Abcam (Cambridge, UK). Biotinylated secondary antibody was from Vector Laboratory (Burlingame, CA, USA). Control experiments were done in the absence of the primary antibody and were negative in all cases.
Cells were fixed in 4% paraformaldehyde, permeabilized in 0.1% Triton X-100, rinsed in phosphate-buffered saline, blocked with 1 mg/ml BSA and incubated with rabbit polyclonal anti-LTβ antibody from Abcam (Cambridge, UK) or with anti-p52 from Santa Cruz Biotechnology (Heidelberg, Germany) for 2 hours followed by anti-rabbit Alexa Fluor 488 for 1 hour. Samples were mounted with Fluorosave (Calbiochem, La Jolla, CA, USA) and analysed with a Zeiss fluorescent microscope equipped with a digital camera (Axiocam, Carl Zeiss, Oberkochen, Germany).
Experiments were performed at least three times. Data are presented either from a representative experiment or as mean ± SEM. Comparisons between groups were analyzed by Student's t test or Wilcoxon matched-pairs signed rank test as indicated.
|
10.1371/journal.ppat.1000057 | Primate Lentiviral Vpx Commandeers DDB1 to Counteract a Macrophage Restriction | Primate lentiviruses encode four “accessory proteins” including Vif, Vpu, Nef, and Vpr/Vpx. Vif and Vpu counteract the antiviral effects of cellular restrictions to early and late steps in the viral replication cycle. We present evidence that the Vpx proteins of HIV-2/SIVSM promote virus infection by antagonizing an antiviral restriction in macrophages. Fusion of macrophages in which Vpx was essential for virus infection, with COS cells in which Vpx was dispensable for virus infection, generated heterokaryons that supported infection by wild-type SIV but not Vpx-deleted SIV. The restriction potently antagonized infection of macrophages by HIV-1, and expression of Vpx in macrophages in trans overcame the restriction to HIV-1 and SIV infection. Vpx was ubiquitylated and both ubiquitylation and the proteasome regulated the activity of Vpx. The ability of Vpx to counteract the restriction to HIV-1 and SIV infection was dependent upon the HIV-1 Vpr interacting protein, damaged DNA binding protein 1 (DDB1), and DDB1 partially substituted for Vpx when fused to Vpr. Our results indicate that macrophage harbor a potent antiviral restriction and that primate lentiviruses have evolved Vpx to counteract this restriction.
| Defense against infection by the primate lentiviruses HIV/SIV is mediated primarily by antibodies that can neutralize the virus and by cytotoxic cells that can recognize and kill other virus-infected cells. However, in the past several years, research has revealed the existence of an additional line of host defense against HIV/SIV. It is now apparent that cells contain factors (also known as cellular restrictions) that potently inhibit virus infection. This has forced primate lentiviruses to evolve a strategy to counteract these cellular restriction factors. For example, HIV/SIV encode an accessory protein called Vif, whose function is to neutralize a cellular restriction to HIV/SIV infection. Our study provides evidence for a novel restriction that is expressed by macrophages and which potently antagonizes HIV and SIV infection. We describe how the virus protects itself from this cellular restriction. The goal is to harness this cellular restriction as the basis for a novel therapeutic strategy against HIV infection.
| The genomes of primate and non-primate lentiviruses encode “accessory” proteins from short open reading frames which are absent from the genomes of simple retroviruses [1]. The function of two of the accessory proteins, the Vif and Vpu proteins, have been defined: Vif antagonizes the antiviral activity of cellular Apobec 3 cytidine deaminases [2] and Vpu antagonizes the activity of tetherin to promote release of virions from the cell surface [3]. In all HIV and SIV lineages, the central viral region (overlapping Vif and Tat open reading frames) encodes at least one gene which is usually termed viral protein R (Vpr). Members of the HIV-2/SIVSM/SIVMAC lineage contain an additional gene in this region termed viral protein X (Vpx) which was originally derived from the African green monkey vpr gene by an ancestral recombination event [4]. Both Vpr and Vpx proteins are packaged into virions through association with the Gag polyprotein [5]–[7] and this points to an early role for these proteins in the virus life cycle (i.e., at a point proceeding de novo production of viral proteins). Most of the information regarding the roles of Vpr and Vpx proteins in primate lentivirus replication has been derived from studies with HIV-1 Vpr. The Vpr protein of HIV-1 has been shown to promote the accumulation of cells in the G2 stage of the cell cycle [8]–[11] and to associate with the DNA repair enzyme Uracil DNA glycosylase[12]. In addition, Vpr has been shown to promote the infection of terminally differentiated macrophages and dendritic cells [13]–[17]. These HIV-1 Vpr-ascribed activities segregate between the Vpx and Vpr proteins of HIV-2/SIVSM: Vpr of HIV-2/SIVSM induces cell cycle arrest and associates with UDG but is dispensable for macrophage infection while Vpx neither induces cell cycle arrest nor associates with UDG [4],[18]. However, Vpx is essential for infection of simian macrophages by SIV in vitro and following infection of simian macrophages by Vpx minus SIVSM, late cDNA product are reduced while 2-LTR cDNAs, which are formed only after completion of reverse transcription, are absent [4],[18]. Whether any of these activities relate to the functional role of Vpr/Vpx proteins in primate lentivirus replication, is unclear. In order to understand the functions of the Vpr/Vpx proteins in macrophage infection, we have focused on Vpx because of its profound impact on macrophage infection. In addition, its effect can be studied independently of other Vpr/Vpx-assigned activities including UDG association and cell cycle arrest.
We previously demonstrated that Vpx of HIV-2/SIVSM was essential for early events in macrophage infection yet dispensable for infection of CD4 lymphocytes [4]. We studied Vpx function in the context of SIVSM PB j which represents a primary isolate [19]. To increase particle infectivity and facilitate analysis of early events in the viral life cycle, viruses were pseudotyped with VSV-G envelope proteins. Although VSV pseudotyping has been shown to alleviate the defects exhibited by other accessory gene mutants such as Nef, pseudotyping did not alleviate the infectivity defect of Vpx-deleted viruses in macrophages. In order to gauge infection of primary macrophages under single cycle conditions, we quantitated viral cDNAs (mainly 2-LTR cDNA) by real time PCR. In this study, where we were dealing with a restriction and the viral target of the restriction was unknown, it seemed prudent to conduct experiments predominantly with viruses intact for all open reading frames as opposed to recombinant indicator viruses.
The profound requirement for Vpx in macrophage infection by HIV-2/SIVSM is illustrated in Figure 1A. 2-LTR cDNA is formed only after completion of viral reverse transcription and translocation of viral cDNA to the nucleus where circularization occurs. Levels of 2-LTR cDNA in macrophages infected with a wild-type SIV and an SIV variant lacking Vpr were indistinguishable (Figure 1A). In contrast, viral 2-LTR cDNA was reduced at least 100 fold in macrophages infected with an SIV variant lacking Vpx (Figure 1A). In COS cells and in HeLa cells, viral cDNA synthesis with wild type and Vpr-deleted or Vpx-deleted viruses were similar. Although 2-LTR cDNA was not detected in macrophages infected with SIVΔ Vpx, late viral cDNAs were evident but at a reduced level. Late cDNAs were reduced 15 fold and 2 fold at 24 and 48 h respectively in SIVΔ Vpx as compared to wild type infection of macrophages (Figure S1). Our original study [4] on the requirement for Vpx in SIV infection of monkey macrophages reported a predominant defect in 2-LTR circle formation and an approximately 3 fold defect in late cDNA synthesis using non quantitative PCR. This is consistent with the defect observed in this study which involves infection of human macrophages with SIV. The greater than 100 fold defect in 2-LTR cDNA formation was recapitulated in macrophage infections with Vpx deleted SIV variants expressing GFP (Figure 1A). This analysis revealed that an SIV variant lacking Vpx was at least 100 fold less infectious than the wild type counterpart (Figure 1A). Although Vpx was necessary for macrophage infection, it was dispensable for infection of COS/HeLa cells (Figure 1A). This suggested the existence of cellular activities, differentially expressed between macrophages and COS or HeLa cells, which impact primate lentivirus infection. One possibility was that COS and HeLa cells contain a cellular activity which promotes virus infection but in macrophages, this activity must be activated by the Vpx protein. An alternative possibility was that macrophages contain a cellular restriction to infection which is counteracted by the Vpx protein and this cellular restriction is not expressed in COS or HeLa cells. To distinguish between these two possibilities, we used a strategy previously adopted to characterize the mechanism by which Vif promotes viral infection [20],[21]. Heterokaryons were generated between macrophages and COS cells and the susceptibility of the heterokaryons to infection by SIVWT and SIVΔ Vpx was compared. When the fusogenic proteins of Newcastle Disease Virus (NDV) were expressed in COS cells, these cells readily underwent fusion with primary macrophages (Figure 1B). Macrophage/COS heterokaryons (double staining cells) were isolated by fluorescence-activated cell sorting (FACS). Double staining cells were not observed when normal COS cells (not expressing NDV proteins) were mixed with macrophages (Figure 1B). As an additional control, macrophage homokaryons were produced using polyethylene glycol (PEG). Both macrophage/COS heterokaryons as well as COS and macrophage homokaryons were infectible by wild type SIV (Figure 1B). In contrast, macrophage homokaryons and macrophage/COS heterokaryons were resistant to SIVΔ Vpx infection (Figure 1B). Since fusion with COS cells did not relieve the block to macrophage infection by SIVΔ Vpx, this indicated that macrophages harbor an antiviral restriction which is counteracted by the Vpx protein and this restriction is absent from COS and HeLa cells.
Vpx and Vpr are virion proteins and would thus be predicted to exert their function in the target cell shortly after infection and prior to de novo synthesis of viral proteins. Therefore, we examined whether Vpx delivered to macrophages would alleviate the restriction in trans to subsequent infection by a Vpx deleted virus. Macrophages were first infected (1° infection) with envelope deleted SIV variants harboring intact or defective Vpx genes (Figure 2). After an additional 8 or 16 hours, macrophages were subsequently super-infected (2° infection) with wild type or Δ Vpx SIV variants. Viral cDNA products were amplified using envelope-specific PCR primers (Figure 2). cDNA products amplified by these envelope specific primers were derived specifically from the secondary (2°) infection since viruses used in the primary infection (1°) lacked an intact envelope gene (Figure 2). Infection of macrophages harboring a wild type Vpx gene alleviated the block to subsequent SIVΔ Vpx super-infection 8 or 16 hours later (Figure 2). In contrast, macrophages initially infected with a ΔVpx SIV remained refractory to subsequent super-infection (Figure 2). Infection of macrophages with SIVWT also removed the restriction to subsequent infection by a Vpx minus SIV variant expressing GFP (Figure 2). This provided evidence that Vpx, delivered to the target cell, can counteract the restriction in trans.
Primate lentiviruses have evolved the accessory protein Vif to counteract the antiviral activity of cellular Apobec 3 cytidine deaminases [22]. Vif achieves this by promoting ubiquitylation and proteasomal destruction of Apobec 3 proteins [23]. To evaluate a possible role for the ubiquitin-proteasome system in the activity of Vpx, we first evaluated whether Vpx itself was ubiquitylated. HA-tagged Vpx and mutants thereof (Figure 3A, lower panel) were co-expressed in 293T cells with 6-Histidine-myc-tagged ubiquitin. Mono and poly ubiquitylated Vpx proteins were purified on nickel beads and Western blotted. Immunoblotting with an HA antibody revealed the presence of mono and poly ubiquitylated forms of Vpx (Ub-Vpx, Figure 3A). We also examined whether Vpx was ubiquinated by endogenous ubiquitin (as opposed to over expressed and tagged ubiquitin). HA-tagged Vpx was expressed in 293 T cells and cell lysates were directly Western blotted and probed with an HA antibody. This revealed the presence of higher molecular weight ubiquitylated forms of Vpx (Figure 3A, right panel). The extent of Vpx ubiquitylation was reduced to various extents in Vpx mutants containing single or multiple lysine to arginine substitutions (Figure 3A). Despite mutagenesis of all four lysines in Vpx, polyubiquitylated forms of the protein were still evident (compare GFP signal with VpxM4 signal). This suggested an involvement of both lysine and nonlysine residues in Vpx ubiquitylation [24],[25]. The ability of the Vpx lysine mutants to support SIV infection of macrophages was next examined. The various mutants were packaged within SIVΔVpx virions and single cycle infection of macrophages was evaluated from synthesis of late viral cDNAs and 2-LTR cDNAs (Figure 3B). The infectivity of the Vpx lysine mutants was impaired to various degrees (Figure 3B). The Vpx mutated lacking all four lysine (VpxM4) exhibited the lowest infectivity for macrophage. However, a mutant lacking the two N-terminal lysines (VpxNM2) appeared to be efficiently ubiquitylated yet this mutant also exhibited a significant infectivity defect (Figure 3B). However, due to technical obstacles in transfection of primary macrophages, we were unable to evaluate the extent of Vpx ubiquitylation of the various lysine mutants in primary macrophages and for this reason, we were unable to directly assess whether the extent of Vpx ubiquitylation was proportional to Vpx biological activity. For subsequent experiments, we focused on the use of Vpx mutant (VpxM4) containing substitutions at all four lysine residues. This mutant was efficiently packaged within virus particles at levels indistinguishable from wild type Vpx (Figure 3C). The packaging of the Vpx lysine mutant in viral particles suggests, at the very least, that this mutant is competent for binding to the p6 domain of the viral Gag polyprotein through which packaging of Vpr and Vpx proteins is mediated [5],[6].
We next examined whether the ability of Vpx to regulate SIV infection of macrophages required proteasome function. Macrophages were treated with three different proteasome inhibitors and then infected with wild type SIV and 2-LTR cDNA was quantitated 24 and 48 hours after infection. Lactacystin had a modest yet significant effect on SIV infection and ALLN and proteasome inhibitor 1 (Prot.1) markedly impaired SIV infection of macrophages (Figure 4). In contrast, neither ALLN nor proteasome inhibitor 1 affected SIV infection of COS cells (Figure 4). Identical results were obtained for HIV-2 in that the proteasome inhibitors compromised macrophage infection but not COS cell infection (Figure S2). In contrast, macrophage infection by HIV-1 (which does not contain Vpx) was not compromised by the proteasome inhibitors (Figure 4, right panel). Since proteasome disruption only impacted virus infection of cells in which Vpx was required for infection, this argued that proteasome inhibition specifically impaired Vpx function rather than impacting virus infection through off-target effects. The proteasome inhibitor lactacystin exerted a more modest but significant effect on SIV and HIV-2 infection of macrophages when compared to the other proteasome inhibitors. However, we were unable to test lactacystin in primary macrophages at higher concentrations because of toxicity. Similar toxic effects of proteasome inhibitors in primary dendritic cells have also limited complete suppression of proteasome function using such inhibitors [17]. Collectively, these experiments indicate the presence of a potent antiviral restriction in macrophages that is counteracted by the Vpx protein and that the proteasome/ubiquitin system is required for the ability of Vpx to counteract this restriction.
We next evaluated whether the antiviral restriction which antagonized HIV-2/SIVSM infection of macrophages was active against HIV-1. We first examined whether the Vpx protein, when packaged in trans within HIV-1 virions, enhanced virus infectivity for primary macrophages. While Vpx had no significant effect on the infectivity of wild type HIV-1, the infectivity of HIV-1Δ Vpr for macrophages was profoundly enhanced by Vpx but not by HIV-1 Vpr (Figure 5A, lower panel). The infectivity enhancement was also apparent in macrophages infected with an HIV-1 variant expressing green fluorescent protein (GFP) (Figure 5B). Thus, while HIV-1 was infectious for macrophages, its ability to infect these cells was markedly enhanced in the presence of Vpx. Vpx had no effect on the infectivity of wild type or ΔVpr HIV-1 for COS cells (Figure 5A, upper panel). A possible explanation for the ability of Vpx to compliment HIV-1 ΔVpr but not wild-type HIV-1 is that Vpx and HIV-1 Vpr proteins compete for packaging within HIV-1 virions. An alternative possibility was that these proteins do not compete for packaging into virions but compete for interaction with the restriction after infection has occurred. Western blotting analysis revealed that both wild type and lysine mutant (VpxM4) Vpx proteins were packaged into wild type and Vpr deleted HIV-1 virions (Figure 5C). This suggested that HIV-1 Vpr competed with Vpx in the target cell following infection and this competition precluded the ability of Vpx to activate the restriction. A prediction of this is that delivery of Vpx to this target cell prior to HIV-1 infection should be sufficient to inactivate the restriction and subsequently enhance macrophage infection by both wild type and Vpr deleted HIV-1. To evaluate this, we bypassed the requirement for Vpx packaging by directly introducing Vpx into the target cell by SIVWT infection. The susceptibility of those cells to infection by wild type or Vpr-deleted HIV-1 variants was then examined. In this case, the infectivity of both wild type and vpr deleted HIV-1 variants for macrophages was enhanced when Vpx was first introduced into the cell by SIVWT infection (Figure 5D). In contrast, prior infection with a SIVΔ Vpx variant did not enhance subsequent HIV-1 infection of macrophages (Figure 5D). Therefore, in the absence of competition by packaged Vpr, Vpx greatly enhanced HIV-1 infectivity for macrophages. We next evaluated whether the ability of Vpx to enhance HIV-1 infectivity depended upon its ubiquitylation. As was the case for SIV, a Vpx mutant lacking ubiquitylation sites (VpxM4) did not enhance HIV-1 infectivity when packaged within HIV-1Δ Vpr virions (Figure 6A). This was also apparent in infections using indicator viruses (Figure 6B). In this case, the ability of Vpx to enhance the infectivity of a Vpr deleted HIV-1 variant expressing GFP was compromised by the M4 mutation. In addition, the ability of Vpx to enhance HIV-1 infectivity required proteasome function in that Vpx failed to enhance permissiveness of macrophages to HIV-1 infection in macrophages in which proteasome function was disrupted by ALLN or proteasome inhibitor 1 (Figure 6C).
Recent studies have demonstrated that the ability of HIV-1 Vpr to induce cell cycle arrest requires the E3 ubiquitin ligase complex scaffolding factor, damaged DNA binding protein 1 (DDB1) [26]–[30]. Therefore, we examined whether the ability of Vpx to counteract the macrophage restriction to SIV and HIV-1 infection was DDB1-dependent. In 293T cells, endogenous DDB1 associated with a wild-type SIV Vpx protein but not with a SIV Vpx mutant lacking lysine residues (VpxM4) (Figure 7A). A specific association of SIV Vpx with DDB1 was apparent in coimmunoprecipitation experiment with either FLAG-tagged Vpx or with HA-tagged Vpx proteins (Figure 7A). If DDB1 is a functional Vpx interactor, we would predict that DDB1 silencing would only impact SIV infection of macrophages in which the restriction was expressed but not in COS cells which lack the restriction. In addition, HIV-1 Vpr did not antagonize a macrophage restriction. The activity of the restriction in HIV-1 was only revealed by the ability of Vpx to enhance HIV-1 infection of macrophages. Therefore, a prediction is that DDB1 silencing should not inhibit infection of macrophages by HIV-1. DDB1 specific siRNAs efficiently reduced DDB1 expression in COS cells and in macrophages (Figure 7B, left panels). While DDB1 silencing had no significant effect on SIV infection of COS cells (p>0.05), SIV infection was significantly impaired (p<0.005) in DDB1-depleted macrophages (Figure 7C, upper right panel). In contrast, macrophage infection by HIV-1 was not affected by DDB1 silencing (Figure 7C, lower right panel). We also used an independent strategy to deplete DDB1 in macrophages to assess its role in virus infection. Similar to the results obtained with siRNA mediated DDB1 depletion, depletion of DDB1 using DDB1-specific shRNAs also specifically impaired the susceptibility of primary macrophages to SIV infection but not HIV-1 infection (Figure S3). Therefore, DDB1 appears to be a specific Vpx cofactor in primary macrophages.
We next examined whether DDB1 was required for the ability of Vpx to counteract the restriction to HIV-1 infection. Since Vpx, when packaged in HIV-1 virions, enhanced macrophage infection, we examined whether Vpx enhanced HIV-1 infection in DDB1 depleted macrophages. While packaging of Vpx in HIV-1 particles markedly increased infectivity for macrophages transfected with a scrambled siRNA (Figure 8A) silencing of DDB1 in macrophages significantly reduced (p<0.002) the ability of Vpx to enhance HIV-1 infection (Figure 8A). However, DDB1 silencing had no significant effect (p>0.05) on the infectivity of HIV-1 which had not packaged Vpx (Figure 8A). Since SIV Vpx but not SIV Vpr was essential for macrophage infection (Figure 1A), we examined whether fusion of DDB1 to SIV Vpr was sufficient to allow SIV Vpr to counteract the macrophage restriction. Packaging of Vpr alone into a Vpr and Vpx deleted SIV (SIVΔXR) did not permit macrophage infection. In contrast, there was a partial and significant (p<0.005) restoration of infectivity when a Vpr-DDB1 fusion was packaged relative to infectivity of virions in which the DDB1 protein was not packaged (Figure 8C). Although ubiquitylation was necessary for the ability of Vpx to counteract the restriction to HIV-1 and SIV infection of macrophages, DDB1 protein was not required for Vpx ubiquitylation (Figure 8C). Mono and poly ubiquitylated forms of Vpx were evident and apparently increased in cells in which DDB1 expression was reduced by RNA interference (Figure 8C). Collectively, these results suggest that DDB1 is required for the ability of Vpx to antagonize a restriction to infect macrophages by HIV-1 and SIV but that DDB1 is not required for Vpx ubiquitylation.
Our study suggests that the function of Vpx is to antagonize an antiviral restriction in macrophages. Vpx exhibits similarities with the Vif protein of primate lentiviruses in that inactivation of the restriction required the proteasome/ubiquitin system. A role for the proteasome/ubiquitin system is provided by our demonstration that ubiquitylation mutants of Vpx are functionally attenuated and treatment of macrophages with proteasome disrupting agents specifically reduces their susceptibility to SIV infection but not HIV-1 infection. The inhibitory effect of proteasome inhibitors on SIV infection of primary macrophages as reported in our study appears to be at odds with studies demonstrating that HIV-1 infection of cell lines is enhanced in the presence of proteasome inhibitors [31]–[35]. The majority of these studies have involved cell lines and one of these studies [31] has suggested that proteasome inhibitors enhance HIV-1 infection by inducing G2/M cell cycle arrest thereby imparting a cellular environment that is more permissive to infection. Our study used primary macrophages and since these cells are terminally differentiated and nondividing, enhancing effects of proteasome inhibitors due to cell cycle arrest would not be manifest. A comparison of our study with the study Goujon et. al. [17] demonstrates that Vpx is essential for infection of macrophage (our study) and of dendritic cells [17]. However, there are some differences in the results obtained with Vpx mutant viruses in these two systems. In the study of Goujon et al. [17], the proteasome inhibitor MG132 marginally (1–2 fold) increased viral DNA accumulation in dendritic cells in the presence of Vpx whereas in our study, proteasome inhibitors markedly inhibited infection of macrophages by SIV but not HIV-1. Since Goujon et al. [17] reported that primary human dendritic cells were highly sensitive to the toxic effects of MG132, it is possible that differences in treatment conditions that can be employed in macrophages versus dendritic cells could account for these differences. The study of Goujon et al. [17] also showed an enhancement of SIV infection in the absence of Vpx. We did not examine the effects of proteasome inhibitors on a Vpx-deleted virus in macrophages because this variant was essentially dead in these cells.
Our study implicates DDB1 as a cellular cofactor of Vpx which is necessary for the ability of Vpx to counteract the macrophage restriction. This is supported by several independent experiments. DDB1 silencing in macrophages specifically impaired their susceptibility to infection by SIV and, in addition, impaired the ability of Vpx to enhance infectivity of macrophages by HIV-1. It is not possible to conclude at present whether DDB1 association accounts, in totality, for the biological activity of Vpx. DDB1 silencing led to a 5–10 fold reduction in SIV infectivity of macrophages whereas there was a 100 fold infectivity defect imparted by deletion of Vpx. However, RNA silencing failed to completely deplete DDB1 from primary macrophages and it is possible that residual DDB1 allowed some retention of Vpx activity in these macrophages. We also present evidence that DDB1 binds to ubiquitylated Vpx and that lysine mutants of Vpx which are inefficiently ubiquitylated exhibit reduced DDB1 binding and are impaired in their ability to support SIV infection of macrophages. Using a Vpx mutant lacking lysine residues, we present evidence that Vpx ubiquitylation is important for association with DDB1 and to counteract the macrophage restriction. Although we attribute loss of Vpx function to lack of ubiquitylation and loss of DDB1 binding, we cannot rule out the possibility that loss of function of the mutant protein was due to indirect effects of the mutations on protein structure. However, at the very least, the Vpx lysine mutant is packaged within virions which suggests that it is competent for interaction with the p6 domain of the Gag polyprotein. As with DDB1 silencing, the reduction in Vpx function imparted by mutation of all four lysines in Vpx caused a no more than a 10 fold defect in Vpx function (for example, see Figure 3B; Figure 6A,B). Therefore, ubiquitylation and DDB1 association may not fully account for the biological activity of Vpx in macrophages. However, polyubiquitylated forms of Vpx were still evident in cells transfected with a Vpx mutant lacking all lysine residues (Figure 3A). This suggests some degree of Vpx ubiquitination on nonlysine residues [24],[25]. Identification and mutagenesis of all ubiquitination residues on Vpx will be required before the degree to which Vpx activity depends upon ubiquitination can fully be assessed. Our study also suggests that DDB1 is not required for Vpx ubiquitylation but that Vpx ubiquitylation is necessary for association with DDB1. Therefore, the loss of function observed with the Vpx lysine mutant is likely to reflect a loss in DDB1 binding. Although SIV Vpr did not counteract the macrophage restriction, fusing it to DDB1 partially conferred this ability. This suggests that the function of Vpx may be to tether DDB1 to the reverse transcription complex upon which the restriction acts. Our study also indicates that DDB1 is required for the ability of Vpx to counter the macrophage restriction to HIV-1 infection. HIV-1 Vpr did not exhibit the ability to counter the macrophage restriction. For this reason, silencing of DDB1 did not impair susceptibility of macrophages to HIV-1 infection. However, the fact that the restriction was active against HIV-1 was revealed by the demonstration that Vpx greatly increased the permissivity of macrophages to HIV-1 infection. In this situation, silencing of DDB1 inhibited the ability of Vpx to enhance macrophage infection by HIV-1. Although Vpx is a virion protein, we do not know if DDB1 itself is packaged within virions. However, since silencing of DDB1 in the target cell inhibited SIV infection, this suggests that Vpx usurps DDB1 after infection of the target cell and likely, within the context of the reverse transcription complex.
Our study also reveals a paradox with regards to the functional consequences of HIV-1 Vpr and HIV-2/SIV Vpx interaction with DDB1. DDB1 mediates the cell cycle arrest property of HIV-1 Vpr. DDB1 was also necessary for the ability of SIV Vpx to counteract the macrophage restriction. However, SIV Vpx, although able to interact with DDB1, does not induce cell cycle arrest. Furthermore, the ability of HIV-1 Vpr to interact with DDB1 does not appear sufficient to confer upon HIV-1 Vpr the ability to efficiently counteract the macrophage restriction. Therefore, there are likely to be different biological outcomes that are dictated by the nature of the interactions that HIV-1 Vpr and SIV Vpx forge with DDB1 and its associated E3 ubiquitin ligase complex components. Further insight into the mechanisms employed by HIV-1 Vpr and HIV-2/SIVSM Vpx to enhance macrophage infection may be revealed once the macrophage restriction itself is identified.
The infectious molecular clone SIVSM PBj1.9 was used for the majority of experiments in this study. This clone, which is representative of the HIV-2/SIVSM group of viruses, was derived from short-term peripheral blood mononuclear cell (PBMC) cultures. Unlike many other HIV-2 and SIVSM clones, PBj1.9 has a complete set of uninterrupted accessory genes and replicates efficiently in macrophages and represents a physiologically relevant virus strain. Mutations which abrogated the translation of Vpx and Vpr genes are as described previously [4]. HIV-GFP (a gift of Paul Clapham, University of Massachusetts Medical School) contains an EGFP gene inserted between the envelope stop codon and nef within the HIV-1NL4-3 backbone. GFP expressing variants of wild type and ΔVpx SIV contain an EGFP gene inserted between Bst 1107I sites within the viral envelope gene (as schematized on Figure 2). Wild type and ΔVpr HIV-1 variants were studied in the context of HIV-1NL4-3. For the generation of viral stocks, 293T cells were transfected with proviral DNAs (25 µg) using a modified calcium phosphate/DNA precipitation method (Stratagene). Viruses were pseudotyped with VSV envelope glycoproteins by cotransfection of proviral DNAs with a plasmid expressing the VSV envelope glycoprotein. For encapsidation of wild type and mutant Vpx and Vpr proteins, 293T cells were cotransfected with proviral DNAs and plasmids expressing Vpx and Vpr proteins. The DNA ratio for pVSV-G, proviral clones and pIRES2-EGFP-Vpx was 1∶14∶1. For encapsidation of Vpr-DDB1 fusion proteins, 293T cells were co-transfected with an SIV deltaVpx/deltaVpr proviral clone, pIRES2-EGFP Vpr-fDDB1 and pVSV-G. The DNA ratio for pVSV-G, proviral clone and DDB1 expression plasmids was 1∶14∶1. HIV-1 and SIV stocks were normalized on the basis of reverse transcriptase activity. Viral infection efficiency was gauged from synthesis of viral cDNA products at early intervals (24 and 48 h) post-infection. PCR conditions for amplification of SIVSM and HIV-1 2-LTR cDNAs are as described previously [4],[36]. cDNA copy numbers were expressed on a per cell basis after quantitation of genomic DNA copy numbers using PCR and primers to the CCR5 gene [36]. Macrophages were initially infected with VSV-pseudotyped SIV variants harboring intact or defective Vpx genes. Viruses used in the initial infection additionally lacked an intact envelope open reading frame. Macrophages were then super-infected with SIV variants which harbored intact envelope genes. As a consequence, cDNA products generated specifically by the super-infecting virus could be identified. SIV cDNA products were amplified in two rounds of PCR with JumpStart™ RedaccuTag™ DNA polymerase (Sigma). First round products were amplified using forward (taacaggaacaccagcaccaaca) and reverse (catctgctttccctgacaa) primers. Second-round products were amplified using forward (taacaggaacaccagcaccaaca) and reverse (aagcataacctggcggtgcaca) primers.
Supernatants from 293T cells transfected with infectious molecular clones were cleared of cellular debris by low-speed centrifugation (1500 g, 10 min) and then filtered (0.45 µm). Virions in clarified supernatants were harvested (10,000 g, 2 h) and resuspended in serum-free medium (500 µl). Concentrated virions were applied to a 15–60% w/v continuous sucrose gradient and virions were resolved at 200,000 g for 16 h. Gradient fractions (0.5 ml) were collected and virus levels in each fraction were measured by reverse transcriptase activity. Virus particles in individual gradients were pelleted and resuspended in sample buffer and the presence of encapsidated Vpx proteins was examined by Western blotting with an αHA antibody.
Peripheral blood monocytes were obtained by elutriation and counter current centrifugation and maintained 2 days in DMEM containing 10% human serum and monocyte colony stimulating factor (MCSF) (RD Systems) and for a further 5 days in medium lacking MCSF prior to use in experiment. 293T, Hela and COS cells were maintained in DMEM containing 10% FBS.
Macrophages or COS cells (8×105) in 24 well plates were directly infected with VSV-G-pseudotyped viruses (1×106 cpm RT/ml or 1 ug p24/ml) in the presence of proteasome inhibitors including Lactacystine (10 uM), ALLN (50 uM) and Proteasome inhibitor 1 (50 uM). After 3–5 h, supernatant was removed and replaced with fresh medium containing proteasome inhibitors. After 24 and 48 h post-infection total DNA was isolated using DNAzol reagent (Invitrogen) and analyzed by real-time PCR assay for 2LTR circles.
For FACS analyis, COS cells and human macrophages were stained with 3.5 µM CellTracker Green CMFDA (5-chloromethylfluorescein diacetate) and 24 µM CellTracker Blue CMAC (7-amino-4-chloromethylcoumarin), respectively. For fluorescence microscopy, COS cells and macrophages were stained with 2.5 µM DiO (3,3′-dioctadecyloxacarbo cyanine perchlorate) and 12 µM DiI (1,1′-dioctadecyl-3,3,3′,3′-tetramethylindocarbocyanine perchlorate) respectively, according to manufacturer's instructions (Molecular Probes).
Generation of macrophage homokaryons was achieved by polyethylene glycol (PEG). Briefly, labeled cells, 15×106 each group, were mixed and centrifuged at 250 g. 50% PEG-1450 was added dropwise to the pellet and cells incubated for 2 min at 37°C with gentle mixing. 1 ml PBS was then added dropwise to the cells over 1 min, followed by 3 ml of 2% FBS/PBS over another 2 minutes. Cells were washed 3 times with 2% FBS/PBS and plated in a 100 mm culture dish (1×107 cells/dish). COS-macrophage and COS-COS cell fusion was achieved using paramyxovirus hemagglutinin-neuraminidase (HN) protein and fusion (F) protein. Briefly, COS cells were transfected with pCAGGS-HN and pCAGGS-F expression vectors encoding HN and F proteins of Newcastle disease virus (gift of Prof. T. Morrison) [37]. Sixteen to twenty hours post-transfection, COS cells were stained, mixed with stained macrophages (ratio 1∶1.5) and plated in 100 mm dishes. COS homokaryons were generated at 1∶1 ratio. After overnight incubation, cells were infected with either SIVWT or SIVΔVpx for 24 h. Cell sorting was performed with a FACSAria flow cytometer using the FACSDiva software (Becton Dickinson). Double-stained cells were sorted. Total DNA was isolated using DNeasy Blood and Tissue Kit (Quiagen) and analyzed by real-time PCR assay for 2LTR circles.
The SIVsm Vpx and HIV-1 Vpr genes were amplified from PBj1.9 and NL4.3 proviral clones respectively, and inserted into a pIRES2-EGFP vector (BD) either with or without a N-terminal minimum HA epitope. The upstream primer for each PCR product provided a Kozak sequence. The Vpx lysine mutants (K68,77,84,85R) were generated by Quikchange XL site-directed mutagenesis (Stratagene). The DDB1 gene was amplified and subcloned from pBj-hp125 (ATCC, MBA-126) and inserted into pIRES2-EGFP as an in frame fusion with the C-terminal of SIV Vpr. A Flag epitope was added to the N-terminal of DDB1 as flanking sequences between Vpr and DDB1. As a control, a N-terminal Flag tagged DDB1 was inserted into pIRES2-EGFP.
293T cells were co-transfected with HA-Vpx, HA-Vpx lysine mutants or a pIRES2-EGFP empty vector and pRGB4-6His-myc-Ubiquitin at a 1∶4 ratio using lipofectamine 2000 (Invitrogen). Non-6His tagged Ubiquitin was included as a control for Ni-NTA pull-down. 36 h after transfection, the 6His-ubiquitin conjugated proteins were purified using Ni-NTA Magnetic Agarose beads (Qiagen) under native conditions [38]. Briefly, cells were lysed in detergent buffer (10 mM Tris-Hcl pH7.5, 150 mM NaCl,1% Triton X-100 and protease inhibitor cocktail) and clarified by centrifugation at 14,000 rpm for 15 min. The cell lysates were incubated with Ni-NTA beads overnight at 4°C in detergent buffer with 300 mM NaCl, 20 mM imidazole and 5 µM MG132. The beads were washed in lysis buffer and attached proteins were eluted in elution buffer (50 mM NaH2PO4, 375 mM NaCl, 1% Triton, 250 mM imidazole pH 8.0).
Virus pellets were lysed in RIPA buffer (50 mM Tris-Hcl pH 7.5, 150 mM NaCl, 1% NP-40, 0.5% NaDoc, 0.1% SDS and protease inhibitor cocktail) lysates of transfected cells or gradient purified virions were boiled in sample buffer, resolved by SDS/PAGE and Western blotted with the following antibodies: HA-Vpx (HA, 16B12, Covance), myc-Ubiquitin (α-ubiquitin, P4G7, Covance; α-Myc 9E10, Sigma), Capsid (polyclonal, ABI), γ-tubulin (GTU-88, Sigma), Flag-Vpx (M2, F3165, Sigma), DDB1 (Goat polyclonal antibody PC718,Calbiochem).
The siRNA sequences for DDB1 silencing in macrophages, COS-1 or 293T cells were
The Scrambled control siRNA sequence was CAGTCGCGTTTGCGACTGG
Macrophages or COS-1 cells were transfected twice with 60 pmol each siRNA using lipofectamine 2000. 24 h after siRNA transfection, cells were infected with RT-normalized virus as indicated. The DDB1 protein knockdown levels were examined at the same time point as cDNA analysis.
For shRNA-mediated DDB1 silencing, macrophages are infected with a TRIP lentiviral vector [39] containing or lacking DDB1 hairpin sequences. 48 h after transduction with shRNA lentivirus vectors, macrophages were infected with VSV-g-pseudotyped SIV or HIV-1 and levels of viral cDNA synthesis was assessed after additional 48 h (96 h post lentivirus vector transduction).
293T cells were transfected with Flag-Vpx, Flag-Vpx lysine mutant (VpxM4) or pIRES2-EGFP vector. 36 h after transfection, cells were harvested and lysed in Co-IP lysis buffer (100 mM NaCl, 50 mM Tris-Hcl pH 7.5, 5 mM MgCl2, 0.5% NP-40, protease inhibitor cocktail) and incubated with Protein A and Protein G beads (Invitrogen) conjugated anti-Flag M2 antibody overnight at 4°C. The beads were washed 4 times in a more stringent wash buffer (400 mM NaCl, 50 mM Tris-Hcl pH 7.5, 5 mM MgCl2, 0.5% NP-40, protease inhibitor cocktail). And bound proteins were boiled and eluted in 2× Laemmli's SDS-sample buffer.
Where indicated, data was analyzed using an unpaired Students t test. p values of 0.05 or lower were considered significant. Statistical analysis was performed using Graph Pad Prizm 5 software. |
10.1371/journal.pgen.1008327 | SMAX1/SMXL2 regulate root and root hair development downstream of KAI2-mediated signalling in Arabidopsis | Karrikins are smoke-derived compounds presumed to mimic endogenous signalling molecules (KAI2-ligand, KL), whose signalling pathway is closely related to that of strigolactones (SLs), important regulators of plant development. Both karrikins/KLs and SLs are perceived by closely related α/β hydrolase receptors (KAI2 and D14 respectively), and signalling through both receptors requires the F-box protein MAX2. Furthermore, both pathways trigger proteasome-mediated degradation of related SMAX1-LIKE (SMXL) proteins, to influence development. It has previously been suggested in multiple studies that SLs are important regulators of root and root hair development in Arabidopsis, but these conclusions are based on phenotypes observed in the non-specific max2 mutants and by use of racemic-GR24, a mixture of stereoisomers that activates both D14 and KAI2 signalling pathways. Here, we demonstrate that the majority of the effects on Arabidopsis root development previously attributed to SL signalling are actually mediated by the KAI2 signalling pathway. Using mutants defective in SL or KL synthesis and/or perception, we show that KAI2-mediated signalling alone regulates root hair density and root hair length as well as root skewing, straightness and diameter, while both KAI2 and D14 pathways regulate lateral root density and epidermal cell length. We test the key hypothesis that KAI2 signals by a non-canonical receptor-target mechanism in the context of root development. Our results provide no evidence for this, and we instead show that all effects of KAI2 in the root can be explained by canonical SMAX1/SMXL2 activity. However, we do find evidence for non-canonical GR24 ligand-receptor interactions in D14/KAI2-mediated root hair development. Overall, our results demonstrate that the KAI2 signalling pathway is an important new regulator of root hair and root development in Arabidopsis and lay an important basis for research into a molecular understanding of how very similar and partially overlapping hormone signalling pathways regulate different phenotypic outputs.
| Karrikins are plant signaling compounds from smoke, which induce germination of fire-following plants. They likely mimic endogenous plant hormones (KAI2-ligand, KL), because Arabidopsis karrikin receptor mutants display shoot developmental phenotypes. Perception of karrikins/KL is very similar to that of another plant hormone class, strigolactones (SLs). Both hormones bind to the related α/β-fold hydrolase receptors KAI2 and D14 respectively, which both interact with the F-box protein MORE AXILLIARY BRANCHES2 (MAX2), for ubiquitylation and subsequent degradation of KL- or SL-signalling specific proteins of the SMXL family. Based on max2 mutant phenotypes it has been suggested that the development of Arabidopsis root architecture and root hairs is regulated by SL signaling. However, max2 does not distinguish between the two signalling pathways. We genetically dissected the role of KL and SL signalling in root and root hair development in Arabidopsis seedlings and show that most root traits are regulated by KL and not by SL signaling: lateral root density is controlled by KL and SL signalling together, while root growth direction, root straightness and root hair development are determined by KL signalling alone. Thus, KL signalling regulates vital plant traits for nutrient and water uptake as well as anchorage to the ground.
| Plant roots continually integrate environmental information to make decisions about their development, and to optimize their growth for optimal nutrient uptake and anchorage. Increased lateral root formation and root hair growth are necessary to compensate for low nutrient availability in the soil by increasing the root surface area for nutrient uptake, while directional growth is required to avoid stressors such as salt, obstacles or to reach moisture [1–5]. Root development is regulated by a number of phytohormones, low-molecular-weight signalling molecules, which mediate localized developmental responses as well as transmission and integration of information across long distances. Among them, SLs have been suggested to act as important regulators of Arabidopsis seedling root architecture and root hair development [6–9]. However, the exact role of SLs in root development remains uncertain, due to interpretational difficulties inherent in the materials used by those studies, namely max2 mutants and the synthetic strigolactone racemic-GR24 (see below, [10]).
Genes involved in SL biosynthesis have been identified in several plant species [10]. The universal SL precursor carlactone is synthesized from β-carotene by a core pathway of three enzymes; the isomerase DWARF27, and the carotenoid cleavage dioxygenases CCD7 and CCD8 (MAX3 and MAX4 in Arabidopsis) [11]. Carlactone is then modified by a variety of enzymes, including the cytochrome P450s of the MAX1 sub-family, to create a range of active SL molecules [12]. SLs are perceived and hydrolysed by the α/β hydrolase receptor DWARF14 (D14) [13–16]. D14 interacts with the SCFMAX2 E3 ubiquitin ligase complex to induce ubiquitylation and subsequent degradation of target proteins, essential to trigger SL signal transduction [15, 17].
A second, closely related signalling pathway also acts through the SCFMAX2 complex [18, 19]. In this pathway MAX2 is thought to interact with KAI2 (KARRIKIN-INSENSITIVE2), an α/β hydrolase receptor protein, which is encoded by an evolutionary older paralog of D14 [20–22]. KAI2 was originally identified as a receptor for karrikins, a family of butenolide compounds found in the smoke of burnt plant material [19, 23]. In fire-following species, karrikins are used as germination cues, indicating the removal of competing plants. However, karrikins promote germination in a range of flowering plant species, which do not germinate after fire [24–26] and KAI2 is required for a number of developmental traits in Arabidopsis not related to germination as well as for arbuscular mycorrhiza symbiosis in rice [19, 27–30]. Because of these roles of KAI2, karrikins are thought to mimic the action of a yet unknown endogenous plant signalling molecule, which is currently denoted KAI2-ligand (KL) [31–33].
Since KAI2 and D14 act through the same F-box protein MAX2, max2 mutants are insensitive to both SLs and karrikins, and display the combined phenotypes of d14 and kai2 mutants [18, 19, 27, 28]. Most studies aimed at understanding the role of SLs in Arabidopsis root development have used max2 mutants—likely for historical reasons because they were available prior to d14 and kai2. However, if only max2 mutants are employed without comparison with the specific receptor mutants, the root phenotypes cannot be reliably attributed to either SL or KL signalling. The second difficulty in interpreting previously published root phenotypes arises from the experimental use of the strigolactone analog GR24, which in standard preparations is a racemic mix of two stereoisomers (rac-GR24). While one stereoisomer (GR245DS) is a potent activator of D14 signalling, the non-natural stereoisomer (GR24ent-5DS) appears to stimulate KAI2 signalling [31, 34]. As such, the indiscriminate use of rac-GR24 has created a legacy of interpretational problems in previous studies, and incorrect attribution of phenotypic effects to SL signalling [10, 34].
Genetic and biochemical evidence indicates that the D14-SCFMAX2 and the KAI2-SCFMAX2 complex target a group of regulators–the SMXL (SMAX1-LIKE) family of proteins with weak homology to ClpB type chaperonins–for ubiquitylation and subsequent proteolytic degradation. In Arabidopsis, the genetically defined degradation targets of KL signalling are SMAX1 (SUPPRESSOR OF MAX2 1) and SMXL2, while the targets of SL signalling are SMXL6, SMXL7 and SMXL8 (hereafter SMXL678) [27, 35–37]. In the shoot, the hormone-induced turnover of SMXL678 proteins is key to correctly shaping shoot architecture [38]. The exact molecular function of the SMXL proteins is poorly understood. SMXL678 and their rice ortholog D53 have been associated with transcriptional regulation, since they physically interact with TOPLESS-RELATED (TPR) co-repressor proteins [27, 39, 40]. Rice D53 interacts with IPA1, a SQUAMOSA PROMOTER-BINDING FAMILY LIKE (SPL) transcription factor in the regulation of shoot branching and may recruit TPR to repress IPA1-mediated transcription [41]. However, they have also been found to be involved in enhancing PIN1 accumulation at the basal membrane of stem xylem parenchyma cells and auxin transport [38]. The role of SMXL proteins in root and root hair development has not been comprehensively addressed. Initial observations suggested mutations of SMXL678 suppress the enhanced lateral root density phenotype of max2 [27], while unexpectedly the increased root skewing phenotype, recently described for kai2 and max2 mutants was also suppressed by smxl678 [29]. These data have been used to propose the existence of non-canonical D14/KAI2 signalling cascades in the context of lateral root development and root skewing [10, 29].
In this study, we dissected the roles of SLs and KLs in the control of root development in Arabidopsis. We aimed to test the important hypothesis that root development might be mediated by non-canonical receptor-target interactions between D14, KAI2 and SMAX1/SMXL2, SMXL678. Our results show that KAI2 is much more important than previously realized in the regulation of root development, and that many effects previously attributed to SL signalling are actually mediated by KAI2 (and therefore KL signalling). We find no evidence for non-canonical receptor-target interactions, but conversely find surprising evidence of non-canonical GR24 ligand-receptor interactions in both KAI2 and D14 signalling.
SLs have previously been described to regulate primary root length (PRL), lateral root density (LRD) and root hair development [6, 8, 9, 42]. We re-assessed the specific roles of SL signalling in root development in mutants specifically affected in SL biosynthesis, namely the SL biosynthesis mutants max3-9, max4-5 and max1-1 (here arranged in pathway order). Surprisingly, we found that SLs only have subtle effects on root architecture. We observed decreased primary root length (PRL) and increased lateral root density (LRD) in SL biosynthesis mutants across many experiments, but rarely at the same time (summarized in S1 Fig). For instance, Fig 1A shows reduction in PRL relative to Col-0 in all SL biosynthesis mutants, but in the same experiment LRD was not altered (S1 Fig). Conversely, Fig 1B shows increased LRD in SL biosynthesis mutants relative to Col-0, but PRL was not altered in the same experiment (S1 Fig). Thus, consistent with previous reports [8], we found that SL signalling has subtle, and possibly mutually exclusive, effects on PRL and LRD of Arabidopsis, which appear to be sensitive to small differences in growth conditions.
We also examined root hair formation in the suite of SL biosynthesis mutants. Contrary to previous assumptions [7] we found that neither root hair density (RHD) nor root hair length (RHL) are altered in any of the SL biosynthesis mutants (Fig 2A, 2C and 2D). Thus, the previously observed root hair phenotypes of max2 mutants must be caused by defects other than SL signalling, for example in KL signalling.
The phenotypes present in SL-specific biosynthesis mutants are insufficient to account for previously described effects of max2 on root development. We therefore hypothesized that KAI2 signalling may play an important role in the regulation of root and root hair development, and we therefore compared and contrasted root development in d14 and kai2 mutants. In the case of LRD, we observed that d14-1 causes increased LRD and/or reduced PRL, consistent with the phenotypes of SL biosynthesis mutants (Fig 1A–1D). We also observed that two allelic kai2 mutants (kai2-1, kai2-2) in the Col-0 background, showed increased LRD of around the same magnitude as d14-1 (Fig 1D, S2A Fig), with no clear effect on PRL (Fig 1C). This phenotype in kai2 was particularly evident at 6dpg, and became less evident at later time points. For d14, the opposite pattern was seen, and the LRD phenotype only became evident at later time points (Fig 1D, S2B Fig). Thus, at least some of the confusion about the role of these pathways in regulation of lateral root development may result from the staging of experiments. Taken together, our results suggest that both SL and KL signalling regulate LRD in Arabidopsis. We further tested this idea by examining LRD in d14 kai2 double mutants. The d14-1 kai2-2 mutant showed a very strong and consistent increase in LRD in comparison to Col-0, d14-1 and kai2-2 (Fig 1D, S2B Fig). The increase in LRD was always greater in d14-1 kai2-2 than in the single mutants (Fig 1D). Thus, both KL and SL signalling regulate LRD in an additive manner, possibly by affecting lateral root development at different developmental stages and time points.
Given the lack of root hair phenotype in SL biosynthesis mutants, we hypothesized that KAI2 and not D14 signalling would regulate root hair development. Consistent with this hypothesis, we observed no RHD or RHL phenotype in d14-1 (Fig 2B–2F). Conversely, RHD and RHL were strongly decreased in two allelic kai2 mutants in Col-0 as well as Ler, and they perfectly phenocopied the root hair phenotype of max2 mutants (Fig 2B, 2E–2H). Thus, the root hair phenotypes previously observed in max2 mutants and attributed to the lack of SL signalling are actually caused by a lack of KL signalling. To confirm this, we assessed whether root hair development can be influenced by exogenous addition of karrikin. Treatment with 1 μM KAR2 increased RHD and RHL relative to control treatments in a KAI2 and MAX2-dependent manner (Fig 2G and 2H), corroborating the role of KL-signalling in promoting root hair development.
In addition to lateral root and root hair phenotypes, we observed that kai2 mutants display increased skewing along the surface of vertically-oriented agar plates, in the Col-0 and in the Ler ecotype (Fig 3A–3D, S3 Fig), consistent with a recent report that described this phenotype in kai2 mutants in Ler [29]. This right-handed skewing is a well-established effect of growing Arabidopsis roots on the surface of agar plates, and probably arises from a combination of circumnutation and thigmotropic responses [43, 44]. Increased skewing is also observed for max2 mutants, but not for SL biosynthesis mutants, nor d14 (Fig 3B and 3C; S3A Fig). The skewing phenotype of the d14-1 kai2-2 double mutant in the Col-0 background is equal to kai2-2 (Col-0), confirming that SL perception is not involved in regulating root growth direction (Fig 3C).
The increased skewing in the kai2 and max2 mutants is accompanied by increased root waving, which is displayed as a decrease in root ‘straightness’ (Fig 3A, 3E and 3F, S3B Fig). Again, this waving phenotype is not observed in d14-1 or SL biosynthesis mutants (Fig 3E, S3B Fig). The waving phenotype is separable from the skewing phenotype, and growth on plates inclined at 45° generally increases waving relative to plates grown at 90°, while altering skewing only in the Ler but not in the Col-0 wild type (S3C–S3G Fig).
Skewing is often associated with epidermal cell file rotation [44]. To determine whether skewing of kai2 and max2 mutants is associated with cell file rotation [45], we inspected epidermal cells between 2 and 3mm above the root tip in kai2 mutants. Cell length was reduced in kai2 and max2 mutants relative to wild-type in both Col-0 and Ler backgrounds (with a concomitant increase in cells/mm) (Fig 4A and 4C, S4A and S4C Fig). However, a careful microscopic inspection of the root surface of kai2 and max2 mutants did not show any signs of epidermal cell file rotation, instead they were clearly vertically orientated (Fig 4B, S4B Fig). This is in contrast to the results of [29], who observed increased cell file rotation in kai2 and max2 mutants in Ler at a 45° growth angle. Since at a 90° growth angle we observed a skewing phenotype but no cell file rotation, we conclude that there is likely no connection between any cell file rotation phenotype in KL perception mutants and their skewing phenotype. Interestingly, also the SL perception mutant d14 displayed the short epidermal cell phenotype but had no skewing phenotype, clearly demonstrating that there is no connection between epidermal cell length and skewing in these receptor mutants (Fig 4A and 4C; S4A and S4C Fig).
It has also been speculated that a smaller root cell diameter in kai2 mutants may cause tissue tensions leading to skewing [29]. We also observed that kai2 mutants in both the Col-0 and Ler background had thinner primary roots than wild-type. Quantification of root diameter at 2.5 mm above the root tip confirmed that the primary roots of kai2 and max2 mutants but not of the d14 mutant are thinner than those of the wild type (Fig 4D, S4D Fig). This indicates that the regulation of root thickness is specific to KL signalling. However, we could genetically separate the thin root diameter from the skewing and waving phenotypes because the root diameter phenotype of max2 could be suppressed by smax1 without altering the waving phenotypes. Conversely, the max2 root diameter phenotype could not be suppressed by smxl2 alone, but smxl2 was sufficient to suppress the skewing phenotype (Fig 5A and 5C; S5A and S5C Fig). Thus, decreased root diameter is unlikely to cause the skewing and waving phenotypes in kai2 and max2 as previously suggested [29].
The mechanism by which KAI2 regulates root skewing has been proposed to include the non-canonical degradation of SMXL678 [29]. We tested this important hypothesis in more detail, by using different combinations of smxl alleles. We observed that, for skewing, smax1 or smxl2 were both independently sufficient to suppress the max2 phenotype (Fig 5A and 5B, S1 Table), indicating that skewing may be very sensitive to the stoichiometry of SMXL proteins or that SMAX1 and SMXL2 act in different cells. smax1 and smxl2 could not suppress the max2 waving phenotype individually, but in combination they were able to completely suppress this phenotype (Fig 5C and 5D, S1 Table), indicating that SMAX1 and SMXL2 act redundantly to promote waving. These results are thus consistent with SMAX1 and SMXL2 acting genetically downstream of KAI2 and MAX2 to regulate root growth patterns. Notably, the effect of kai2, smax1 and smxl2 on skewing was consistent between plants grown in Munich [M] and Leeds [L].
Consistent with the results of [29], we observed a reduction in skewing in smxl678 max2-1 relative to max2-1 in plants grown in Munich [M] (Fig 5E). However, this was not the case in Leeds [L], where root skewing was often increased in smxl678 relative to wild-type, and in which there was an additive increase in skewing in smxl678 max2-1 (Fig 5G). We also did not observe any suppression of the max2-1 waving phenotype in smxl678 (Fig 5F). Thus, our analysis of smxl678 mutants indicates that SMXL678 proteins likely do not act downstream of KAI2/MAX2 in the regulation of root growth patterns, but rather, that SMXL678 regulates skewing in parallel to the KAI2-SMAX1/SMXL2 pathway.
Previous results showed that the max2 LRD phenotype was suppressed in a smxl678 background but not in a smax1 background [27], suggesting that the max2 LRD phenotype arises solely from excess SMXL678 protein accumulation. Since our results show that both D14 and KAI2 regulate LRD, this would again imply non-canonical regulation of SMXL678 by KAI2. To again test this hypothesis, we re-examined the regulation of LRD using more recently-available smax1 smxl2 double mutants [35]. We found that smax1 smxl2 was as efficient in reducing LRD of max2 as smxl678 (Fig 6). However, consistent with a role of both SL and KL signalling in regulating LRD neither smax1 smxl2 nor smxl678 appeared to be completely epistatic to max2 (Fig 6). The most parsimonious explanation for these results is that the max2 LRD phenotype arises from the accumulation of both SMAX1/SMXL2 and SMXL678, and that SL and KL signalling act together in the regulation of LR development by their canonical pathways: SL signalling by promoting SMXL678 turnover, and KL signalling by promoting SMAX1 SMXL2 turnover.
We also assessed, whether regulation of RHD and RHL by KAI2 occurs through canonical or non-canonical signalling. For both RHD and RHL, we found that smax1 smxl2 have increased RHD and RHL, and are epistatic to max2-1 in both of these phenotypes. smxl2 but not smax1 single mutants display an increased RHL with respect to the wild type, suggesting that SMXL2 may be more important in regulating RHL than SMAX1. Conversely, smxl678 mutants have no RHD or RHL phenotype, and no effect on the max2 phenotype (Fig 7A–7F). This is consistent with our observation that kai2 and not d14 phenocopies the root hair phenotype of max2 and that root hair development is regulated by KL signalling under standard conditions.
As a final test for non-canonical signalling in root development, we examined ligand-receptor interactions, using the easily scorable, karrikin-responsive root hair phenotypes as a system. Exogenous application of rac-GR24 was previously shown to promote root hair elongation [7, 42]. In light of the effects of KAI2 mutations on root hair development, we hypothesized that rac-GR24, and in particular the GR24ent-5DS stereoisomer, would modulate RHD and RHL, in a manner dependent on KAI2 [34]. Similar to KAR2, rac-GR24 treatment increased both RHD and RHL in Col-0 (Fig 8A and 8B), and this effect was dependent on MAX2 as previously reported [7, 42]. However, unexpectedly, it was independent of KAI2, suggesting that rac-GR24 might promote RHD and RHL via D14 (Fig 8A and 8B). We assessed this in detail and quantified RHD and RHL after treatment with the pure stereoisomers GR245DS and GR24ent-5DS, which are thought to specifically activate D14 and KAI2, respectively [34]. We observed that both GR245DS and GR24ent-5DS promote RHD and RHL in the wild-type, but their effects in d14 and kai2 mutants were intriguingly divergent from expectations. In d14-1, only GR24ent-5DS promotes RHD (as expected), but both GR245DS and GR24ent-5DS promote RHL to a similar degree in kai2, suggesting that both can be perceived by KAI2 to promote RHL (Fig 3A and 3B). Furthermore, both stereoisomers cause increased RHD and RHL in kai2-2, although the ‘canonical’ D14 ligand GR245DS has a significantly stronger effect than GR24ent-5DS (Fig 8A and 8B). Neither stereoisomer promoted RHD and RHL in the d14-1 kai2-2 double and max2-1 mutants (Fig 8), confirming that no additional unknown receptor is involved in the response to rac-GR24. The first major implication of these results is that D14 can act to promote root hair development, when stimulated with ligand, even if that is not the standard function of D14 (Fig 2). The second major implication is that in roots, contrary to previous suggestions for the regulation of Arabidopsis hypocotyl elongation [34], D14 can perceive GR24ent-5DS ligands when KAI2 is absent, and KAI2 can perceive GR245DS ligands when D14 is absent.
Since these results are unexpected we wondered whether the GR24 stereoisomers we used are really pure and determined their purity by nuclear magnetic resonance (NMR), circular dichroism (CD) spectroscopy and polarimetry (S6 Fig). Both 1H-NMR, 13C-NMR and CD as well as rotation values determined by means of polarimetric measurements confirmed the purity of the compounds and recapitulated previously published NMR- and CD-spectra for (+)-5-Desoxystrigol and (–)-ent-5-Desoxystrigol [46, 47]. Since the stereoisomers are pure, we conclude that they do not specifically act through KAI2 or D14 but that both molecules can bind to and trigger both receptors in the context of root hair development.
Previous Arabidopsis hypocotyl elongation assays suggested specific roles of GR245DS and GR24ent-5DS in triggering D14- vs KAI2-mediated signalling respectively because GR245DS suppressed hypocotyl elongation specifically in kai2 mutants and GR24ent-5DS in d14 mutants [34]. We re-examined the effects of the GR24 stereoisomers on hypocotyl elongation (S7 Fig). Similar to root hair elongation and contrary to a previous report [34] the d14-1 mutant responds equally to GR245DS and GR24ent-5DS with a decrease in hypocotyl growth, showing that in the hypocotyl KAI2 can mediate responses to both molecules. The kai2-1 mutant also responds to both molecules but to a lesser extent to GR24ent-5DS, suggesting together with the above results that D14 is more effective in mediating responses to its previously suggested ligand GR245DS than to GR24ent-5DS [34]. Similar to root hair development, the d14-1 kai2-2 double mutant and the max2-1 mutant do not respond to any molecule in this assay, confirming that in the hypocotyl response to the GR24 stereoisomers no additional receptor is involved. In summary, we show that GR245DS and GR24ent-5DS can activate both signalling through KAI2 and through D14 in the regulation of RHL as well as hypocotyl elongation.
Root systems flexibly adapt their architecture and morphology to heterogeneous soil environments and to the physiological needs of the plant. A network of plant hormone signalling pathways is essential for translating environmental signals and physiological states into developmental outputs [48]. Strigolactones (SLs) have been assumed to play an important role in modulating root development [7–9]. Here we demonstrate that under standard growth conditions KL signalling plays a much larger role than SL signaling in shaping root and root hair development (Fig 9).
Previous reports showed increased LRD in max2 and suppression of lateral root emergence by rac-GR24 [8, 9]. Our study indicates that these effects are mediated through both the KAI2 and D14 signalling pathways, in an additive manner. We observed that lateral root density (LRD) is consistently higher in kai2 mutants than wild type (particularly at earlier time points). We found SL biosynthesis and perception mutants also displayed subtle changes in root architectural parameters, such as primary root length (PRL) and LRD. In a range of experiments with SL mutants, we either observed strongly decreased PRL or strongly increased LRD, but not both phenotypes together. This suggests that the effects of SL signalling on PRL or LRD are to some extent mutually exclusive, and that expression of one phenotype reduces expression of the other, which may explain some of the previous contradictory reports regarding effects of SLs on root development [8, 9]. We also found that the time after germination matters for the LRD phenotypes. Thus, confusion about the role of SLs in LR development may also reflect differences in the physiological timing of observations within experiments. The d14 kai2 double mutant showed a much larger increase in LRD compared to the single mutants, indicating that both signalling pathways contribute additively to modulating LRD, and that previously reported max2 phenotypes reflect a lack of both signalling pathways. This is further supported by suppression of the max2 LRD phenotype by mutants in both the targets of KL signalling (SMAX1/SMXL2) and SL signalling (SMXL678).
A major finding of our work is the important role of KL signalling in root hair development. Root hair density (RHD) and root hair length (RHL) are strongly reduced in kai2 and max2 mutants and increased in smax1 smxl2 mutants, as well as by karrikin treatment of wild type roots. Our results thus present compelling evidence that KL signalling is a key regulator of root hair development. KAI2 being a major regulator of root hair development rather than D14 seems to make sense from an evolutionary point of view. Root hair development and tip growth in Arabidopsis rely on conserved functions and genes that also operate in the development of rhizoids of Marchantia polymorpha gametophytes, which appear to be homologous to root hairs [49–51]. D14 occurs only in genomes of seed plants while KAI2 is already present in algae [19, 20, 22]. Thus, it is possible that KAI2-SMAX1 module is part of an ancient and conserved pathway regulating tip growth of epidermal cells.
We did not find any impact of d14 and SL biosynthesis mutants on root hair development in our study. However, we found that D14 signalling can be triggered to promote root hair development, if the correct ligand is present and KAI2 is absent. This is very similar to the hypocotyl, in which D14-mediated SL perception can regulate hypocotyl elongation, but is not actually required to do so [19, 34]. This suggests that there may be a role for D14 signalling in root hair development under certain environmental conditions, when SL levels are very high, for example under phosphate starvation [52]. Previous studies, [53, 54] found a small decrease in RHD of the SL biosynthesis mutant max4-1, which could be rescued by adding GR245DS [54]. This is inconsistent with our observations here, but might reflect differences in the growth conditions used, and indeed these studies used low phosphate media. Thus, further investigation of the role of D14 signalling in environment-dependent root hair development is warranted.
No single signalling pathway for control of root skewing and straightness has been identified, but several studies have exposed different pathways impinging on these root behaviors (reviewed in Roy and Bassham. 2014). The activities of multiple hormones, such as auxin and ethylene, are among the candidates [55, 56]. Here we demonstrate that KL signalling is a novel regulator of root skewing and root straightness. The increased skewing and waving phenotypes of KL perception mutants were found in both the Col-0 and Ler background although Ler shows an intrinsically higher tendency to skew than Col-0. Our results are broadly consistent with the recent report of [29], but our interpretation of the cause of the phenotype differs. Swarbreck et al. [29], speculated that skewing may be caused by increased epidermal cell file rotation and/or smaller root diameter of kai2 mutants. Under our conditions, we did not observe epidermal cell file rotation in kai2 and max2, but rather shorter epidermal cells. Since both kai2 and d14 have a reduced epidermal cell length, but skewing only occurs in kai2, we conclude that epidermal cell length is not related to skewing. Interestingly, in the experiments in which epidermal cell length was inspected, PRL was not significantly altered. This implies that a compensatory increase in epidermal cell division must occur in both the KL and SL perception mutants, which would be consistent with increased cell division in the primary root meristem. Alternatively, the epidermal cell length may differ among different root zones thus compensating for the shorter epidermal cell length in the zone 2–3 mm above the root tip. We also show that the reduced root diameter of KL perception mutants does not cause either skewing or waving since smax1 alone suppresses the root diameter but not the waving phenotype of max2, and smxl2 suppresses the max2 skewing but not the root diameter phenotype.
We have previously highlighted some phenotypic characteristics suggesting that KL and SL signalling in the root might not act through the canonical KAI2-SMAX1 and D14-SMXL678 receptor-target pairs [10]. The main reason for this suggestion was that max2 mutants had stronger LRD phenotypes than SL biosynthesis mutants [7–9], which suggested that KAI2 regulates lateral root emergence rather than or in addition to D14, while mutations of the genes encoding the canonical SL signalling targets SMXL678 were able to completely suppress the max2 LRD phenotype with smax1 being unable to do so [10, 27]. Similarly, Swarbreck et al. [29] suggested that non-canonical signalling may occur in skewing responses, since smxl678 mutants can completely suppress the max2 skewing phenotype, which arises solely through lack of KAI2 signalling.
We have now robustly tested this hypothesis, and find no evidence for non-canonical KL and SL signalling in roots under our growth conditions. Using smax1 smxl2 double mutants, we show that every effect of loss of KAI2 activity can be suppressed by loss of SMAX1 and SMXL2 (or only one of the two), and that similarly, all effects of loss of D14 activity can be suppressed by loss of SMXL678. In the case of LRD, we show that smax1 smxl2 mutants can suppress the phenotype of max2, demonstrating that the canonical KL signalling targets are involved in regulating lateral root emergence and that SMXL2 compensates for the absence of functional SMAX1 in lateral root development [27]. The suppression of the max2 LRD phenotype by smxl678 as well as smax1 smlx2 is consistent with our observation that both D14 and KAI2 regulate LRD. Thus, the accumulation of both SMAX1/SMXL2 and SMXL678 contributes to max2 LRD phenotypes and there is no need to invoke non-canonical receptor-target pairs to explain the effects of KAI2 and D14 on LRD.
We also reject the idea that KL signalling regulates skewing through SMXL678 [29]. We find that smxl2 mutations are sufficient to suppress skewing in max2, consistent with canonical KAI2-SMAX1/SMXL2 signalling acting in this response. It is certainly interesting that smxl678 mutants suppress skewing of max2 under some conditions, which does not reflect any known effect of D14 signalling. However, we show that this phenotype is highly variable, and under our growth conditions in Leeds, smxl678 mutants actually increased root skewing additively with max2. Thus, although SMXL678 can certainly regulate skewing, this appears to be unrelated to the clearly defined and consistent effect of KL signalling on skewing. In fact, it appears consistent with the observation that rac-GR24 treatment–which stimulates SMXL678 degradation–causes an increase in root skewing in the wild type [29]. The location-dependent skewing behaviour of smxl678 mutants suggests that the role of SMXL678 in skewing may strongly depend on environmental conditions, and it will be interesting to identify the mechanisms underlying this phenomenon in the future.
The case is even more clear-cut for RHL, RHD, root straightness and root diameter, for which only kai2 and max2 mutants show a phenotypic difference to wild type, and which can only be suppressed by mutating SMAX1 and SMXL2. Interestingly, the smxl2 mutant alone has longer root hairs than wild-type showing for the first time a phenotype in which SMXL2 plays a more important role than SMAX1, although it is alone not sufficient to suppress the max2 phenotype. In the case of root diameter, mutation of SMAX1 is sufficient to suppress the max2 phenotypes (S2 Table). This partial redundancy of SMAX1 and SMXL2 is also seen in seed germination, hypocotyl growth and leaf shape [27, 35]. This likely arises from different expression patterns of the two genes: in tissues where only one of the two proteins is expressed, removing this one is sufficient to suppress the phenotype. Conversely, in the case of skewing, removing either SMAX1 or SMXL2 alone suffices to suppress the max2 phenotype (S1 Table), suggesting that skewing is particularly sensitive to SMAX1/SMXL2 levels or stoichiometry or that SMAX1 or SMXL2 regulate skewing in different tissues.
In contrast to the lack of evidence for non-canonical receptor-target interactions, we uncovered unexpected evidence for non-canonical receptor ligand interactions in the context of root development. The two stereoisomers of rac-GR24, GR245DS and GR24ent5DS have been suggested to specifically activate D14 and KAI2 respectively in the regulation of hypocotyl growth [34]; and GR24ent5DS showed only a very low efficiency in inhibiting shoot branching in Arabidopsis and rice [34, 57]. However, our study shows that there is very little specificity of the two receptors for the two stereoisomers, as both d14 and kai2 mutants respond to both with increased RHL and even with decreased hypocotyl elongation. This result is strengthened by confirming the purity of the employed compounds via NMR and CD. It has been shown by differential scanning fluorimetry (DSF) in vitro that D14 can bind both GR245DS and GR24ent-5DS but KAI2 only bound GR24ent-5DS [31]. However, the situation in vivo may be different and binding of both ligands to both α/β hydrolase receptors D14 and KAI2 may be stabilized through receptor protein complexes. Although binding of the ‘wrong’ stereoisomer to the α/β hydrolase receptor may be less efficient than binding of the ‘correct’ one, it may suffice to trigger developmental responses, which are very sensitive to removal of SMXL proteins, or which may require additional interaction partners in the receptor complex that stabilize the complex in presence of the hypo-specific ligand. Independent of the mechanism, our results show that GR245DS and GR24ent-5DS cannot safely be used to specifically trigger D14 and KAI2-mediated signalling, respectively. This also implies that the community urgently needs an affordable synthetic SL that triggers D14 in a highly specific manner.
Overall our results show that KL signaling and therefore SMAX1 and SMXL2 play an important role in controlling root architecture and root hair development (Fig 9). However, some traits such as LRD and epidermal cell length are regulated by both SMAX1/SMXL2 and SMXL678. Key challenges for future studies will be to understand how exactly SMXL proteins regulate root architecture. Ruyter-Spira et al. [8] previously suggested that the impact of SLs on root development might be best understood as a reflection of their effect on the auxin landscape, and we hypothesize that this may also be the case for KAI2 signalling. Most of the traits we have examined are known to be regulated by auxin, and SL signalling in the shoot is known to modulate auxin transport by regulating PIN protein abundance [27, 58]. Thus, it is very possible that the KAI2-SMAX1/SMXL2 and D14-SMXL678 pairs regulate the auxin landscape of the root, for example by controlling the abundance of auxin transport proteins. Such a scenario might underlie the variability in phenotypes observed in the mutants in our study (for instance, the strong variation in smxl678 skewing phenotype), since environmental parameters such as light or temperature are known to affect endogenous auxin levels [59, 60].
We do not currently know enough about the upstream inputs into the KL signalling pathway to understand the aetiology of KAI2-induced root development, but undoubtedly the phenotypes described here will provide important clues and tools in this regard. SL production increases in several plant species upon phosphate starvation [12, 61–63] and the effect of SL biosynthesis on root architecture was suggested to depend on the sucrose level in the medium and thus on the carbon-status of the plants [8], but it is yet unknown whether KL signalling is also influenced by mineral nutrient levels. However, expression of KAI2 does respond to light conditions, and thus KL signalling could potentially integrate light cues into root development [64]. Indeed, it is likely that both signalling pathways are influenced by multiple abiotic and perhaps biotic stimuli, and it will be exciting to learn how SL and KL signalling tune root development to environmental conditions.
Arabidopsis thaliana genotypes were in Columbia-0 (Col-0) or Landsberg erecta (Ler) parental backgrounds. The following mutants were used: Ler: max2-8 [18], kai2-1, kai2-2 [18], Col-0: kai2-2 [28], max3-9 [65], max4-5, d14-1 kai2-2 [66], d14-1 [19], max1-1, max2-1, max2-2 [67], smax1-2, max2-1 smax1-2 [37], smax1-2 smxl2-1, max2-1 smax1-2 smxl2-1 [35], smxl6-4 smxl7-3 smxl8-1, max2-1 smxl6-4 smxl7-3 smxl8-1 [27].
For analysis of root growth, Arabidopsis thaliana seeds were grown in axenic conditions on 12x12cm square plates containing 60 ml agar-solidified medium. Seed were surface sterilized either by vapour sterilization, or by washing with 1 ml of 70% (v/v) ethanol and 0.05% (v/v) Triton X-100 with gentle mixing by inversion for 6 minutes at room temperature, followed by 1 wash with 96% ethanol and 5 washes with sterile distilled water. For primary root length and lateral root density plants were grown in Cambridge and Leeds on plates containing ATS medium [68] supplemented with 1% sucrose (w/v) and solidified with 0.8% ATS. For measurements of skewing, waving, cell length, root diameter, root hair density and root hair length, seedlings were grown in Munich on plates containing 0.5X Murashige & Skoog medium, pH5.8 (½ MS) (Duchefa, Netherlands), supplemented with 1% sucrose and solidified with 1.5% agar. Plates were stratified at 4°C for 2–3 days in the dark, and then transferred to a growth cabinet under controlled conditions at 22°C, 16-h/8-h light/dark cycle (intensity ~120 μmol m-2 s-1). Unless otherwise indicated, the plates were placed vertically.
rac-GR24 was purchased from Chiralix (Nijmegen, The Netherlands), GR24ent5DS and GR245DS from Strigolab (Turin, Italy), and KAR2 from Olchemim (Olomouc, Czech Republic). For treatment with rac-GR24, GR24ent5DS or GR245DS, 1 mM stock solutions were prepared in 100% acetone. KAR2 was dissolved in 70% methanol for the preparation of 1 mM stock. The volume required to reach the final concentration of these different stock solutions was added to molten media prior to pouring Petri dishes. In each experiment, an equivalent volume of solvent was added to Petri dishes for untreated controls.
For quantification of primary root length and lateral root number, seedlings were grown as described above in Cambridge and Leeds for 10 days post germination (dpg). This allowed for the emergence of lateral roots sufficient for quantification in wild-type seedlings. A dissecting microscope was used to count emerged lateral roots in each root system, and images of the plates were then taken using a flatbed scanner. Primary root length was quantified using Image J. Separate experiments were primarily used to assess root skewing (see below), but root skewing angles were also measured from these images generated in these experiments.
The root slanting assay was modified from the method described by [69]. Arabidopsis seedlings were grown in Munich under the conditions described above (except for Fig 8G for which plants were grown in Leeds). Images were taken 5 days post germination (dpg) using an Epson Perfection V800 Pro Scanner. Images were analysed using the Simple Neurite Tracer plug-in of Fiji (https://imagej.net/Fiji/Downloads) to determine the following parameters as illustrated in Fig 4; root length (L), ratio of the straight line between the hypocotyl-root junction and the root tip (Lc), and vertical axis (Ly). These measurements were taken from at least 60 individual roots per genotype and used to calculate the root skewing angle (α) and root straightness (Lc/L) as previously described [70, 71].
Root hair growth was examined in Munich on the same Arabidopsis roots, which were used for determining root skewing and straightness. Images were taken at 2 mm from the root tip of a minimum of 8 roots per genotype and treatment with a Leica DM6 B microscope equipped with a Leica DFC9000 GT camera. The number of root hairs was determined by counting the root hairs between 2 and 3 mm from the root tip on each root, and root hair length was measured for 10–18 different root hairs per root using Fiji. The root hair position was determined following the method described by [72] for 5–15 root hairs per root and a minimum of 8 roots per genotype.
Using the same images as for root hair quantification, root diameter, root cell length and number of cells were analysed in Munich using Fiji. Root diameter was measured at 2.5 mm from the root tip. The number of cells was defined as the number of epidermal cells that crossed a 1-mm-long straight line drawn between 2 to 3 mm from the root tip. Root cell length was measured for at least 10 different epidermal cells per individual root in a minimum of 10 roots per genotype, between 2 to 3 mm from the root tip.
Statistical analyses were performed in R-studio, using one-way Analysis of Variance (ANOVA), followed by Tukey HSD or Dunnett´s post hoc test.
Sequence data for the genes mentioned in this article can be found in The Arabidopsis Information Resource (TAIR; https://www.arabidopsis.org) under the following accession numbers: MAX3, AT2G44990; MAX4, AT4G32810; MAX1, AT2G26170; D14, AT3G03990; KAI2, AT4G37470; MAX2, AT2G42620; SMAX1, AT5G57710; SMXL2 AT4G30350; SMXL6, AT1G07200; SMXL7, AT2G29970; SMXL8, AT2G40130.
|
10.1371/journal.ppat.1007816 | Host-specialized fibrinogen-binding by a bacterial surface protein promotes biofilm formation and innate immune evasion | Fibrinogen is an essential part of the blood coagulation cascade and a major component of the extracellular matrix in mammals. The interface between fibrinogen and bacterial pathogens is an important determinant of the outcome of infection. Here, we demonstrate that a canine host-restricted skin pathogen, Staphylococcus pseudintermedius, produces a cell wall-associated protein (SpsL) that has evolved the capacity for high strength binding to canine fibrinogen, with reduced binding to fibrinogen of other mammalian species including humans. Binding occurs via the surface-expressed N2N3 subdomains, of the SpsL A-domain, to multiple sites in the fibrinogen α-chain C-domain by a mechanism analogous to the classical dock, lock, and latch binding model. Host-specific binding is dependent on a tandem repeat region of the fibrinogen α-chain, a region highly divergent between mammals. Of note, we discovered that the tandem repeat region is also polymorphic in different canine breeds suggesting a potential influence on canine host susceptibility to S. pseudintermedius infection. Importantly, the strong host-specific fibrinogen-binding interaction of SpsL to canine fibrinogen is essential for bacterial aggregation and biofilm formation, and promotes resistance to neutrophil phagocytosis, suggesting a key role for the interaction during pathogenesis. Taken together, we have dissected a bacterial surface protein-ligand interaction resulting from the co-evolution of host and pathogen that promotes host-specific innate immune evasion and may contribute to its host-restricted ecology.
| Many bacterial pathogens are specialized for a single host-species and rarely cause infections of other hosts. Our understanding of the bacterial factors underpinning host-specificity are limited. Here we demonstrate that a canine host-restricted bacterial pathogen, Staphylococcus pseudintermedius, produces a surface protein (SpsL) that has the ability to preferentially bind to canine fibrinogen with high strength. This host-specific interaction has evolved via binding to a tandem repeat region of the fibrinogen α-chain which is divergent among mammalian species. Importantly, we found that the strong binding interaction with canine fibrinogen promotes bacterial aggregation and biofilm formation as well as inhibiting neutrophil phagocytosis. Our findings reveal the host-adaptive evolution of a key bacterium-host interaction that promotes evasion of the host immune response.
| Many bacteria evolve strict mutualistic relationships with their host species with limited capacity to colonize and cause disease in other hosts. In contrast, other bacteria have the ability to expand into new host-species leading to the emergence of new pathogenic clones. Our understanding of the bacterial and host factors that underpin pathogen-host ecology is very limited. However, bacterial surface proteins are central mediators of host colonization, and tissue tropism and, as such, are likely to play a critical role in determining host ecology [1, 2]. For example, the choline-binding protein A, of the major human pathogen Streptococcus pneumoniae, binds to polymeric immunoglobulin receptor, secretory component, secretory IgA, and factor H of complement from humans but not from other animal species tested [1]. In addition, the human host-restricted Streptococccus pyogenes expresses surface-anchored M protein that binds exclusively to human CD46 mediating binding and invasion of epithelial cells. Adaptive diversification of bacterial surface proteins can also have a major impact on tissue tropism and disease manifestation. For example, uropathogenic Escherichia coli virulence has arisen due to mutations in the fimbrial adhesin FimH, promoting high affinity binding to the urinary epithelium [3]. Similarly, a single non-synonymous mutation in a fibronectin-binding autolysin of Staphylococcus saprophyticus, associated with a selective sweep, has been linked to the pathogenesis of urinary tract infection in humans [4]. Additionally, single amino acid substitutions in the fibronectin-binding protein A (FnBPA) of Staphylococcus aureus, are associated with cardiac device infections and bacteremia in humans due to increased binding affinity for fibronectin [5–7].
Fibrinogen is a highly abundant protein in blood and is required for blood coagulation, thrombosis and host immune defense [8]. This large glycoprotein is composed of three chains, termed the α-, β-, and γ-chains, that form a dimer of trimers [8]. During coagulation, thrombin cleaves the fibrinogen α- and β-chains allowing fibrin formation, with the γ-chain binding directly to platelets to produce the blood clot [8]. Bacterial pathogens have evolved many mechanisms to bind to host fibrinogen to disrupt blood coagulation as well as promote host cell adherence, immune evasion and abscess formation [9, 10]. The importance of this interaction is highlighted by the large number of fibrinogen-binding proteins of bacteria that have been identified, with S. aureus encoding at least 9 fibrinogen-binding proteins [9–12]. It is unclear if each of these proteins confer an exclusive function via distinct fibrinogen-binding sites, or if convergent evolution is driving a high redundancy for fibrinogen-binding. In S. aureus there are fibrinogen-binding proteins that exhibit host-specificity and those that exhibit a broader host tropism. In the case of clumping factor B (ClfB) a host-restrictive fibrinogen-binding phenotype is observed due to the interaction with a sequence unique to the human fibrinogen α-chain [13]. Conversely, clumping factor A (ClfA) interacts with fibrinogen from multiple hosts, such as human, canine, and murine, due to an interaction with the fibrinogen γ-chain [14]. A single residue substitution of Q407A in the ovine fibrinogen γ-chain is sufficient to eliminate binding of ClfA to ovine fibrinogen [14]. As FnBPA adheres to the same region in the fibrinogen γ-chain, it is assumed that it exhibits the same host phenotype but this has not been directly investigated [15].
Staphylococcus pseudintermedius naturally colonizes the nares and perineum of healthy dogs and is a major cause of canine skin infections, particularly in canine breeds that are genetically pre-disposed to atopic dermatitis including boxers, German shepherds, golden retrievers, Dalmatians, Labrador retrievers, French bulldogs, West Highland white terriers, Jack Russell terriers, and shar-peis [16–19]. Although S. pseudintermedius occasionally causes zoonotic infections of humans via dog bite wounds, it is highly host-restricted and there is limited evidence for colonization or transmission among non-canine host species such as humans and cats [17]. Of note, S. pseudintermedius demonstrates a host-specific preference for corneocytes collected from healthy dogs in comparison to human healthy volunteers [20], suggesting the existence of bacterial surface factors that promote a canine host-tropism. However, the bacterial factors underpinning the canine host-restricted ecology of S. pseudintermedius are unknown. Previously, we identified a complement of 18 genes encoding cell wall-associated proteins of S. pseudintermedius strain ED99 and discovered 2 fibrinogen-binding proteins, SpsD and SpsL, which exhibited host-dependent variation in fibrinogen-binding after heterologous expression by Lactococcus lactis [21]. While SpsD was encoded by closely-related species of staphylococci associated with non-canine host-species, SpsL was specific for S. pseudintermedius [21] and was shown to be required for abscess formation in a murine skin infection model [22]. Previous sequence and structural analysis of SpsL identified a similar domain architecture to the fibronectin-binding proteins of S. aureus with an amino acid identity of ~27% [21]. In S. pseudintermedius strain ED99, SpsL is a protein of 930 amino acids that contains the typical signatures of a cell wall-associated protein with an N-terminal signal peptide and C-terminal LPKTG anchor motif [21]. The N-terminal of SpsL consists of an A-domain with N1, N2, and N3-subdomains that would be predicted to mediate fibrinogen-binding [21]. The C-terminal contains 7 tandem repeats that share 91–100% pairwise identity and 24% protein identity to the fibronectin-binding repeats of FnBPA [21]. These C-terminal repeats of SpsL have been demonstrated to confer fibronectin-binding that can mediate internalization of S. pseudintermedius into canine epithelial cells [23].
Here we dissect the interaction of SpsL with fibrinogen and demonstrate its functional consequences. Using atomic force microscopy (AFM) we quantify the enhanced binding force of SpsL for canine fibrinogen, which is dependent on a tandem repeat region in the fibrinogen α-chain that is genetically diverse in mammalian species. Further, we demonstrate that the strong host-specific interaction with canine fibrinogen is required for SpsL-mediated aggregation and biofilm formation, and promotes neutrophil opsonophagocytosis suggesting a key role for SpsL-fibrinogen binding in the pathogenesis and canine host ecology of S. pseudintermedius.
Previously, heterologous expression of SpsL in L. lactis revealed a host-dependent binding to fibrinogen [21]. In order to investigate this preliminary observation, we examined the capacity of wild-type S. pseudintermedius strain ED99 to bind to immobilized fibrinogen from different host species. At mid-exponential growth phase, the highest binding was observed to canine and human fibrinogen, with very limited binding to bovine and ovine fibrinogen (Fig 1A). A S. pseudintermedius ED99 mutant deficient in expression of a fibrinogen-binding protein SpsD (ED99ΔspsD), cultured to early-exponential growth phase, demonstrated binding to fibrinogen that was equivocal to wild-type ED99 for canine, ovine and human fibrinogen but reduced for bovine fibrinogen (p<0.001) (Fig 1B). In contrast, a mutant deficient in SpsL (ED99ΔspsL), cultured to mid-exponential growth phase, exhibited highly reduced binding to fibrinogen from all host species (p<0.001) (Fig 1C), with complete loss of fibrinogen-binding by a mutant deficient in both SpsL and SpsD (ED99ΔspsLΔspsD), at both early-exponential (Fig 1B) and mid-exponential growth phases (Fig 1C). Re-introduction of the deleted spsL gene restored fibrinogen-binding as did complementation of ED99ΔspsLΔspsD with a plasmid (pALC2073::spsL) encoding SpsL (Fig 1C and 1D). In summary, these data indicate that S. pseudintermedius ED99 has host-specific interactions with fibrinogen that are primarily mediated by SpsL. However, these adherence assays do not allow quantification of the strength of the binding interaction between SpsL and fibrinogen.
In order to compare the molecular forces driving the binding of SpsL to canine and human fibrinogen, we used atomic force microscopy (AFM) [24, 25]. Firstly, for single-cell force spectroscopy (SCFS, Fig 2A), single bacteria were attached onto AFM cantilevers, and force-distance curves were collected between the cell probes and fibrinogen-coated surfaces (Fig 2A). The adhesion forces obtained for three representative cells interacting with either human or canine fibrinogen are presented (Fig 2A). While there was substantial variation between cells, the binding probability was always higher for canine rather than for human fibrinogen (85% vs 56%; means ± 12 and 28, from a total of n = 1,139 and 1,176 curves). Also, binding forces were stronger for canine fibrinogen (355 ± 354 pN from n = 228 adhesive curves; 2,077 ± 1,157 pN (n = 388), and 1,024 ± 427 pN (n = 352), for cell #1, cell #2 and cell #3, respectively), than for human fibrinogen (149 ± 84 pN (n = 85), 744±467 pN (n = 362), and 541±266 pN (n = 216)). Next, we used single-molecule force spectroscopy (SMFS) with fibrinogen-coated AFM tips to quantify the strength of single bonds (Fig 2B and 2C). Canine fibrinogen (Fig 2B) always showed very large forces (1,237 ± 754 pN from n = 258 adhesive curves; 1,554 ± 828 pN (n = 308), and 2,630 ± 1,393 pN (n = 137), for cell #1, cell #2 and cell #3, respectively). Of note, these high forces are in the range of values reported previously for the high-affinity “dock, lock and latch” binding of SdrG to fibrinogen [26]. For human fibrinogen, these strong forces were also observed but much less frequently (Fig 2C). Taken together these data are consistent with a high affinity dock, lock and latch-based mechanism for the binding of SpsL to canine fibrinogen.
To investigate the region of SpsL required for fibrinogen-binding, an array of recombinant truncates of SpsL were generated and purified from E. coli as described in the Supplemental Methods section (S1A Fig). From structural and functional studies of related staphylococcal surface proteins, we predicted that the N2N3 subdomains of SpsL would be sufficient for fibrinogen-binding [27–29]. However, in ELISA-like assays, none of the purified recombinant truncates of SpsL exhibited binding to fibrinogen, with all peptides tested demonstrating binding equivocal to the negative control (fibronectin-binding domain of SpsD) (S1B–S1D Fig). In contrast, fibronectin-binding could be detected in the SpsL recombinant protein construct containing a single fibronectin-binding repeat (S1E Fig) [23]. Similarly, full-length or truncated SpsL A-domain fragments expressed and purified from S. pseudintermedius ED99ΔspsLΔspsD supernatant did not adhere to canine fibrinogen, in contrast to a positive control of recombinant SpsD N2N3 purified from E. coli (S1F Fig). However, heterologous overexpression of SpsL on the surface of a fibrinogen-binding deficient S. aureus strain (SH1000ΔclfAΔclfBΔfnbAΔfnbB) [30] promoted high levels of adherence to canine fibrinogen (S1G Fig). Taken together, these data suggest that SpsL requires bacterial cell surface attachment to mediate fibrinogen-binding. Accordingly, subsequent experiments employed SpsL constructs expressed on the surface of S. pseudintermedius ED99.
SpsL fragments representing the A-domain and N2N3 subdomains were expressed on the surface of the S. pseudintermedius fibrinogen-binding deficient mutant ED99ΔspsLΔspsD (Fig 3A). As reported for other bacterial cell wall-associated proteins, we considered that the C-terminal repeat region may be required to project the fibrinogen-binding domain from the cell surface, and that a small region of the N1 subdomain may be required for secretion and cell surface expression [31, 32]. To address this issue, chimeric proteins were generated that replace the SpsL fibronectin-binding repeats with the ClfA SD repeats (that do not exhibit any known ligand-binding activity) [33] and that contain a 21 amino acid region of the N1 subdomain (N121; 181 VSKEENTQVMQSPQDVEQHVG 201) (Fig 3A). Analysis of the binding of these constructs to immobilized fibrinogen and expression analysis by Western blot indicated the requirement for the N121 peptide for cell surface expression, with the N2N3+SD and N2N3 constructs not expressed on the cell surface (Fig 3B and 3C). Importantly, both the chimeric A-domain (A+SD) and N2N3-subdomain (N121+N2N3+SD) proteins exhibited binding to canine fibrinogen that was equivocal to the full length SpsL protein (Fig 3B). The binding of the chimeric N121+N2N3+SD protein to fibrinogen from multiple host species indicate that the SpsL N2N3 subdomains are sufficient for host-specific fibrinogen-binding, with the 21 amino acids of the N1 subdomain required for cell surface expression, (Fig 3D) suggesting that SpsL mediates ligand-binding in a manner analogous to the dock, lock and latch-binding mechanism described for other staphylococcal cell wall-associated proteins [34]. To investigate this further, we modelled the structure of SpsL N2N3 subdomains, based on the crystal structure of ClfA (pdb 1N67) [35]. The structural model predicted classical DE-variant IgG folds made up of β-sheets typical of staphylococcal fibrinogen-binding proteins (S2A Fig). From this model we identified a putative latch region, 502 NSASGSG 508, required for the dock, lock and latch binding mechanism (S2A Fig). Deletion of this putative latch region in a surface-expressed SpsL construct had no effect on surface expression or fibronectin-binding but abrogated adherence to both canine and human fibrinogen (p<0.001) (Fig 3E and S2B Fig). In addition to the AFM data, these results, suggest that the SpsL N2N3-subdomains expressed on the bacterial surface mediate fibrinogen-binding via a mechanism analogous to the dock, lock, and latch binding model.
Staphylococcal proteins have evolved the ability to bind fibrinogen through multiple distinct interactions with different regions of host fibrinogen [13, 14, 36]. Previously it has been identified that S. pseudintermedius strain 326 is capable of binding to the fibrinogen α-chain with binding to the β-, and γ-chains not investigated [37]. To identify the binding site of SpsL, recombinant versions of the α-, β-, and γ-chains of human fibrinogen were expressed in and purified from E. coli and employed in bacterial binding assays. Both S. pseudintermedius ED99 and ED99ΔspsD demonstrated specific adherence to the human α-chain, but not to the β- or γ-chains revealing the α-chain as the receptor for SpsL binding (Fig 4A). To further refine the location of the SpsL binding site, 6 overlapping fragments of the canine fibrinogen α-chain were synthesized (NCBI reference sequence: XP_532697.2), purified from E. coli, and analyzed for adherence to ED99ΔspsLΔspsD expressing full length SpsL (Fig 4B and 4C). SpsL demonstrated binding to two of the overlapping fragments (250–450 and 400–600 amino acids) that span the α-connector region of fibrinogen containing unordered tandem repeats (residues P283-S419) (Fig 4B). Purification of equivalent fragments derived from the human α-chain revealed equivocal binding to the 400–600 fragment but reduced binding to the human 250–450 fragment (Fig 4D). These data indicate that the canine fibrinogen α-chain contains strong (250–450) and weaker (400–600) SpsL binding sites while human fibrinogen contains just the weaker binding site (400–600). To confirm that the canine tandem repeat region is responsible for the host-specific interaction of SpsL with fibrinogen, we generated chimeric full-length proteins where the tandem repeat region from human and canine fibrinogen α-chains, respectively (P283-G421), were exchanged. The addition of the canine α-chain tandem repeats provides stronger binding to the human fibrinogen α-chain and in contrast the addition of the human α-chain tandem repeats provides weaker binding to the canine fibrinogen α-chain (Fig 4E). As a control, we examined the binding of ClfB expressed on the surface of SH1000ΔclfAΔclfBΔfnbAΔfnbB. ClfB is a S. aureus fibrinogen-binding surface protein that binds specifically to repeat 5 of the human α-chain tandem repeats [13]. The host-specificity of ClfB was confirmed with specific binding observed to the human 250–450 fragment (S3A Fig), and the canine chimeric protein containing the human tandem repeat sequence (S3B Fig) but not to the canine 250–450 fragment or the canine alpha chain. These data demonstrate that the tandem repeat region of the fibrinogen α-chain is responsible for the host-specific interaction of SpsL.
The canine tandem repeat region of the fibrinogen α-chain contains 7 repeats of 18 amino acids and a partial repeat of 11 amino acids (S3C Fig) [38]. The generation of recombinant fragments spanning the tandem repeats of the fibrinogen α-chain (S3D Fig), revealed that SpsL is capable of binding to multiple regions in the canine tandem repeat region (S3E Fig). In addition, deletion of the whole tandem repeat region, in the canine α-chain, confirmed the presence of a weaker binding site (Fig 4F), which we localized to the region adjacent to the tandem repeats (S423-E474) in both canine and human fibrinogen (S3F and S3G Fig). Overall, these data demonstrate that SpsL mediates binding to multiple locations in the fibrinogen α-chain, and that the strong canine-specific interaction is dependent on the unique tandem repeat sequence present in the canine fibrinogen α-chain.
The number of tandem repeats in the bovine fibrinogen α-chain have been reported to differ between cattle breeds [39]. To investigate if this is also the case for dogs we investigated publicly available canine sequences of the fibrinogen α-chain but, due to the repetitive nature of the tandem repeat region, the paired-end short sequence reads were not sufficient to support assembly and robust analysis. To overcome this, we isolated genomic DNA from 11 different canine breeds and PCR-amplified DNA specific for the P283-E474 region of the fibrinogen α-chain, followed by DNA sequencing. Sequence analysis, in comparison to NCBI reference sequence: XP_532697.2, revealed that the region of weaker binding (S423-E474) is conserved among the canine breeds examined (S4A Fig). In contrast, the French bulldog and Labrador retriever exhibited heterozygous alleles that contain an additional repeat unit in the stronger binding site (Fig 4G). This heterozygous allele, common to both breeds, contains a duplication of repeat 4 and amino acid substitutions that result in the replacement of repeats 6 and 7 with repeat 8 (XP_532697.2:p.[S347_S348insTRPGSTGPGSAGTWS;S373N;L394P]) (Fig 4G). The unique French bulldog allele contains substitutions that convert repeat 5 to repeat 4, and repeat 7 to repeat 8 (XP_532697.2:p.[S347_T351del;G352R;T361A;L394P]), with a unique bulldog allele replacing repeat 6 with repeat 8 (XP_532697.2:p.S373N) (Fig 4G). Overall, these analyses demonstrate that the canine-specific binding site of SpsL in the tandem repeat region of the canine fibrinogen α-chain has undergone genetic diversification during the evolution of different breeds of dog.
The evolution of a strong canine-specific fibrinogen-binding interaction for SpsL suggests an important role in canine host-pathogen interactions. Accordingly, we investigated the impact of the interaction on phenotypes relevant to pathogenesis. Firstly, we considered if the interaction could promote inhibition of opsonophagocytosis as reported previously for fibrinogen-binding proteins of S. aureus ClfA and Efb [40, 41]. Accordingly, FITC-labelled bacteria were opsonized with bovine, canine, human, or ovine fibrinogen or bovine fibronectin and analyzed for phagocytosis by human neutrophils. As expected, full length SpsL, but not the chimeric A-domain protein (A+SD), inhibited phagocytosis in the presence of fibronectin (p<0.001) (Fig 5A). However, the ability of SpsL to inhibit neutrophil phagocytosis in the presence of fibrinogen was demonstrated to be host-specific with opsonophagocytosis inhibited in the presence of canine and human fibrinogen (p<0.001) but not in the presence of bovine or ovine fibrinogen (Fig 5A).
We next examined the role of SpsL-canine fibrinogen-binding on S. pseudintermedius aggregation and biofilm formation. The aggregation of S. aureus has been demonstrated to be important for the development of bloodstream infections [42, 43] and catheter-related infections [44]. In particular, it has been reported that fibrinogen-dependent S. aureus aggregation can stimulate the activation of virulence through a quorum-sensing dependent mechanism [45]. In order to examine the potential role of the canine-specific fibrinogen-binding in S. pseudintermedius aggregation, we attempted to block binding of S. pseudintermedius to the canine fibrinogen α-chain by including soluble fibrinogen in a bacterial adherence assay. Instead of blocking adherence, we found that soluble canine fibrinogen α-chain, but not human fibrinogen α-chain, supported the formation of surface bound aggregates (Fig 5B). Deletion of the weaker binding site (S423-E474) in the canine fibrinogen α-chain, had no effect on bacterial aggregation (Fig 5B). However, deletion of the stronger binding site (the tandem repeat region), resulted in complete abrogation of bacterial aggregation (Fig 5B). This demonstrates that SpsL promotes surface bound bacterial aggregation in a host-restricted manner. To further investigate the impact of fibrinogen on the aggregation of S. pseudintermedius we performed static biofilm assays in the presence or absence of fibrinogen from different host species (Fig 5C). Coating with canine fibrinogen supported enhanced biofilm formation among bacterial cells expressing SpsL than wells coated with either human or bovine fibrinogen demonstrating that the strong interaction of SpsL with canine fibrinogen promotes the initial attachment stage of biofilm formation (Fig 5C). Overall, these data demonstrate that the high strength interaction of SpsL with canine fibrinogen promotes bacterial aggregation and biofilm formation.
In order to investigate if other staphylococcal fibrinogen-binding proteins exhibit similar host-specificity, we generated constructs expressing chimeric SpsL proteins that contain the fibrinogen-binding N2N3 subdomains of ClfB or FnBPA but maintain the SpsL promoter, signal peptide, fibronectin-binding repeats, and cell wall anchor (Fig 6A). The generation of chimeric proteins was favored over the expression of native ClfB or FnBPA proteins to limit variation in cell surface expression. The N2N3 subdomains of these proteins were selected because of the host-restriction of ClfB to human fibrinogen α-chain and the similar domain architecture of SpsL and FnBPA. As expected from our previous analysis, SpsL showed similar binding to both canine and human fibrinogen (Fig 6B). However, the high strength interaction of SpsL with canine fibrinogen is essential for bacterial aggregation (Fig 6C) and biofilm formation (Fig 6H). SpsL-ClfBN2N3 demonstrated a similar binding curve to SpsL but exhibited specific binding to human fibrinogen (Fig 6D) as previously predicted [13]. This human fibrinogen-binding was not sufficient to mediate bacterial aggregation (Fig 6E) or biofilm formation (Fig 6H) demonstrating that not all staphylococcal fibrinogen-binding proteins are capable of mediating these infection-related phenotypes. In contrast, SpsL-FnBPAN2N3 exhibited a binding pattern predicted from an interaction with the fibrinogen γ-chain with equivocal binding to bovine, canine, and human fibrinogen and reduced ovine fibrinogen-binding (Fig 6F). SpsL-FnBPAN2N3 is capable of mediating bacterial aggregation (Fig 6G) and biofilm formation (Fig 6H) in the presence of fibrinogen from all hosts tested suggesting that FnBPA does not have a host-restricted tropism. From this comparative analysis we can conclude that SpsL is unique in promoting bacterial aggregation and biofilm formation in a manner that corresponds to the host-restricted ecology of S. pseudintermedius.
The factors underpinning bacterial host-tropism are not well understood but often involve surface proteins mediating interactions with host cells and the extracellular matrix [1]. The genus Staphylococcus includes species such as S. aureus which have a multi-host tropism with the capacity to switch between different host species. In contrast, some species such as S. pseudintermedius are highly host-restricted and although S. pseudintermedius can occasionally cause zoonotic infections of humans (typically through dog bite wounds), the capacity to spread in human populations has not been reported. The bacterial factors underpinning the host-restricted ecology of S. pseudintermedius are unknown. Previously, we demonstrated that SpsL contributed to abscess formation in a murine model of subcutaneous infection indicating that it is a virulence factor during the pathogenesis of skin infection [22]. The poor binding of SpsL to murine fibrinogen suggests that this effect is not mediated by the interaction of SpsL with murine fibrinogen [22]. Here we demonstrate that SpsL mediates high strength binding to canine fibrinogen in a host-specific manner and that this host-adaptation conferred the ability to mediate bacterial aggregation and biofilm formation. The role of SpsL-fibrinogen binding in canine pathogenesis cannot be formally tested in vivo by experimental infections of dogs in the UK due to ethical constraints. However, our ex vivo binding and cellular infection data reveal multiple pathogenic traits that depend on the host-specific interaction of SpsL and canine fibrinogen, suggesting a key role in the host ecology of S. pseudintermedius.
Cell surface proteins of S. aureus have been reported to contribute to tissue or disease tropism in humans. For example, the fibrinogen- and loricrin-binding protein, ClfB, exhibits greater adherence to skin corneocytes taken from atopic dermatitis patients with low levels of natural moisturizing factor suggesting a role in niche adaptation [46, 47]. ClfB interacts with the human tandem repeat region of the fibrinogen α-chain [13] but unlike SpsL, ClfB binds to a single site; namely repeat unit 5, and exclusively binds to human fibrinogen [13]. In addition to ClfB, the bone sialoprotein-binding protein (Bbp) and the extracellular fibrinogen-binding protein (Efb) also bind to the fibrinogen α-chain via distinct RGD-integrin-binding sites inhibiting thrombin-induced coagulation and platelet aggregation, respectively [36, 48]. In contrast, SpsL interacts with multiple sites in the canine fibrinogen α-chain; namely within the tandem repeats and their flanking regions, (Fig 4). Similarly, the serine-rich repeat glycoproteins of Streptococcus agalactiae, Srr1 and Srr2, bind to repeat units 6, 7, and 8 of the tandem repeat region of the human fibrinogen α-chain via a variation of the dock, lock, and latch binding mechanism, with Srr2 displaying a stronger binding affinity than Srr1 [49]. The enhanced binding affinity of Srr2 was linked with increased adherence to endothelial cells, which may be important for Group B Streptococcus-associated meningitis [49]. The ability of the Srr and SpsL proteins to adhere to more than one site in the tandem repeat region of the fibrinogen α-chain may have evolved as a mechanism for overcoming extant genetic diversity in this region between individuals within a host species as observed in the current study for SpsL (Fig 4G) [39, 50].
We were unable to detect binding of soluble SpsL proteins to canine fibrinogen by ELISA suggesting that immobilization and surface presentation is essential for SpsL functionality, even when full length SpsL is expressed as a recombinant protein (S1 Fig). To address this, we utilized AFM, demonstrating that bacterial surface-associated SpsL binds to fibrinogen via extremely strong binding forces (around 2000 pN) that are in the range of the strength measured for the dock, lock and latch interaction between fibrinogen and the structurally-related SdrG and ClfA [26, 51]. Dock, lock and latch forces have been shown to originate from hydrogen bonds between the ligand peptide backbone and the adhesin [52, 53], and are activated by mechanical tension, as observed with catch bonds [54]. Of note, ClfB has much greater affinity for loricrin when expressed on the bacterial cell surface rather than as a recombinant protein with the C-terminal stalk enhancing binding affinity [25]. A similar mechanism may be required for SpsL adherence to fibrinogen with the C-terminal repeat domain enhancing the ligand-binding affinity of the N2N3 subdomains. It is increasingly being recognized that analysis of protein-protein interactions on the bacterial cell surface is more physiologically relevant than testing the interaction of recombinant polypeptides [25, 55].
Our data reveal that the high strength canine-specific binding of SpsL facilitates several virulence phenotypes not previously reported for S. pseudintermedius including surface-bound bacterial aggregation. When S. aureus forms fibrinogen-dependent aggregates, agr-mediated quorum sensing is activated leading to the up-regulation of virulence gene expression [45]. Consequently, the inhibition of S. aureus aggregation in vivo has been linked with decreases in mortality from sepsis and protection from lethal lung injury [43, 56]. We also discovered that SpsL facilitates fibrinogen-dependent biofilm formation, a phenotype not previously reported for S. pseudintermedius. Such fibrinogen-dependent biofilms are observed in S. aureus strains isolated from skin infections [57], a phenomenon implicated in indwelling medical device infections [58]. In this regard, inhibition of fibrin formation reduced the development of S. aureus biofilms in a murine catheter infection model [58], and molecules targeting SpsL could be beneficial in preventing canine indwelling device infections caused by S. pseudintermedius. Finally, we have demonstrated that SpsL binding to soluble fibrinogen inhibits neutrophil phagocytosis, suggesting a role for SpsL in innate immune evasion. Taken together, we have dissected the host-dependent binding of a bacterial surface protein and demonstrated its importance for multiple pathogenic traits, providing new insights into the host-specific ecology of a major bacterial pathogen.
Chicken immunization was performed using unembryonated hen’s eggs at the Scottish national blood transfusion service (Pentland Science Park, Midlothian, UK). The procedures performed were carried out under the authority of the UK Home Office Project License PPL 60/4165 and Animals (Scientific Procedures) Act 1986 regulations.
Human venous blood was taken from consenting adult healthy volunteers in accordance with a human subject protocol approved by the national research ethics service (NRES) committee London City and East under the research ethics committee reference 13/LO/1537. Passive volunteer recruitment was conducted at the Roslin Institute (University of Edinburgh). Written consent was taken from each volunteer before blood collection and after an outline of the risks was provided. All blood collection samples were anonymized.
The bacterial strains and plasmids used in this study are listed in S1 Table. S. pseudintermedius and S. aureus strains were routinely cultured in Brain Heart Infusion broth at 37°C with shaking and supplemented with 10 μg ml-1 chloramphenicol as required. E. coli strains were cultured in Luria broth at 37°C with shaking supplemented with 100 μg ml-1 ampicillin, 15 μg ml-1 tetracycline, or 25 μg ml-1 kanamycin as required.
Fibrinogen isolated from bovine, human, and ovine plasma (Sigma-Aldrich) and bovine fibronectin (EMD Millipore) was sourced commercially. Canine fibrinogen was purified from Beagle sodium citrate whole blood (Lampire Biological Products) using a previously described method [59]. All fibrinogen samples were purified to remove contaminating fibronectin using Gelatin-Sepharose 4B (GE Healthcare). Depletion of fibronectin was confirmed by Western blot analysis using 1 μg ml-1 rabbit anti-fibronectin IgG (abcam) and 0.2 μg ml-1 goat anti-rabbit IgG-HRP (abcam).
Solid phase adherence assays were performed using S. pseudintermedius and S. aureus strains expressing pALC2073 or pCU1 constructs cultured to an OD600nm of 0.6 and induced for protein expression with 3 μg ml-1 anhydrotetracycline for 2 h. Cells were washed and suspended in PBS to OD600nm of 1.0. Wells were coated overnight at 4°C with fibrinogen from multiple hosts or recombinant α-chain fragments in a 96-well MaxiSorp plate (Nunc). After blocking with 8% (w/v) milk-PBS, bacteria were applied to the wells for 2 h at 37°C. After washing, adherent cells were fixed with 25% (v/v) formaldehyde (Sigma) for 30 min and stained with 0.5% (v/v) crystal violet (Sigma) for 3 min. The cell-associated stain was solubilized with 5% acetic acid (v/v) and analyzed using a Synergy HT plate reader (BioTek) at 590 nm wavelength.
For aggregation experiments, the same procedure was followed as stated above with the addition of either soluble fibrinogen or recombinant fibrinogen α-chain to the bacteria using two-fold serial dilution and then incubation for 2 h at 37°C.
The primers used in this study are listed in S2 Table. Initial expression constructs of the human and canine fibrinogen α-chains were synthesized by Integrated Design Technologies (IDT) using the DNA sequence of a female Boxer (NCBI reference sequence: XP_532697.2) as highlighted in S1 Table. For typical restriction-ligation cloning procedures, the region of interest was amplified (PfuUltra II Fusion HS Polymerase—Agilent) and blunt cloned into pSC-B using the StrataClone Blunt PCR Cloning Kit (Agilent). Restriction digestion of the plasmid of interest (pQE30, pT7, or pALC2073) and the blunt cloned PCR product was performed at 37°C for at least 2 h and purified using the Monarch Gel Extraction Kit (NEB). All digested plasmids were treated with Antarctic Phosphatase (NEB) before overnight ligation with T4 DNA Ligase (NEB) at a 3:1 molar ratio of insert:plasmid. Dialysis of the 20 μl ligation reactions was performed using 0.025 μm filter circular discs (Millipore) before electroporation into the appropriate E. coli strain–DC10B [60], DH5α (Invitrogen), or XL-1 Blue (Agilent). All plasmid constructs were verified by Sanger sequencing (Edinburgh Genomics, University of Edinburgh) before transformation into E. coli BL21 DE3 (Invitrogen), or appropriate S. pseudintermedius strain.
Some expression constructs were also produced using sequence ligase independent cloning (SLIC) as described previously [61]. Briefly, primers were designed to amplify the gene of interest as well as sequence complementation to the expression plasmid. Primers were also designed to amplify the plasmid of interest, pQE30, pALC2073 or pCT using Platinum PCR Supermix (Invitrogen) or PfuUltra II Fusion HS Polymerase (Agilent). All PCR products were purified using Monarch PCR & DNA Cleanup kit or Monarch Gel Extraction kit (NEB). T4 DNA Polymerase (NEB) was used to generate DNA overhangs on both the insert and plasmid PCRs with step-wise temperature increments used to anneal the complementary DNA sequences. The heat annealed constructs were electro-transformed into E. coli DC10B [60] or DH5α (Invitrogen) and verified using Sanger sequencing (Edinburgh Genomics, University of Edinburgh).
S. pseudintermedius and S. aureus competent cells were produced using a method outlined previously [60]. Plasmids for electroporation were concentrated to 1 μg μl-1 using Pellet Paint co-precipitant (Novagen) and 5 μg used for the electro-transformation as previously described [23].
Recombinant hexa-Histidine-tagged proteins expressed in E. coli were cultured to OD600nm of 0.6 and induced using 1 mM IPTG at either 37°C for 4 h or 16°C overnight. Recombinant α-chain proteins were purified under denaturing conditions (8M urea, 100 mM monosodium phosphate, 10 mM Tris-HCl) using Ni-NTA agarose (Invitrogen) and gravity flow columns (Bio-Rad). Bacterial lysis was performed in pH 8.0 binding buffer at room temperature with tilting for at least 1 h. Lysates were pelleted at 16000 x g for 20 min and the supernatant filter sterilized. Lysates were tilted at room temperature with conditioned Ni-NTA agarose for 1 h. The column was washed with pH 6.3 wash buffer and the protein eluted with pH 4.5 elution buffer. After analysis by 4–20% Mini-PROTEAN TGX precast gel (Bio-Rad), protein quantification was performed using a BCA assay (Novagen).
S. pseudintermedius cells were cultured to exponential phase (OD600nm of 0.4–0.6). Cells were washed with PBS and suspended in lysis buffer (50 mM Tris-HCl, 20 mM MgCl2, pH 7.5) supplemented with 30% (w/v) raffinose and cOmplete protease inhibitor (Roche). Cell wall proteins were solubilized by incubation with 400 μg ml-1 lysostaphin at 37°C for 20 min. Supernatant samples were collected after protoplast recovery by centrifugation at 6000 x g for 20 min. The production of cell lysate samples was generated by lysing cell pellets in PBS on the One-Shot cell disruptor (Constant Systems) with 2 passes at 40 Kpsi.
Recombinant His-tag SpsL N2N3 protein was used as antigen for chicken immunization and antibody generation at the Scottish national blood transfusion service (Pentland Science Park). The Eggspress IgY purification kit (Gallus Immunotech) was used to purify antibody from egg yolk. Further purification of the antibody was performed using CNBr-activated Sepharose 4B (GE Healthcare). This antibody was used in Western blot analysis to detect the expression of SpsL using 1 μg ml-1 chicken anti-SpsL N2N3 IgY and 0.5 μg ml-1 F(ab’)2 rabbit anti-chicken IgG-HRP (Bethyl Laboratories).
Genomic DNA was isolated from whole canine blood using the method described previously [65]. The region of interest in the fibrinogen α-chain was amplified using Q5 Hot Start high-fidelity DNA polymerase (NEB) and purified using Monarch PCR & DNA Cleanup kit (NEB). Purified PCR products were analyzed by Sanger sequencing (Eurofins) and DNAStar SeqMan Pro 14 (Lasergene). Sequence alignment was performed using MegAlign (Lasergene) and PRALINE [66].
Biofilm assays were performed using S. pseudintermedius strains expressing pALC2073 constructs of full length SpsL or A-domain+SD. Strains were grown in TSB supplemented with 0.5% (v/v) glucose and 3% (v/v) NaCl. 96-well tissue culture plates were coated overnight at 4°C with 100 nM bovine, canine, human, or ovine fibrinogen with some wells left uncoated. Overnight cultures were diluted to an OD600nm of 0.05 and 100 μl applied to the plate and incubated at 37°C for 24 h. The plates were washed three times with PBS and the bacteria fixed with 25% (v/v) formaldehyde (Sigma) for 30 min. After washing, the plates were stained with 0.5% (v/v) crystal violet (Sigma) for 3 min and then solubilized with 5% acetic acid (v/v). Plates were analyzed using a Synergy HT plate reader (BioTek) at 595 nm wavelength.
50 ml of venous blood was drawn from healthy volunteers and mixed with 6 ml of acid-citrate-dextran (Sigma). Human neutrophils were isolated as outlined previously [67] and suspended to a final concentration of 2.5 x 106 cells ml-1 in RPMI-1640 (Gibco) containing 0.05% human serum albumin (Sigma). 2.5 x 106 CFU of bacteria, previously labelled with FITC using a method previously described [68], were opsonized with 50 nM of extracellular matrix protein at 37°C for 15 min and diluted to 1 ml in RPMI-1640 containing 0.05% human serum albumin. 2.5 x 105 CFU were then opsonized with 10% human serum in 2 ml 96-well v-bottomed plates (Corning) at 37°C for 15 min. 2.5 x 105 neutrophils were added to the opsonized bacteria (MOI of 1) and incubated at 37°C for 15 min with shaking at 750 rpm. The samples were fixed with 1% (v/v) paraformaldehyde (Fisher Scientific) and incubated at 4°C for at least 30 min. Phagocytosis was measured in comparison to serum-only controls using the BD LSRFortessa X20 cell analyzer.
Data is presented in Prism 6 (Graphpad) with statistical analysis performed using Minitab 16. All data was analyzed for normality, using the Anderson-Darling test, and equal variance before choosing the method of statistical analysis. Multiple comparisons were performed were appropriate. ELISA-type binding assays and bacterial adherence assays were analyzed at one protein concentration. For data displaying statistical significance, the following symbols are used, * p≤0.05, ** p≤0.01, and *** p≤0.001.
|
10.1371/journal.pcbi.1006166 | Variability in pulmonary vein electrophysiology and fibrosis determines arrhythmia susceptibility and dynamics | Success rates for catheter ablation of persistent atrial fibrillation patients are currently low; however, there is a subset of patients for whom electrical isolation of the pulmonary veins alone is a successful treatment strategy. It is difficult to identify these patients because there are a multitude of factors affecting arrhythmia susceptibility and maintenance, and the individual contributions of these factors are difficult to determine clinically. We hypothesised that the combination of pulmonary vein (PV) electrophysiology and atrial body fibrosis determine driver location and effectiveness of pulmonary vein isolation (PVI). We used bilayer biatrial computer models based on patient geometries to investigate the effects of PV properties and atrial fibrosis on arrhythmia inducibility, maintenance mechanisms, and the outcome of PVI. Short PV action potential duration (APD) increased arrhythmia susceptibility, while longer PV APD was found to be protective. Arrhythmia inducibility increased with slower conduction velocity (CV) at the LA/PV junction, but not for cases with homogeneous CV changes or slower CV at the distal PV. Phase singularity (PS) density in the PV region for cases with PV fibrosis was increased. Arrhythmia dynamics depend on both PV properties and fibrosis distribution, varying from meandering rotors to PV reentry (in cases with baseline or long APD), to stable rotors at regions of high fibrosis density. Measurement of fibrosis and PV properties may indicate patient specific susceptibility to AF initiation and maintenance. PV PS density before PVI was higher for cases in which AF terminated or converted to a macroreentry; thus, high PV PS density may indicate likelihood of PVI success.
| Atrial fibrillation is the most commonly encountered cardiac arrhythmia, affecting a significant portion of the population. Currently, ablation is the most effective treatment but success rates are less than optimal, being 70% one-year post-treatment. There is a large effort to find better ablation strategies to permanently cure the condition. Pulmonary vein isolation by ablation is more or less the standard of care, but many questions remain since pulmonary vein ectopy by itself does not explain all of the clinical successes or failures. We used computer simulations to investigate how electrophysiological properties of the pulmonary veins can affect rotor formation and maintenance in patients suffering from atrial fibrillation. We used complex, biophysical representations of cellular electrophysiology in highly detailed geometries constructed from patient scans. We heterogeneously varied electrophysiological and structural properties to see their effects on rotor initiation and maintenance. Our study suggests a metric for indicating the likelihood of success of pulmonary vein isolation. Thus either measuring this clinically, or running patient-specific simulations to estimate this metric may suggest whether ablation in addition to pulmonary vein isolation should be performed. Our study provides motivation for a retrospective clinical study or experimental study into this metric.
| Success rates for catheter ablation of persistent atrial fibrillation (AF) patients are currently low; however, there is a subset of patients for whom pulmonary vein isolation (PVI) alone is a successful treatment strategy [1]. PVI ablation may work by preventing triggered beats from entering the left atrial body, or by converting rotors or functional reentry around the left atrial/pulmonary vein (LA/PV) junction to anatomical reentry around a larger circuit, potentially converting AF to a simpler tachycardia [2]. It is difficult to predict whether PVI represents a sufficient treatment strategy for a given patient with persistent AF [1], and it is unclear what to do for the majority of patients for whom it is not effective.
Patients with AF exhibit distinct properties in effective refractory period (ERP) and conduction velocity (CV) in the PVs. For example, paroxysmal AF patients have shorter ERP and longer conduction delays compared to control patients [3]. AF patients show a number of other differences to control patients: PVs are larger [4]; PV fibrosis is increased; and fiber direction may be more disorganised, particularly at the PV ostium [5]. There are also differences within patient groups; for example, patients for whom persistent AF is likely to terminate after PVI have a larger ERP gradient compared to those who require further ablation [1, 3].
Electrical driver location changes as AF progresses; drivers (rotors or focal sources) are typically located close to the PVs in early AF, but are also located elsewhere in the atria with longer AF duration [6]. Atrial fibrosis is a major factor associated with AF and modifies conduction. However, there is conflicting evidence on the relationship between fibrosis distribution and driver location [7, 8].
It is difficult to clinically separate the individual effects of these factors on arrhythmia susceptibility and maintenance. We hypothesise that the combination of PV properties and atrial body fibrosis determines driver location and, thus, the likely effectiveness of PVI. In this study, we tested this hypothesis by using computational modelling to gain mechanistic insight into the individual contribution of PV ERP, CV, fiber direction, fibrosis and anatomy on arrhythmia susceptibility and dynamics. We incorporated data on APD (action potential duration, as a surrogate for ERP) and CV for the PVs to determine mechanisms underlying arrhythmia susceptibility, by testing inducibility from PV ectopic beats. We also predicted driver location, and PVI outcome.
All simulations were performed using the CARPentry simulator (available at https://carp.medunigraz.at/carputils/). We used a previously published bi-atrial bilayer model [9], which consists of resistively coupled endocardial and epicardial surfaces. This model incorporates detailed atrial structure and includes transmural heterogeneity at a similar computational cost to surface models. We chose to use a bilayer model rather than a volumetric model incorporating thickness for this study because of the large numbers of parameters investigated, which was feasible with the reduced computational cost of the bilayer model.
As previously described, the bilayer model was constructed from computed tomography scans of a patient with paroxysmal AF, which were segmented and meshed to create a finite element mesh suitable for electrophysiology simulations. Fiber information was included in the model using a semi-automatic rule based method that matches histological descriptions of atrial fiber orientation [10]. The left atrium of the bilayer model consists of linearly coupled endocardial and epicardial layers, while the right atrium is an epicardial layer, with endocardial atrial structures including the pectinate muscles and crista terminalis. The left and right atrium of the model are electrically connected through three pathways: Bachmann’s bundle, the coronary sinus and the fossa ovalis. Tissue conductivities were tuned to human activation mapping data from Lemery et al. [9, 11].
The Courtemanche-Ramirez-Nattel human atrial ionic model was used with changes representing electrical remodelling during persistent AF [12], together with a doubling of sodium conductance to produce realistic action potential upstroke velocities [9], and a decrease in IK1 by 20% to match clinical restitution data [13]. Regional heterogeneity in repolarisation was included by modifying ionic conductances of the cellular model, as described in Bayer et al. [14], which follows Aslanidi et al. and Seemann et al. [15, 16]. Parameters for the baseline PV model were taken from Krueger et al. [17].
The following PV properties were varied as shown in schematic Fig 1: APD, CV, fiber direction, the inclusion of fibrosis in the PVs and the atrial geometry. These are described in the following sections.
To investigate the effects of PV length and diameter on arrhythmia inducibility and arrhythmia dynamics, bi-atrial bilayer meshes were constructed from MRI data for twelve patients. All patients gave written informed consent; this study is in accordance with the Declaration of Helsinki, and approved by the Institutional Ethics Committee at the University of Bordeaux. Patient-specific models with electrophysiological heterogeneity and fiber direction were constructed using our modelling pipeline, which uses a universal atrial coordinate system to map scalar and vector data from the original bilayer model to a new patient specific mesh. Late gadolinium enhancement MRI (average resolution 0.625mm x 0.625mm x 2.5mm) was performed using a 1.5T system (Avanto, Siemens Medical Solutions, Erlangen, Germany). These LGE-MRI data were manually segmented using the software MUSIC (Electrophysiology and Heart Modeling Institute, University of Bordeaux, Bordeaux France, and Inria, Sophia Antipolis, France, http://med.inria.fr). The resulting endocardial surfaces were meshed (using the Medical Imaging Registration Toolkit mcubes algorithm [18]) and cut to create open surfaces at the mitral valve, the four pulmonary veins, the tricuspid valve, and each of the superior vena cava, the inferior vena cava and the coronary sinus using ParaView software (Kitware, Clifton Park, NY, USA). The meshes were then remeshed using mmgtools meshing software (http://www.mmgtools.org/), with parameters chosen to produce meshes with an average edge length of 0.34mm to match the resolution of the previously published bilayer model [9]. Two atrial coordinates were defined for each of the LA and RA, which allow automatic transfer of atrial structures to the model, such as the pectinate muscles and Bachmann’s bundle. These coordinates were also used to map fiber directions to the bilayer model.
To investigate the effects of PV electrophysiology on arrhythmia inducibility and dynamics, we varied PV APD and CV by modifying the value of the inward rectifier current (IK1) conductance and tissue level conductivity respectively. IK1 conductance was chosen in this case to investigate macroscopic differences in APD [19], although several ionic conductances are known to change with AF [20]. Modifications were either applied homogeneously or following a ostial-distal gradient. This gradient was implemented by calculating geodesic distances from the rim of mesh nodes at the distal PV boundary to all nodes in the PV and from the rim of nodes at the LA/PV junction to all nodes in the PV. The ratio of these two distances was then used as a distance parameter from the LA/PV junction to the distal end of the PV (see Fig 1).
IK1 conductance was multiplied by a value in the range 0.5–2.5, resulting in PV APDs in the clinical range of 100–190ms [3, 21, 22]. This rescaling was either a homogeneous change or followed a gradient along the PV length. Gradients of IK1 conductance varied from the baseline value at the LA/PV junction, to a maximum scaling factor at the distal boundary. PV APDs are reported at 90% repolarisation for a pacing cycle length of 1000ms. LA APD is 185ms, measured at a LA pacing cycle length of 200ms.
To cover the clinically observed range of PV CVs, longitudinal and transverse tissue conductivities were divided by 1, 2, 3 or 5, resulting in CVs, measured along the PV axis, in the range: 0.28–0.67m/s [3, 21–24]. To model heterogeneous conduction slowing, conductivities were varied as a function of distance from the LA/PV junction, ranging from baseline at the junction to a maximum rescaling (minimum conductivity) at the distal boundary. The direction of this gradient was also reversed to model conduction slowing at the LA/PV junction [5].
Motivated by the findings of Hocini et al. [5], interstitial fibrosis was modelled for the PVs with a density varying along the vein, increasing from the LA/PV junction to the distal boundary. This was implemented by randomly selecting edges of elements of the mesh with probability scaled by the distance parameter and the angle of the edge compared to the element fiber direction, where edges in the longitudinal fiber direction were four times more likely to be selected than those in the transverse direction, following our previous methodology [25]. To model microstructural discontinuities, no flux boundary conditions were applied along the connected edge networks, following Costa et al. [26]. An example of modelled PV interstitial fibrosis is shown in S1A Fig.
For a subset of simulations, interstitial fibrosis was incorporated in the biatrial model based on late gadolinium enhancement (LGE)-MRI data, using our previously published methodology [25]. In brief, likelihood of interstitial fibrosis depended on both LGE intensity and the angle of the edge compared to the element fiber direction (see S1B Fig). LGE intensity distributions were either averaged over a population of patients [27], or for an individual patient. The averaged distributions were for patients with paroxysmal AF (averaged over 34 patients), or persistent AF (averaged over 26 patients). For patient-specific simulations, the model arrhythmia dynamics were compared to AF recordings from a commercially available non-invasive ECGi mapping technology (CardioInsight Technologies Inc., Cleveland, OH) for which phase mapping analysis was performed as previously described [28].
PV fiber direction shows significant inter-patient variability. Endocardial and epicardial fiber direction in the four PVs was modified according to fiber arrangements described in the literature [5, 29, 30]. Six arrangements were considered, as follows: 1. circular arrangement on both the endocardium and epicardium; 2. spiralling arrangement on both the endocardium and epicardium; 3. circular arrangement on the endocardium, with longitudinal epicardial fibers; 4. fibers progress from longitudinal at the distal vein to circumferential at the ostium, with identical endocardial and epicardial fibers; 5. epicardial layer fibers as per case 4, with circumferential endocardial fibers; 6. as per case 4, but with a chaotic fiber arrangement at the LA/PV junction. These fiber distributions are shown in S2 Fig.
Cases 4–6 were implemented by setting the fiber angle to be a function of the distance along the vein, measured from the LA/PV junction to the distal boundary, varying from circumferential at the junction to longitudinal at the distal end (representing a change of 90 degrees). The disorder in fiber direction at the LA/PV junction for case 6 was implemented by taking the fibers of case 4 and adding independent standard Gaussian distributions scaled by the distance from the distal boundary, resulting in the largest perturbations at the ostium.
Arrhythmia inducibility was tested by extrastimulus pacing from each of the four PVs individually using a clinically motivated protocol [31], to simulate the occurrence of PV ectopics. Simulations were performed for each of the PVs, to determine the effects of ectopic beat location on inducibility. Sinus rhythm was simulated by stimulating the sinoatrial node region of the model at a cycle length of 700ms throughout the simulation. Each PV was paced individually with five beats at a cycle length of 160ms, and coupling intervals between the first PV beat and a sinus rhythm beat in the range 200–500 ms. Thirty-two pacing protocols were applied for each model set up: eight coupling intervals (coupling interval = 200, 240, 280, 320, 360, 400, 440, 480ms), for each of the four PVs. Inducibility is reported as the proportion of cases resulting in reentry; termed the inducibility ratio.
The effects of PVI were determined for model set-ups that used the original bilayer geometry and in which the arrhythmia lasted for greater than two seconds. PVI was applied two seconds post AF initiation in each case by setting the tissue conductivity close to zero (0.001 S/m) in the regions shown in S3 Fig.
For each case, ten seconds of arrhythmia data were analysed, starting from two seconds post AF initiation, to identify re-entrant waves and wavefront break-up using phase. The phase of the transmembrane voltage was calculated for each node of the mesh using the Hilbert transform, following subtraction of the mean [32]. Phase singularities (PSs) for the transmembrane potential data were identified by calculating the topological charge of each element in the mesh [33], and PS spatial density maps were calculated using previously published methods [14]. PS density maps were then partitioned into the LA body, PV regions, and the RA to assess where drivers were located in relation to the PVs (see S3 Fig). The PV region was defined as the areas enclosed by, and including, the PVI lines; the LA region was then the rest of the LA and left atrial appendage. The PV PS density ratio was then defined as the total PV PS count divided by the total model PS count over both atria.
A difference in APD between the model LA and PVs was required for AF induction. Modelling the PVs using LA cellular properties resulted in non-inducibility, whereas, modelling the LA using PV cellular properties resulted in either non-inducibility or macroreentry.
The effects of modifying PV APD homogeneously or following a gradient are shown in Table 1. Simulations in which PV APD was longer than LA APD were non-inducible (PV APD 191ms). As APD was decreased below the baseline value (181ms), inducibility initially increased and then fluctuated. Comparing cases with equal distal APD, arrhythmia inducibility was significantly higher for APD following a ostial-distal gradient than for homogeneous APD (p = 0.03 from McNemar’s test).
PS location was also affected by PV APD. PV PS density was low in cases of short APD, an example of which is shown in Fig 2 where reentry is no longer seen around the LA/PV junction in the case of short APD (120ms). This change was more noticeable for cases with homogeneous PV APD than for a gradient in APD; PV reentry was observed for the baseline case and a heterogeneous APD case, but not for a homogeneous decrease in APD.
Arrhythmia inducibility decreased with homogeneous CV slowing (from 0.38 i.e. 12/32 at 0.67m/s to 0.03 i.e. 1/32 at 0.28m/s). In the baseline model, reentry occurs close to the LA/PV junction due to conduction block when the paced PV beat encounters a change in fiber direction at the base of the PVs, together with a longer LA APD compared to the PV APD. In this case, the wavefront encounters a region of refractory tissue due to the longer APD in the LA. However, when PV CV is slowed homogeneously, the wavefront takes longer to reach the LA tissue, giving the tissue enough time to recover, such that conduction block and reentry no longer occurs. Modifying conductivity following a gradient means that, unlike the homogeneous case, the time taken for the extrastimulus wavefront to reach the LA tissue is similar to the baseline case, so the LA tissue might still be refractory and conduction block might occur. In the case that conduction was slowest at the distal vein, the inducibility was similar to the baseline case (see Table 2, GA, inducibility is 0.38 at baseline and 0.34 for the cases with CV slowing). Cases with greatest conduction slowing at the LA/PV junction (see Table 2, GB) exhibit an increase in inducibility (from 0.38 to 0.53) when CV is decreased because of the discontinuity in conductivity at the junction.
Fig 2 shows that reentry is seen around the LA/PV junction in cases with both baseline and slow CV, indicating that the presence of reentry at the LA/PV junction is independent of PV CV.
PV conduction properties are also affected by PV fiber direction. Modifications in fiber direction increased inducibility compared to the baseline fiber direction (baseline case: 0.38; modified fiber direction cases 1-6: 0.53-0.63). The highest inducibility occurred with circular fibers at the ostium (cases 1 and 4, 0.63), independent of fiber direction at the distal PV end. This inducibility was reduced if the epicardial fibers were not circular at the ostium (case 3, 0.56), or if fibers were spiralling (case 2, 0.56) instead of circular.
Next we investigated the interplay between PV properties and atrial fibrosis. LA fibrosis properties were varied to represent interstitial fibrosis in paroxysmal and persistent AF patients, incorporating average LGE-MRI distributions [27] into the model. These control, paroxysmal and persistent AF levels of fibrosis were then combined with PV properties varied as follows: baseline CV and APD (0.67m/s, 181ms), slow CV (0.51m/s), short APD (120ms), slow CV and short APD. PS distributions in Fig 2 show that reentry occurred around the LA/PV junction in the case of baseline PV APD for control or paroxysmal levels of fibrosis, but not for shorter PV APD. Modifying PV CV did not affect whether LA/PV reentry is observed. Rotors were found to stabilise to regions of high fibrosis density in the persistent AF case.
Models with PV fibrosis had a higher inducibility compared to the baseline case (0.47 vs. 0.38) and a higher PV PS density since reentry localised there. Fig 3 shows an example with moderate PV fibrosis (A) in which reentry changed from around the RIPV to the LIPV later in the simulation; adding a higher level of PV fibrosis resulted in a more stable reentry around the right PVs (B).
The relationship between LA fibrosis and PV properties on driver location was investigated on an individual patient basis for four patients. For patients for whom rotors were located away from the PVs (Fig 4 LA1), increasing model fibrosis from low to high increased the model agreement with clinical PS density 2.3 ± 1.0 fold (comparing the sensitivity of identifying clinical regions of high PS density using model PS density between the two simulations). For other patients, lower levels of fibrosis were more appropriate (2.1 fold increase in agreement for lower fibrosis, Fig 4 LA2), and PV isolation converted fibrillation to macroreentry in the model.
Arrhythmia inducibility showed a large variation between patient geometries (0.16–0.47). Increasing PV area increased inducibility to a different degree for each vein: right superior PV (RSPV) inducibility was generally high (> 0.75 for all but one geometry) independent of PV area; left superior PV (LSPV) inducibility increased with PV area (Spearman’s rank correlation coefficient of 0.36 indicating positive correlation; line of best fit gradient 0.27, R2 = 0.3); left inferior PV (LIPV) and right inferior PV (RIPV) inducibility exhibited a threshold effect, in which veins were only inducible above a threshold area (Fig 5A). There is no clear relationship between PV length and inducibility. PV PS density ratio increased as PV area increased (Fig 5B, Spearman’s rank correlation coefficient of 0.41 indicating positive correlation). Fig 5C shows that rotor and wavefront trajectories depend on patient geometry, exhibiting varied importance of the PVs compared to other atrial regions.
PVI outcome was assessed for cases with varied PV APD (both with a homogeneous change or following a gradient), with the inclusion of PV fibrosis and with varied PV fiber direction because these factors were found to affect the PV PS density ratio. PVI outcome was classified into three classes depending on the activity 1 second after PVI was applied in the model: termination, meaning there was no activity; macroreentry, meaning that there was a macroreentry around the LA/PV junctions; AF sustained by LA rotors, meaning there were drivers in the LA body. These classes accounted for different proportions of the outcomes: termination (27.3% of cases), macroreentry (39.4%), or AF sustained by LA rotors (33.3%). Calculating the PV PS density ratio before PVI for each of these classes shows that cases in which the arrhythmia either terminated or changed to a macroreentry are characterised by a statistically higher PV PS density ratio pre-PVI than cases sustained by LA rotors post-PVI (see Fig 6, t-test comparing termination and LA rotors shows they are significantly different, p<0.001; comparing macroreentry and LA rotors p = 0.01). High PV PS density ratio may indicate likelihood of PVI success.
In this computational modelling study, we demonstrated that the PVs can play a large role in arrhythmia maintenance and initiation, beyond being simply sources of ectopic beats. We separated the effects of PV properties and atrial fibrosis on arrhythmia inducibility, maintenance mechanisms and the outcome of PVI, based on population or individual patient data. PV properties affect arrhythmia susceptibility from ectopic beats; short PV APD increased arrhythmia susceptibility, while longer PV APD was found to be protective. Arrhythmia inducibility increased with slower CV at the LA/PV junction, but not for cases with homogeneous CV changes or slower CV at the distal PV. The effectiveness of PVI is usually attributed to PV ectopy, but our study demonstrates that the PVs affect reentry in other ways and this may, in part, also account for success or failure of PVI. Both PV properties and fibrosis distribution affect arrhythmia dynamics, which varies from meandering rotors to PV reentry (in cases with baseline or long APD), and then to stable rotors at regions of high fibrosis density. PS density in the PV region was high for cases with PV fibrosis. The measurement of fibrosis and PV properties may indicate patient specific susceptibility to AF initiation and maintenance. PV PS density before PVI was higher in cases in which AF terminated or converted to a macroreentry; thus, high PV PS density may indicate likelihood of AF termination by PVI alone.
PV repolarisation is heterogeneous in the PVs [23], and exhibits distinct properties in AF patients, with Rostock et al. reporting a greater decrease in PV ERP than LA ERP in patients with AF, termed AF begets AF in the PVs [21]. Jais et al. found that PV ERP is greater than LA ERP in AF patients, but this gradient is reversed in AF patients [3]. ERP measured at the distal PV is shorter than at the LA/PV junction during AF [5, 22]. Motivated by these clinical and experimental studies, we modelled a decrease in PV APD, which was applied either homogeneously, or as a gradient of decreasing APD along the length of the PV, with the shortest APD at the distal PV rim. An initial decrease in APD increased inducibility (Table 1), which agrees with clinical findings of increased inducibility for AF patients. Applying this change following a gradient, as observed in previous studies, led to an increased inducibility compared to a homogeneous change in APD. Similar to Calvo et al. [34] we found that rotor location depends on PV APD (Fig 2). Thus PV APD affects PVI outcome in two ways; on the one hand, decreasing APD increases inducibility, emphasising the importance of PVI in the case of ectopic beats; on the other hand, PV PS density decreases for cases with short PV APD, and PVI was less likely to terminate AF.
Multiple studies have measured conduction slowing in the PVs [3, 5, 21–24]. We modelled changes in tissue conductivity either homogeneously, or as a function of distance along the PV. Simply decreasing conductivity and thus decreasing CV, decreased inducibility (Table 2). Kumagai et al. reported that conduction delay was longer for the distal to ostial direction [22]. We found that modifying conductivity following a gradient, with CV decreasing towards the LA/PV junction, resulted in an increase in inducibility in the model. This agrees with the clinical observations of Pascale et al. [1]. This suggests that PVI should be performed in cases in which CV decreases towards the LA/PV junction as these cases have high inducibility. Changes in CV may also be due to other factors, including gap junction remodelling, modified sodium conductance or changes in fiber direction [5, 29].
A variety of PV fiber patterns have been described in the literature and there is variability between patients. Interestingly, all of the PV fiber directions considered in our study showed an increased inducibility compared to the baseline model. Verheule et al. [29] documented circumferential strands that spiral around the lumen of the veins, motivating the arrangements for cases 1 and 4 in our study; Aslanidi et al. [15] reported that fibers run in a spiralling arrangement (case 2); Ho et al. [30] measured mainly circular or spiral bundles, with longitudinal bundles (cases 3 and 5); Hocini et al. [5] reported longitudinal fibers at the distal PV, with circumferential and a mixed chaotic fiber direction at the PV ostium (case 6). Using current imaging technologies, PV fiber direction cannot be reliably measured in vivo.
In our study, fiber direction at the PV ostium was found to be more important than at the distal PV; the greatest inducibility was for cases with circular fibers at the ostium on both endocardial and epicardial surfaces, independent of fiber direction at the distal PV end. Similar to modelling studies by both Coleman [35] and Aslanidi [15], inducibility increased due to conduction block near the PVs.
PVs may be larger in AF patients compared to controls [4, 36], and this difference may vary between veins; Lin et al. found dilatation of the superior PVs in patients with focal AF originating from the PVs, but no difference in the dimensions of inferior PVs compared to control or to patients with focal AF from the superior vena cava or crista terminalis [37]. We found that inducibility increased with PV area for the LSPV, LIPV and RIPV, but not for the RSPV (see Fig 5). In addition, PV PS density ratio increased with total PV area, suggesting that PVI alone is more likely to be a successful treatment strategy in the case of larger veins. However, Den Uijl et al. found no relation between PV dimensions and the outcome of PVI [38]. Rotors were commonly found in areas of high surface curvature, including the LA/PV junction and left atrial appendage ostia, which agrees with findings of Tzortzis et al. [39]. However, there were differences in PS density between geometries, with varying importance of the LA/PV junction (Fig 5), demonstrating the importance of modelling the geometry of an individual patient.
Myocardial tissue within the PVs is significantly fibrotic, which may lead to slow conduction and reentry [5, 30, 40]. More fibrosis is found in the distal PV, with increased connective tissue deposition between myocardial cells [41]. We modelled interstitial PV fibrosis with increasing density distally, and found that the inclusion of PV fibrosis increased PS density in the PV region of the model due to increased reentry around the LA/PV junction and wave break in the areas of fibrosis. This, together with the results in Fig 6, suggests that PVI alone is more likely to be a successful in cases of high PV fibrosis. There are multiple methodologies for modelling atrial fibrosis [25, 42, 43], and the choice of method may affect this localisation.
Population based distributions of atrial fibrosis were modelled for paroxysmal and persistent patients, together with varied PV properties. The presence of LA/PV reentry depends on both PV properties and the presence of fibrosis; reentry is seen at the LA/PV junction for cases with baseline PV APD, but not for short PV APD, and stabilised to areas of high fibrosis in persistent AF, for which LA/PV reentry no longer occurred. This suggests that rotor location depends on both fibrosis and PV properties. This finding may explain the clinical findings of Lim et al. in which drivers are primarily located in the PV region in early AF, but AF complexity increased with increased AF duration, and drivers are also located at sites away from the PVs [6]. During early AF, PV properties may be more important, while with increasing AF duration, there is increased atrial fibrosis in the atrial body that affects driver location. This suggests that in cases with increased atrial fibrosis in the atrial body, ablation in addition to PVI is likely to be required.
Simulations of models with patient-specific atrial fibrosis together with varied PV properties performed in this study offer a proof of concept for using this approach in future studies. The level of atrial fibrosis and PV properties that gave the best fit of the model PS density to the clinical PS density varied between patients. Measurement of PV ERP and conduction properties using a lasso catheter before PVI could be used to tune the model properties, together with LGE-MRI or an electro-anatomic voltage map.
It is difficult to predict whether PVI alone is likely to be a successful treatment strategy for a patient with persistent AF [44]. This will depend on both the susceptibility to AF from ectopic beats, together with electrical driver location, and electrical size. Our study describes multiple factors that affect the susceptibility to AF from ectopic beats. Measurement of PV APD, PV CV and PV size will allow prediction of the susceptibility to AF from ectopic beats. Arrhythmia susceptibility increased in cases with short PV APD, slower CV at the LA/PV junction and larger veins, suggesting the importance of PVI in these cases.
The likelihood that PVI terminates AF was also found to depend on driver location, assessed using PS density. Our simulation studies suggest that high PV PS density indicates likelihood of PVI success. Thus either measuring this clinically using non-invasive ECGi recordings, or running patient-specific simulations to estimate this value may suggest whether ablation in addition to PVI should be performed. In a recent clinical study, Navara et al. observed AF termination during ablation near the PVs, before complete isolation, in cases where rotational and focal activity were identified close to these ablation sites [45]. These data may support the PV PS density metric suggested in our study. Our simulations show that PV PS density depends on PV APD, the degree of PV fibrosis and to a lesser extent on PV fiber direction. To the best of the authors’ knowledge, there are no previous studies on the relationship between fibrosis in the PVs, or PV fiber direction, and the success rate of PVI. Measuring atrial electrogram properties, including AF cycle length, before and after ablation may indicate changes in local tissue refractoriness [46]. PV APD can be estimated clinically by pacing to find the PV ERP; and PV fibrosis may be estimated using LGE-MRI, although this is challenging, as the tissue is thin. PV fiber direction data is not currently available clinically, which limits the predictive ability of the model. Areas of high PV PS density on ECGi need to be carefully interpreted in terms of expected accuracy of the inverse solution on the PVs and the incidence of false phase singularity detection [47]. In addition, multiple mechanisms may underlie areas of high PS density. Importantly, not all PSs sustain and drive AF, and represent suitable targets for ablation.
Limitations to this study include that PV branching structures were not considered since PVs were trimmed at the highest level that results in a single PV rim at each distal PV. Mansour et al. found that just 56% of patients had four PVs with separate ostia [48], 29% of patients had an additional PV, and 17% a common PV trunk. Although some studies have reported differences in ERP between the endocardium and epicardium [23], we modelled the endocardium and epicardium ERP identically. Furthermore, we modelled changes in APD by modifying IK1 only and did not consider other ionic conductances or methods for parametrisation [20, 49, 50]. We used a bilayer model, rather than a volumetric model incorporating thickness, which will affect rotor drift [51]. In addition, we did not model changes in connexins [29] or cell morphology [52]. Furthermore, we modelled 2 seconds of activity following PVI in the model, where these ablation lesions were applied simultaneously rather than sequentially as in the clinic, and we did not model long term AF recurrence. Finally, we did not consider the case of AF sustained by focal beats; we either considered the inducibility due to PV ectopics, or maintenance due to reentry.
Our computational modelling study suggests that measurement of fibrosis and PV properties may indicate patient specific susceptibility to AF initiation and maintenance. In addition, high PV PS density pre-ablation indicates likelihood of PVI success in our simulations, motivating a retrospective clinical study into this metric.
|
10.1371/journal.pmed.1002085 | Dietary Diversity, Diet Cost, and Incidence of Type 2 Diabetes in the United Kingdom: A Prospective Cohort Study | Diet is a key modifiable risk factor for multiple chronic conditions, including type 2 diabetes (T2D). Consuming a range of foods from the five major food groups is advocated as critical to healthy eating, but the association of diversity across major food groups with T2D is not clear and the relationship of within-food-group diversity is unknown. In addition, there is a growing price gap between more and less healthy foods, which may limit the uptake of varied diets. The current study had two aims: first, to examine the association of reported diversity of intake of food groups as well as their subtypes with risk of developing T2D, and second, to estimate the monetary cost associated with dietary diversity.
A prospective study of 23,238 participants in the population-based EPIC-Norfolk cohort completed a baseline Food Frequency Questionnaire in 1993–1997 and were followed up for a median of 10 y. We derived a total diet diversity score and additional scores for diversity within each food group (dairy products, fruits, vegetables, meat and alternatives, and grains). We used multivariable Cox regression analyses for incident diabetes (892 new cases), and multivariable linear regression for diet cost. Greater total diet diversity was associated with 30% lower risk of developing T2D (Hazard ratio [HR] 0.70 [95% CI 0.51 to 0.95]) comparing diets comprising all five food groups to those with three or fewer, adjusting for confounders including obesity and socioeconomic status. In analyses of diversity within each food group, greater diversity in dairy products (HR 0.61 [0.45 to 0.81]), fruits (HR 0.69 [0.52 to 0.90]), and vegetables (HR 0.67 [0.52 to 0.87]) were each associated with lower incident diabetes. The cost of consuming a diet covering all 5 food groups was 18% higher (£4.15/day [4.14 to 4.16]) than one comprising three or fewer groups. Key limitations are the self-reported dietary data and the binary scoring approach whereby some food groups contained both healthy and less healthy food items.
A diet characterized by regular consumption of all five food groups and by greater variety of dairy, fruit, and vegetable subtypes, appears important for a reduced risk of diabetes. However, such a diet is more expensive. Public health efforts to prevent diabetes should include food price policies to promote healthier, more varied diets.
| Diet is a known modifiable risk factor for chronic diseases, and poor quality diets are linked with risk of type 2 diabetes.
A varied diet is advocated as being critical to healthy eating, but people can vary in consumption of different food groups, and also of different subtypes within major food groups.
We analysed self-reported diet data and data on new-onset type 2 diabetes diagnosis in middle- and older-aged women and men from the EPIC-Norfolk cohort study.
Total diet diversity and diversity within major food groups has not previously been studied in relation to health outcomes.
This large United Kingdom study provides evidence that reported intake of a diet that is diverse in subtypes within the dairy, fruit, and vegetable food groups is independently associated with lower type 2 diabetes risk.
People who reported consuming all five food groups had a 30% reduced incidence of type 2 diabetes, but the cost of such a diet was 18% higher (£4.15/day [4.14 to 4.16]) than a diet comprising three or fewer food groups.
Diversity of food groups and subtypes within dairy, fruits, and vegetables is important for chronic disease prevention.
Health promotion efforts need to incorporate financial strategies to support greater dietary diversity.
| Non-communicable diseases present a significant challenge to both high-income and low-income countries, with growing numbers of people experiencing the health and economic burden of one or more chronic conditions [1]. Diet is a key modifiable risk factor for multiple chronic diseases, with poor quality diets being a leading cause of type 2 diabetes (T2D), cardiovascular diseases, hypertension, and certain cancers [2]. It is estimated that diets that do not match nutritional guidelines contribute to 70,000 premature deaths in the United Kingdom [3]. Inadequate consumption of fruits and vegetables in particular is estimated to contribute to 5% of excess mortality globally [2]. Many national and international policies acknowledge the importance of supporting individuals in achieving a healthy balanced diet, and numerous dietary guidelines emphasise the critical role of the consumption of a diet that is varied and includes different foods from different food groups [2,4–7].
Previous aetiological work has tended to examine the association between diet and health by studying individual nutrients, certain food groups, or overall diet quality. Although greater intake of different food subtypes (minor food groups) from each major food group is crucial for nutritional adequacy [8], indices of diet quality rarely include a measure of dietary diversity and none address variety within food groups other than for fruits and vegetables [9,10]. Recent prospective studies in the EPIC cohort indicated that consuming a higher number of different items within the fruit (0–58) and/or vegetable (0–59) food groups was associated with a reduced risk of T2D [11] and certain cancers [12,13], independent of known confounders and quantity of intake. Furthermore, specific subtypes of dairy products are also likely to matter for T2D, specifically low-fat fermented items such as yoghurt [14]. Consumption of a higher number of major food groups has been associated with lower all-cause and cause-specific mortality [15,16]. More recently, however, analysis in a multi-ethnic cohort concluded that a higher number of different food items (between 0 and 120) consumed at least twice a week was not associated with incident T2D [17].
It is possible that a diet that is comprised of all five major food groups could still rely on consumption of a narrow range of foods within each food group. In that sense, it would have overall diversity at the major food group consumption level but would not be varied in terms of different subtypes of foods. Therefore, we aimed to investigate how variation between and within each major food group was related to diabetes risk. We hypothesised that greater diversity across major food groups would be associated with lower T2D incidence and that there would be an independent impact of greater diversity of minor food groups within each major group. A secondary aim was to assess the monetary cost associated with dietary diversity, and we expected a greater cost associated with greater diversity.
A prescribed informed consent statement was signed by all participants in the EPIC-Norfolk study. The study was approved by the Norwich District Health Authority Ethics Committee.
The EPIC-Norfolk study is a population-based prospective cohort study that has been described in detail elsewhere [18]. In brief, EPIC-Norfolk included 25,639 participants (55% women) aged 40–79 years (99.7% white) who were recruited from age-sex registers of general practices in a geographically circumscribed area in the East of England, and who attended a clinical assessment at cohort entry (1993–1997). Participants were followed up using an 18-mo postal questionnaire, a second clinical assessment (1998–2000), and a second postal questionnaire (2002–2004). We excluded participants with known diabetes at baseline (n = 855), unknown diabetes status (n = 5), or missing information on potential confounders (n = 1,541), providing a final sample of 23,238 individuals for analysis (S1 Fig).
New T2D cases were ascertained from multiple sources: two follow-up health and lifestyle questionnaires providing self-reported information on doctor-diagnosed diabetes or medications; medications brought to the second clinical exam; and record linkage. Record linkage to external sources included the listing of any EPIC-Norfolk participant in the general practice diabetes register, local hospital diabetes register, hospital admissions data with screening for diabetes-related admissions, and Office of National Statistics mortality data with coding for diabetes. Participants who self-reported a history of diabetes which could not be confirmed against any other sources were not considered as confirmed cases. Follow-up was censored at date of diagnosis of T2D, 31 July 2006, or date of death, whichever came first.
A semi-quantitative Food Frequency Questionnaire (FFQ) was used to assess habitual dietary intake at baseline, asking respondents to “estimate average food use during the last year” for 130 of the most commonly consumed food and beverage products. The FFQ provided a standard serving size for each product with nine standard response categories, from never or less than once/month to six or more/day [19]. A separate question was concerned with daily intake of milk, with six possible responses from none to more than one pint.
We used raw frequency data to construct a summary score to assess total diet diversity based on a count of five major food groups used in current food guides for eating well: dairy products, fruits, vegetables, grain/cereal products, and meat and alternatives (protein) [8,20,21]. We also constructed additional scores for dairy diversity (milk, cheese, yoghurt), fruit diversity (vitamin A-rich, citrus and berry, other), vegetable diversity (vitamin A-rich, dark green leafy, starchy tubers, other), “meat and alternatives” diversity (flesh meat—red (including processed), organ meat, flesh meat—poultry, fish and seafood, eggs, legumes/beans and nuts and seeds), and grain diversity (whole grains, non-wholegrains). We assigned individual FFQ items to specific subtypes within each major food group based on previous work [8] and United Nation’s Food and Agriculture Organization food group classification guidance [22] (see S1 Table). Similar to other studies [23], items consumed at least twice per week were considered to constitute habitual intake and counted in the relevant food group. A participant scored zero when they reported intakes of an item to be once a week, 1–3 a month, or never/less than once a month. FFQ responses of one pint (0.5683 L) or more than one pint counted as daily milk intake based on dietary guidelines of 3 cups/d (1.249 imperial pint) [24]. Mixed dishes (e.g., soups, quiche) were separated into main components using codebook description of standard recipes [25], and assigned to relevant food groups and subtypes when ingredients contributed at least 10% to the dish’s total weight or were listed among the top five components. For items with unavailable codebook recipes, we used online lists of ingredients for common brands (e.g., Heinz oxtail soup). Each diversity score increased by one when a different food group was consumed; the score increased regardless of the quantity of an item from a given group or the number of possible items from the same group. We also calculated a composite score for diversity of intake of all food group subtypes (0–18).
Covariables based on completed health and lifestyle questionnaires included education level (four categories), UK Registrar General’s occupational social class (six categories), smoking status (three categories) [26], overall physical activity (four categories), and history of myocardial infarction, stroke, or cancer and family history of diabetes (binary). Waist circumference, height, and weight were measured to standard protocol, and body mass index (BMI) calculated as kg/m2.
The monetary cost of the reported diets was estimated by linking food price data for individual foods to the EPIC FFQ’s nutrient composition database as described previously [27]. Retail prices for each of the 289 component food items in the FFQ were obtained by using standardized and published price collection methods [28]. In brief, each food and drink item in the FFQ was priced by using MySupermarket.com, a website for comparing supermarket food prices nationwide in the United Kingdom. For each of the 289 items in the FFQ, we selected the lowest, non-sale price from among the five nationwide retailers on the website at that time (June 2012): Tesco, Sainsbury’s, Asda, Waitrose, and Ocado, which together had a 68% market share at that time [29]. For packaged food (including most fresh produce), we selected the middle size of the range of size options or the larger size if only two options were available. As described previously [28,30], prices were adjusted for preparation losses and cooking fraction to yield an adjusted food price of £/100 g edible portion. The addition of this new variable to the EPIC-Norfolk’s food and nutrient database [31] allowed the derivation of dietary cost for each participant. The variable associated with each individual's diet was cost per day (£/d).
Means with standard deviations and frequencies were used to describe the characteristics of the cohort across three levels of the total diet diversity score (≤3, 4, or 5). Covariance matrices were used to assess the strength of relationships between diversity scores. Multivariable Cox regression analyses were used to examine the relationship between each diversity score and the risk of developing T2D. Hazard ratios (HR) and 95% confidence intervals (95% CI) were estimated using a series of models: model 1 adjusted for age, sex, BMI, and total energy intake (Kcal) (n = 23,912); model 2 additionally adjusted for lifestyle factors (smoking status, alcohol intake (units/week) and physical activity level) plus family history of diabetes (n = 23,705); and model 3 further adjusted for socioeconomic status (education and occupational social class) (n = 23,238). Using model 3, the independent relationship of total diet diversity and T2D was then examined by separately including each specific food group diversity score and by including all five specific food group diversity scores. In addition, the independent relationship of each specific food group diversity score with T2D was examined by including (1) the total diet diversity score, (2) the four other specific food group diversity scores, or (3) the total diet diversity score and all other specific food group diversity scores.
Sensitivity analyses included the total quantity of intake of all items from the relevant food group in model 3 to control for the relationship between the diversity of food groups and the number of foods reported, which is independently associated with nutrient adequacy [32]. Waist circumference, as a marker of central adiposity, was also included in model 3, as it may be an independent risk factor of cardio-metabolic conditions [33]. Vegetable diversity was re-examined after excluding all potato items and, alternatively, restricting to baked and boiled potatoes given the high consumption in the UK of fried potato products, which would contribute to higher fat and energy intakes. Analyses were also repeated after additionally excluding participants with self-reported chronic conditions. We also undertook a sensitivity analysis in the sub-sample of EPIC-Norfolk (n = 10,787) in whom HbA1c was measured at baseline to exclude individuals (n = 262) who had a baseline HbA1c ≥6.5% (or ≥48 mmol/mol), which is indicative of prevalent but undiagnosed diabetes.
Multivariable linear regression was used to assess cross-sectional associations at baseline between each diversity score and diet cost, adjusting for age, sex, and total energy intake (n = 23,238). We used regression coefficients for post-estimation calculation of adjusted means (95% CI). Statistical analyses were conducted using Stata version 13.1.
The average duration of follow up was 10 (±1.5) y, and we identified 892 new cases of T2D over 245,045 person-years of follow up. On average, participants reported consuming 4.7 (0.6) major food groups at least twice or more per week. Very few participants reported consuming foods from two groups (0.45%), one group (0.07%), or none (0.01%); while most reported consuming four (21.29%) or five groups (74.43%) and some consumed only three groups (3.75%). Within the specific food groups, there was more evidence of heterogeneity in reported diets between individuals. A diversity score of zero was observed in 13.4% of participants for dairy products, 7.8% for fruits, and 8.1% for meat (and alternatives). For participants who scored three for total diet diversity, we found that 80% scored zero for dairy, 62% for fruit, 10% for vegetables, 55% for meat, and 9% for grain. And among participants who scored four for total diet diversity, there were 47% scoring zero for dairy, 24% for fruit, 1% for vegetables, 27% for meat and 1% for grain.
Total diet diversity was positively correlated with diversity within each specific food group: dairy (r = 0.52), fruits (r = 0.42), vegetables (r = 0.28), meat and alternatives (r = 0.35), and grains (r = 0.21). The specific food group diversity scores were not correlated with each other, except for the scores for diversity in fruits and vegetables (r = 0.23), and vegetables and meat and alternatives (r = 0.25). Table 1 shows that participants who reported regular consumption of a diet with greater total diet diversity had more favourable socioeconomic and lifestyle profiles.
As shown in Table 2, the total diet diversity score and the specific food group diversity scores for dairy products, fruits, and vegetables were each inversely associated with risk of developing T2D (Model 1). Participants who reported meeting the recommendation to consume foods from all five food groups had a 30% lower incidence of T2D (HR 0.70 [0.51, 0.95]), but those consuming only four major food groups did not have a lower risk (HR 0.85 [0.62, 1.18]) compared to those reporting intakes of three or fewer food groups. Similarly, those participants who reported the greatest level of diversity of consumption of dairy products, fruits, or vegetables had a 38% (HR 0.62 [0.47, 0.83]), 35% (HR 0.65 [0.50, 0.84]), and 33% (HR 0.67 [0.52, 0.86]), respectively, lower risk of T2D compared to the individuals with the least variation of subtypes within a specific food group. In the case of these three specific food groups, there was a significant linear trend with the risk of developing diabetes being inversely related to the degree of food group diversity. There was no association with diversity within the meat or grain food groups. Adjustment for family history and lifestyle factors (Model 2) and additionally for socioeconomic status (Model 3) did not appreciably alter the HRs. We also observed a strong inverse association between the summary score for diversity of all food group subtypes and risk of developing type 2 diabetes (p for trend <0.01) (S2 Table).
After additionally mutually adjusting for all diversity scores within specific food groups, the inverse association of total diet diversity with diabetes risk was attenuated and became non-significant (p = 0.47) (Table 3, Model 6). In analyses adjusting for the association of other specific food group diversity scores, the inverse association of dairy, fruit, and vegetable diversity with T2D remained statistically significant (Table 4, Model 2). However, after accounting for total diet diversity and all other specific food group diversity scores, only dairy and vegetable diversity were significantly independently associated with diabetes risk (Table 4, Model 3).
Inclusion of total quantity of all items from a given food group attenuated results for total diet diversity and dairy diversity, although inverse associations were amplified for vegetable diversity and unaffected for fruit diversity. Results were unaffected in sensitivity analyses after including waist circumference or excluding participants with self-reported chronic conditions. After excluding participants with a baseline level of HbA1c ≥ 6.5%, greatest fruit diversity and vegetable diversity showed stronger inverse associations with T2D (HR 0.42 [0.23, 0.75] and HR 0.56 [0.32, 0.97], respectively) as did total diet diversity (HR 0.53 [0.28, 1.00]) (S3 Table). Finally, inverse associations were stronger for total diet diversity and similar for vegetable diversity when we counted only baked and boiled potatoes, or did not count potato items consumed at least twice a week (S4 Table).
The adjusted mean diet cost was 18% higher for participants consuming all five major food groups (£4.15/day [4.14 to 4.16]), and 7% for four major food groups (£3.85/day [3.83 to 3.88]), compared to those reporting limited diversity (£3.53/day [3.48 to 3.59]) (Table 5 and S2 Fig). The comparison for the costs of diversity within the specific food groups suggested differences between extreme categories of 7% (£0.29/day) for dairy diversity (4.25 [4.22, 4.29] versus 3.96 [3.93, 3.99]), 22% (£0.81/day) for fruit diversity (4.43 [4.41, 4.45] versus 3.63 [3.59, 3.67]), 30% (£1.01/day) for vegetable diversity (4.40 [4.38, 4.42] versus 3.39 [3.35, 3.44]), 42% (£1.47/day) for meat (and alternatives) diversity (4.93 [4.87, 5.00] versus 3.48 [3.44, 3.51]), and -1% (£0.06/day) for grain diversity (4.05 [4.04, 4.06] versus 4.11 [4.08, 4.14]) (Table 5 and S2 Fig). The summary score for diversity of all food group subtypes was also associated with a significant added diet cost (p for trend < 0.001) (Table 5).
This prospective population-based cohort study of 23,238 British adults suggested that individuals who report regular weekly consumption of all five major food groups subsequently had a lower risk of developing type 2 diabetes as did people who consumed diets that were rich in variability within the dairy, fruit, and vegetable food groups. The association of total diet diversity was attenuated after accounting for diversity within the five food groups. However, greater diversity within dairy, fruit, and vegetable food groups remained predictive of diabetes, independent of diversity in each of the other food groups. The cost of a diet that was varied was significantly higher than the cost of one that was the least diverse.
Previous epidemiological studies show that several diet quality indices are associated with 9%–13% reduced risk of T2D [9,10]. However, these studies do not separately examine the role of dietary diversity in relation to T2D. Variety of foods is only considered as a component of a few diet quality indices (e.g., Healthy Eating Index and Dietary Guidelines Index) [9,10]. Studies using diet diaries have shown the risk of T2D is lower by 14%–21% in people reporting higher vegetable intake, particularly green leafy vegetables [11,34], and by 15%–28% with greater reported dairy product intake, specifically yoghurt consumption [14,35]. Previous studies have also examined the broader health impact of total diet diversity, showing higher risk of mortality in people who reported consuming diets with only two food groups or fewer per week when measured by 24 h recall [15,16]. To date, the only published study that examined diversity within specific food groups (fruits and vegetables), showed that people who reported consuming 12 different fruit and vegetable items per week had a 39% lower risk of developing T2D [11]. Despite common advice to consume a varied diet [2,5,7], we are not aware of studies investigating how the number of different food groups and different subtypes within each food group included in a diet are associated with risk of diabetes. Our findings suggest that individuals who meet the recommendation to consume a healthy diet with food items from each of the five food groups had a reduced risk of developing T2D. More notably, our results further showed that people reporting regular consumption of the full range of food subtypes within dairy, fruit, and vegetable food groups also had a reduced risk of T2D.
The biological pathways linking the inverse associations of total diet diversity and diversity within three specific food groups with T2D risk are unclear. A recent study using FFQ data in older adults reports that greater diversity of foods consumed was significantly positively correlated with a more diverse intestinal microbiota, suggesting that dietary diversity influences microbiota composition [36]. Complementary experiments in that study further showed that dietary changes toward lower diversity resulted in losses in the range of different intestinal microbiota and that reduced microbiota diversity was associated with poorer health outcomes [36]. Greater within-group diversity may also have a specific role for health by providing a balance of the multitude of micronutrients, dietary fibre, and other bioactive compounds necessary for maintaining physical functioning [37]. The particular benefits of fruit and vegetable diversity may derive from the inclusion of phytochemicals that are more specific to certain subgroups that individuals with more varied intakes might consume preferentially [38]. For example, greater vegetable diversity may provide individuals with specific subgroups that contain high concentrations of flavonoids and carotenoids, which have known health benefits [39].
While diverse diets may be healthier, they are also more costly [40–42]. Others have reported a difference of 12% in total weekly food expenditure when comparing top and bottom ranges of food variety [43,44]. In the current study, the adjusted mean cost of the whole diet was 18% higher for participants consuming all five food groups compared to those consuming only three or fewer groups. In light of global 5-a-day campaigns emphasising fruit and vegetable variety, it is important for public health efforts to acknowledge that the adoption of diets including all vegetable and all fruit subtypes may be substantially more costly for consumers and may especially exacerbate existing socioeconomic inequalities in diet. Others also note the higher cost of better quality diets [28,45]. Modelling evidence indicates that combining food taxes and subsidies as a multifaceted policy intervention could best support individuals in making healthy food choices so as to prevent chronic conditions and to help reduce health disparities [46,47]. Given the rising price of healthy food groups, there is a need for a comprehensive food pricing strategy to target the increase in the diversity of foods individuals consume, particularly within fruits and vegetables. Further work should investigate how to develop and implement such a policy approach and to evaluate the impact on equity.
The strengths and weaknesses of this study deserve attention. Strengths include a large sample size, prospective study design, thorough assessment of new cases of T2D with self-report information supplemented by external sources, use of established classification of food groups, and comprehensive information on covariables, thereby minimising sources of bias and confounding. In particular, we examined the exposure to different subtypes within each major food group using two approaches (separate within-group scores and a composite score of all food subtypes). The greater magnitude of effect on diabetes incidence and more pronounced diet cost using the composite score further corroborates the primary findings. Another strength of our study was the availability in a subgroup of HbA1c data at baseline, allowing us to confirm that our findings were unaffected by undiagnosed cases of T2D at baseline. However, some potential limitations merit discussion. First, as an observational study, results may be limited by residual confounding or confounding by unmeasured factors. Second, dietary data were based on self-report from FFQ and therefore may be prone to error and bias [19]. In particular, participants who reported diets with limited variety of food groups may have poor completion of the FFQ. Nonetheless, we took an over-inclusive scoring approach, which likely captured diets that had lower levels of diversity. Moreover, FFQ data are suitable for ranking individuals according to habitual intakes and our scoring approach using frequency information avoided the many assumptions used to estimate absolute intakes [19]. However, the diversity scores were limited by the fact that they were based on a simple yes/no for consumption at least twice a week, regardless of the amount consumed, the number of items consumed within a given food group, or the potential healthfulness of an item (e.g., whole-fat versus low-fat milk, lean meats versus red and processed meats, fried fish versus baked fish). In addition, our study did not account for changes in diet diversity and/or changes in other lifestyle factors over follow-up, and our price data were from 2012 because information was not available retrospectively for study baseline. Finally, the EPIC-Norfolk data provides strong external validity and generalisability only to other predominantly European-descended and middle-aged populations.
This large epidemiological study in a population-based cohort is the first to report an association of total diet diversity and diversity within specific food groups with lower risk of diabetes. These findings support current public health recommendations encouraging consumption of all major food groups and also of different types of fruits, vegetables, and dairy products as part of a regular balanced diet. However, the additional cost of greater diversity deserves attention toward a comprehensive food pricing strategy. Future work should investigate how to develop and implement such a policy approach, including the consideration of financial incentives to actively support lower-income groups in achieving a healthy, mixed diet.
|
10.1371/journal.pntd.0000196 | Decision Tree Algorithms Predict the Diagnosis and Outcome of Dengue Fever in the Early Phase of Illness | Dengue is re-emerging throughout the tropical world, causing frequent recurrent epidemics. The initial clinical manifestation of dengue often is confused with other febrile states confounding both clinical management and disease surveillance. Evidence-based triage strategies that identify individuals likely to be in the early stages of dengue illness can direct patient stratification for clinical investigations, management, and virological surveillance. Here we report the identification of algorithms that differentiate dengue from other febrile illnesses in the primary care setting and predict severe disease in adults.
A total of 1,200 patients presenting in the first 72 hours of acute febrile illness were recruited and followed up for up to a 4-week period prospectively; 1,012 of these were recruited from Singapore and 188 from Vietnam. Of these, 364 were dengue RT-PCR positive; 173 had dengue fever, 171 had dengue hemorrhagic fever, and 20 had dengue shock syndrome as final diagnosis. Using a C4.5 decision tree classifier for analysis of all clinical, haematological, and virological data, we obtained a diagnostic algorithm that differentiates dengue from non-dengue febrile illness with an accuracy of 84.7%. The algorithm can be used differently in different disease prevalence to yield clinically useful positive and negative predictive values. Furthermore, an algorithm using platelet count, crossover threshold value of a real-time RT-PCR for dengue viral RNA, and presence of pre-existing anti-dengue IgG antibodies in sequential order identified cases with sensitivity and specificity of 78.2% and 80.2%, respectively, that eventually developed thrombocytopenia of 50,000 platelet/mm3 or less, a level previously shown to be associated with haemorrhage and shock in adults with dengue fever.
This study shows a proof-of-concept that decision algorithms using simple clinical and haematological parameters can predict diagnosis and prognosis of dengue disease, a finding that could prove useful in disease management and surveillance.
| Dengue illness appears similar to other febrile illness, particularly in the early stages of disease. Consequently, diagnosis is often delayed or confused with other illnesses, reducing the effectiveness of using clinical diagnosis for patient care and disease surveillance. To address this shortcoming, we have studied 1,200 patients who presented within 72 hours from onset of fever; 30.3% of these had dengue infection, while the remaining 69.7% had other causes of fever. Using body temperature and the results of simple laboratory tests on blood samples of these patients, we have constructed a decision algorithm that is able to distinguish patients with dengue illness from those with other causes of fever with an accuracy of 84.7%. Another decision algorithm is able to predict which of the dengue patients would go on to develop severe disease, as indicated by an eventual drop in the platelet count to 50,000/mm3 blood or below. Our study shows a proof-of-concept that simple decision algorithms can predict dengue diagnosis and the likelihood of developing severe disease, a finding that could prove useful in the management of dengue patients and to public health efforts in preventing virus transmission.
| Dengue fever/dengue haemorrhagic fever (DF/DHF) is a re-emerging disease throughout the tropical world. The disease is caused by four closely related dengue viruses, which are transmitted by the Aedes mosquitoes, principally Aedes aegypti [1]. DHF and dengue shock syndrome (DSS) represent the severe end of the disease spectrum, which if not properly managed, would result in significant mortality. The pathophysiology of severe DHF and DSS is characterized by plasma leakage as a result of alteration in microvascular permeability [2]. There is as yet no vaccine or specific antiviral therapy for DF/DHF and management of cases remains largely supportive [3].
Dengue illness is often confused with other viral febrile states, confounding both clinical management [4]–[6] and disease surveillance for viral transmission prevention [7]. This difficulty is especially striking during the early phase of illness, where non-specific clinical symptoms and signs accompany the febrile illness [4]. More definitive symptoms, such as retro-orbital pain, and clinical signs, such as petechiae, do not appear until the later stages of illness, if at all. Definitive early dengue diagnosis thus requires laboratory tests and those suitable for use at this stage of illness are either costly, such as RT-PCR for dengue; not sufficiently rapid, such as virus isolation; or undergoing field trials, such as ELISA for NS1 protein of dengue virus [8],[9]. Furthermore, many dengue endemic places lack the necessary laboratory infrastructure or support [7] and thus a scheme for reliable clinical diagnosis, using data that can be obtained routinely, would be useful for early recognition of dengue fever, not only for case management but also for dengue surveillance. The current World Health Organization (WHO) scheme for classifying dengue infection (Table S1) makes use of symptoms and signs that are often not present in the first few days of illness, and thus not a guide for early diagnosis, but are instead designed for monitoring disease progression for clinical management of the severe DHF/DSS. Other attempts at identifying clinical features for the diagnosis of dengue disease have made use of univariate or multivariate analysis of clinical symptoms and signs, haematological or biochemical parameters [10],[11]. Although such studies provide a list of symptoms and signs that could be associated with dengue disease, how these should be applied for clinical diagnosis is not apparent. Evidence-based triage strategies that identify individuals likely to have dengue infection in the early stages of illness are needed to direct patient stratification in clinical investigations, management and healthcare resource planning.
To address this goal, we show here that a decision tree approach can be useful to develop an intuitive diagnostic algorithm, using clinical and haematological parameters, that is able to distinguish dengue from non-dengue disease in the first 72 hours of illness. We also demonstrate a proof-of-concept that such an approach can be useful for early dengue disease prognostication.
All results have been summarized in terms of means and standard deviation for continuous variables using independent sample T-test. Shapiro-Wilk normality test was used to check for non-normally distributed parameters whereby a p value <0.05 indicated that the parameter was unlikely to originate from a normal distribution. Non-normally distributed parameters were log-transformed and rechecked for normality. If the log-transformation still resulted in non-normal distribution, non-parametric Kruskal-Willis test was used for continuous variables whereas Student's t test was exploited for normally distributed continuous variables. For dichotomous variables, Chi-square test was used in case of expected frequencies that were higher than 5, whereas Fisher's exact test was performed when the expected table values were smaller than 5. Cases with missing values were excluded from the analysis and thus, the number of cases used for calculations varied between different covariates. All calculations were performed using Systat for Windows (SYSTAT Software Inc. San Jose, CA). A two-tailed p value <0.05 was considered as statistically significant.
We constructed a decision tree for dengue diagnosis with 1,200 patients with acute febrile illness. Of these, 1,012 were recruited from the EDEN study and 188 from Vietnam. The EDEN cohort consisted of 173 DF, 3 DHF and 836 non-dengue cases while the Vietnam cohort consisted of 168 DHF and 20 DSS cases, resulting in a total of 364 dengue and 836 non-dengue cases used for our diagnostic tree construction.
The resulting diagnostic algorithm is shown in Figure 1. The first splitting parameter is a platelet count of 196,000/mm3 blood or less followed by the total white cell or lymphocyte counts, body temperature, haematocrit or neutrophil count and another platelet count at presentation. The predicted diagnosis is shown in colours, with red indicating probable dengue, brown indicating likely dengue, green indicating likely non-dengue and blue indicating probable non-dengue (Figure 1A). Each of the nodes showed statistical significance in the proportion of dengue and non-dengue cases, with the odds ratio calculated as shown in Figure 1B. The performance of this algorithm is shown in Figure 2. The overall error rate estimated after k-fold cross validation was 15.7%, with a sensitivity and specificity of 71.2% and 90.1%, respectively (Figure 2B).
We also examined if the use of a decision tree would be useful for prognostication.
For the EDEN cohort, we used a platelet count of less than 50,000/mm3 on days 5 to 7 of illness as a marker of severe disease. This level of thrombocytopenia has been shown to be associated with the development of complications such as bleeding and shock in adults [16]–[18].
Fifteen cases were excluded from this analysis as they were either admitted to private hospitals where access to the clinical information was not available to us, or were foreigners who returned to their country of origin to seek medical treatment. Thus, 161 Singaporean dengue cases were analysed and the pruning confidence was set to 25% with minimal cases defined as 16.
The best performing decision algorithm made used of platelet count, the crossover value (Ct) of the real-time RT-PCR for dengue viral RNA (a marker for viremia levels) and the presence of anti-dengue IgG antibodies, as the first, second and third splitting parameters, respectively (Figure 3). All 3 parameters were obtained from the first visit. All three DHF cases were correctly classified using this algorithm, one into the group with a platelet count of 108,000 mm3 or less, the other two into the group with pre-existing anti-dengue IgG antibodies. The predicted outcome of disease is shown in colours, with red indicating probable severe dengue, brown indicating likely severe dengue, green indicating likely non-severe dengue and blue indicating probable non-severe dengue (Figure 3A). The statistical significance of each node of the algorithm and their odds ratio with severe dengue are shown in Figure 3B. The performance of this algorithm is shown in Figure 4. The overall error rate using k-fold crossover validation analysis was 20.5%, with a sensitivity of 78.2% and specificity of 80.2% (Figure 4B).
The use of data obtained from the 89 hospitalised cases alone resulted in a very similar decision algorithm, although the AUCs were substantially lower than the above analysis due largely to the smaller dataset. Taken together, these indicate that the prediction algorithm as defined in Figure 3A is stable.
We next examined the clinical outcomes of the patients grouped according to the decision algorithm in Figure 3A. The results are summarised in Table 1. Each of the four groups of patients showed different rates of hospitalisation, duration of hospitalisation and the proportion of clinically severe cases. The latter was defined as patients who met the criteria for DHF; had a systolic blood pressure less than 90 mmHg; a serum transaminase of greater than 1000 which suggests severe liver involvement and who received blood transfusion. The results indicate that statistically significant differences were observed between the groupings as indicated in Table 1.
The lack of evidence-based diagnostic algorithm for early dengue diagnosis as well as prognostic triage strategies limits effective patient management, use of healthcare resources and disease surveillance efforts. For instance, over 80% of the total dengue cases in Singapore are admitted for hospitalised care, mostly to monitor for signs of clinical deterioration. Prognostication in the early stages of dengue illness could significantly influence clinical management and the use of healthcare resources, particularly during dengue outbreaks, such as occurred in Singapore in 2005 where up to 8% of all acute hospitals beds available were occupied by dengue cases [19].
To identify the decision algorithms, we have used a C4.5 decision tree classifier, which has several advantages over other statistical tools [14]. Briefly, decision algorithms are in principle simple to understand and are able to handle both nominal and categorical data. Importantly, they are also able to handle missing values, which are commonly encountered in clinical studies. In contrast, logistic regression and discriminant analyses require much more data preparation and appropriate handling of missing values for reliable calculations [15]. Decision algorithms are also easy to interpret, use and validate using common statistical techniques. Importantly, it provides a means to identify parameters that would be significantly associated with disease when analysed in sub-groups but not in the total study population. To our understanding, this is the first time decision tree modelling has been used to identify prognostic markers for dengue disease.
While dengue is predominantly a paediatric disease, dengue in adults has become an increasingly recognised problem, both in dengue-endemic regions [6],[20] as well as in adult travellers returning from the tropics [5]. The case recruitment in Singapore has thus focused on adult cases. Since the course of disease in all but three of the Singaporean adult cases were consistent with DF instead of DHF, we have included 188 DHF/DSS paediatric and adult cases from Vietnam in order to ensure that the diagnostic algorithm developed here is robust across a spectrum of dengue presentations.
The decision algorithm for the diagnosis of dengue within the first three days of illness made use of a combination of platelet count, total white cell count, body temperature, absolute lymphocyte and neutrophil counts, in sequential order (Figure 1A). Each node of the decision tree has statistically significant odds ratio ranging from 5.9 to 13.8 (Figure 1B).
Although the tree has an optimal combined sensitivity and specificity of 71.2% and 90.3%, respectively, its usage can be adjusted according to the objective in which it is used for. In an outbreak where the aim is to identify all dengue cases for laboratory investigation and clinical follow-up, the tree could be used to exclude dengue cases whereupon all cases except those predicted as probable non-dengue (shown in blue in Figure 1A) are tested for dengue virus. When applied hypothetically to an outbreak similar to that observed in Singapore in 2005 where 29% of the acute febrile cases recruited into our study was dengue, the positive and negative predictive values of the tree are 57.7% and 94.4%, respectively. Conversely, when the dengue prevalence is low as was encountered in our EDEN study between 2006 and August 2007 where only 43 out of 555 (7.7%) cases presenting with acute febrile illness were dengue, increasing the specificity of clinical diagnosis by selecting patients with probable dengue (shown in red in Figure 1A) would result in a positive and negative predictive values of 51.1% and 97.7%, respectively. Such a level of positive predictive value would be useful to guide the selection of patients for virological surveillance, a critical part of any dengue prevention program [7], [21]–[23].
Upon diagnosis, current dengue management strategies require daily observation for signs of clinical deterioration, particularly for clinical or laboratory evidence of hemorrhage or plasma leakage. In situations of high prevalence of dengue illness, such an approach can quickly overwhelm limited healthcare resources. It would be advantageous to be able to stratify dengue cases for clinical follow-up and management based on the likely outcome of disease. We thus searched for an algorithm that could be used for prognostication. Since the incidence of DHF is low in Singapore [20] and hospitalisation of the dengue cases is subject to variation arising from physician-to-physician differences in decision-making, we have used platelet count nadir of 50,000/mm3 or less at 5 to 7 days after onset of illness as an objective end-point for our analysis. This level of thrombocytopenia has been found to be associated with increased risks of haemorrhage and shock in adults with DF [16]–[18]. We were unable to include the DHF and DSS cases recruited in Vietnam for the tree construction as daily laboratory parameters comparable to those collected for the Singapore cohort were not available.
The decision algorithm for prognostication (Figure 3A) uses the platelet count as the first splitting criteria, followed by the dengue virus genome copy number estimated by real-time RT-PCR as the second splitting criteria for those with platelet count greater than 108,000/mm3 blood. The second splitting criterion is a marker of viral load. Although we have used the Ct value of our real-time RT-PCR in this analysis, it is likely that other parameters that provide estimates for the viral load could be substituted for the viral genome copy numbers. The development of NS1 antigen ELISA that is currently being evaluated in several places could be one such alternative. The third splitting criterion uses the presence of anti-dengue IgG antibody, indicating secondary infection.
Although thrombocytopaenia [16]–[18], high viremia [24] and presence of pre-existing anti-dengue antibodies [24]–[26] have previously been reported to be associated with severe disease, how these factors should be used clinically for prognostication has never been described. Furthermore, using these parameters singly also presents difficulties since these parameters are dynamic and the window period in which these parameters offer peak predictive values is extremely short [9]. The use of a decision tree approach could thus provide clinicians with an algorithm to guide the evaluation of a panel of critical laboratory parameters and the sequential order these should be considered within the first 72 hours of illness.
Each of the four groups of patients under the decision algorithm for prognosis (Figure 3A) also showed significant differences in clinical outcome (Table 1). While only three cases met all the criteria for DHF according to the WHO dengue classification, the clinical records of another 20 cases showed that they either had a period of hypotension (systolic blood pressure of less than 90mmHg) or severe liver inflammation (liver transaminases>1000), both without documented pleural effusion, ascites or rise in serial hematocrit, or received platelet/blood transfusion. These clinical parameters have been previously observed in severe dengue [15],[16] and we have taken these cases collectively as clinically severe outcomes. Of these 23 cases, 19 (82.6%) were predicted by our tree as either probable severe dengue or likely severe dengue with data obtained in the first three days of illness. Conversely, 91.8% and 100% of the patients in the groups predicted by our tree as either likely non-severe dengue or probable non-severe dengue, respectively, did not show severe clinical outcomes (Table 1).
The use of such a prognostic algorithm could prove useful in segregating patients according to likely clinical outcomes to guide clinical management and follow-up visits. Although our EDEN cohort in Singapore has focused on dengue in the adult population, our findings demonstrate a proof-of-concept that the use of simple haematological and virological parameters is predictive of disease outcome, and can be built upon to develop prognosis-based protocols for dengue case management that begins at the primary healthcare setting.
Our study represents the first to demonstrate that decision algorithms for dengue diagnosis and prognosis can be developed for clinical use. While a large multi-centre prospective study will be needed for these algorithms to be applied globally, our analysis indicates that a decision tree approach can differentiate dengue from non-dengue febrile illness and predict outcome of disease. |
10.1371/journal.pgen.1005309 | Temporal Coordination of Carbohydrate Metabolism during Mosquito Reproduction | Hematophagous mosquitoes serve as vectors of multiple devastating human diseases, and many unique physiological features contribute to the incredible evolutionary success of these insects. These functions place high-energy demands on a reproducing female mosquito, and carbohydrate metabolism (CM) must be synchronized with these needs. Functional analysis of metabolic gene profiling showed that major CM pathways, including glycolysis, glycogen and sugar metabolism, and citrate cycle, are dramatically repressed at post eclosion (PE) stage in mosquito fat body followed by a sharply increase at post-blood meal (PBM) stage, which were also verified by Real-time RT-PCR. Consistent to the change of transcript and protein level of CM genes, the level of glycogen, glucose and trehalose and other secondary metabolites are also periodically accumulated and degraded during the reproductive cycle respectively. Levels of triacylglycerols (TAG), which represent another important energy storage form in the mosquito fat body, followed a similar tendency. On the other hand, ATP, which is generated by catabolism of these secondary metabolites, showed an opposite trend. Additionally, we used RNA interference studies for the juvenile hormone and ecdysone receptors, Met and EcR, coupled with transcriptomics and metabolomics analyses to show that these hormone receptors function as major regulatory switches coordinating CM with the differing energy requirements of the female mosquito throughout its reproductive cycle. Our study demonstrates how, by metabolic reprogramming, a multicellular organism adapts to drastic and rapid functional changes.
| Mosquitoes transmit numerous devastating human diseases due to their obligatory hematophagy that is required for the efficient reproduction. Metabolism must be synchronized with high energetic needs of a female mosquito for host seeking, blood feeding and rapid egg development. Each reproductive cycle is divided into two phases that are sequentially governed by juvenile hormone (JH) and 20-hydroxyecdysone. During the pre-blood meal phase, the JH receptor Methoprene-tolerant (Met) controls carbohydrate metabolism (CM) pathways and its RNA interference (RNAi) silencing caused up-regulation of CM enzymes at the transcript and protein levels activating glycolytic flux and depletion of storage and circulating sugars. During the second, post blood meal phase, CM was regulated by the ecdysone receptor EcR and its RNAi silencing had a dramatic effect opposite to that of Met RNAi. Thus, we show that Met and EcR function as regulatory switches coordinating carbohydrate metabolism with energetic requirements of the female mosquito reproductive cycle.
| The ability of multicellular organisms to maintain metabolic homeostasis and respond to changing energy requirements during development, reproduction and stress represents an essential adaptation critical for survival and evolutionary success. Thus, it is important to decipher regulatory mechanisms coordinating metabolic pathways; understanding these mechanisms in organisms facing extreme and fluctuating energy demands is particularly valuable.
Female mosquitoes, which are obligatory blood feeders, serve as disease vectors [1]. Pathogens, taking advantage of this blood dependency, use mosquitoes as vectors spreading serious human diseases. Despite continuing efforts and advances in insect control, mosquitoes pose an enormous threat, killing over a million people each year. The situation is aggravated by the lack of effective vaccines, fast growing insecticide resistance, social complexities and ecological changes [2]. A detailed understanding of the reproductive biology of the mosquito may provide vital information to take us a step closer to more effective vector-control strategies.
Hematophagous mosquitoes possess numerous distinct physiological features that play a critical role in the stunning environmental adaptations of these disease vector insects. These include a powerful system of odorant receptors, an extremely efficient host-seeking behavior, adaptations for blood feeding and digestion, ability to excrete large amounts of solutes, and rapid egg development [3]. Hematophagy puts extremely high energy demands on a female mosquito at different stages throughout its reproductive cycle. Therefore, metabolic pathways must be synchronized with energy needs of a reproducing female mosquito. However, regulatory mechanisms governing temporal coordination of metabolism at the molecular level have not been well understood in mosquitoes.
Each reproductive cycle of a female mosquito is divided into two phases, which are governed by alternating titers of two major insect hormones—a sesquiterpenoid juvenile hormone (JH) and a steroid hormone 20-hydroxyecdysone (20E). JH guides the female mosquito development from the adult eclosion from pupae to blood feeding. During this JH-controlled post eclosion (PE) phase, which lasts 3–5 days, a female mosquito matures and prepares itself for events associated with subsequent blood feeding, while actively seeking hosts.
Ingestion of blood leads to dramatic events in a female mosquito, including digestion of a huge meal, powerful excretion, a high level of gene expression and rapid egg maturation. During this post blood meal (PBM) phase, a female mosquito faces an intense degree of metabolic activity. 20E is the major regulator of the PBM phase of the female mosquito reproductive cycle, and its action is mediated by a nuclear receptor, the Ecdysone Receptor (EcR) [4]. The mosquito fat body serves as the nutrient sensor organ, detecting the nutrients derived from a blood meal and blood-derived nutrients are utilized for the production of YPPs in fat body cells [1].
In this study, we investigated whether the two major regulators of female mosquito reproductive cycles, JH and 20E, are involved in the temporal coordination of CM throughout the female mosquito reproductive cycle. Our results show that the JH receptor, Met, and the EcR synchronize CM with energy requirements of a reproducing female mosquito. Deciphering the regulatory mechanisms governing mosquito CM has shed light on adaptations of an organism dealing with intense energetic stress.
We characterized transcript abundance of genes encoding CM enzymes in the fat body of the female Aedes aegypti mosquito throughout the first reproductive cycle. For this analysis, we used two time course fat body microarray transcriptomes spanning the entire first gonadotrophic cycle; one that encompassed eight time points from 6h to 72h PE, and a second one covering nine time points from 3h to 72h PBM (S1 and S2 Tables). Both transcriptomes were obtained from custom-made Agilent microarray chips that contained probe sets corresponding to 15,321 genes in the A. aegypti genome [5]. DEG sets during PBM were calculated by comparing transcripts from each of the nine time points with that at 72h PE using the same filtering criteria as those for the PE genes [5]. CM gene transcripts were abundant during the first 24h PE, while later, by 72h PE, there was a considerable decline in their levels (Fig 1 and S1A Fig). Following a blood meal, most CM genes exhibited significant up-regulation reaching their maximal expression by 36h PBM, dropping back to early PE levels by 72h PBM (Fig 1 and S1A Fig). Overall, the second wave of the CM gene activity during the PBM phase was considerably higher than the first during the PE phase. The genes encoding glycogen/sugar metabolism (10 of 17 enzyme coding genes) and glycolysis (13 of 28 enzyme coding genes) exhibited particularly pronounced fluctuations in their expression levels (Fig 1 and S1B and S1C Fig). The genes encoding the citrate cycle exhibited a similar trend, but to a lesser extent (Fig 1).
To authenticate our microarray analysis results, transcript levels of genes encoding key enzymes of CM pathways in fat body samples were determined using real-time PCR (qPCR) (Fig 2A and S2 Fig). In agreement with microarray data, the transcript of the glycogen phosphorylase gene (GLY) encoding the key glycogen degrading enzyme was high at the beginning of the PE phase, but its expression at 72h PE was dramatically reduced (Fig 2A). This is in contrast to a small decrease in the mRNA levels of genes encoding the enzymes involved in glycogen biosynthesis—glycogen synthase (GYS) (S2 Fig). These gene dynamics suggest predominance of glycogen accumulation during the PE phase. During the PBM phase, the GLY transcript was greatly elevated at 36h PBM (Fig 2A), while that of the GYS gene showed only a moderate increase, suggesting a trend opposite to PE in the utilization of sugar reserves during the PBM stage (S2 Fig). Transcripts of genes encoding enzymes for trehalose metabolism, trehalose-6-phosphate (TPS) and trehalose-6-phosphatase (TPP) and trehalase (TREA) (Fig 2A and S2 Fig), the enzymes responsible for transforming glucose to trehalose, declined during the PE phase. During PBM, each of these three genes had a dramatic peak of expression at 36h PBM (Fig 2A and S2 Fig). Transcript levels of nine glycolytic genes, determined using qPCR, were in agreement with microarray data showing differential PE and PBM expression of these genes (Fig 2A and S2 Fig). Our analysis included three genes encoding the rate-limiting enzymes of glycolysis—hexokinase (HEX), phosphofructokinase (PFK), pyruvate kinase (PYK) (Fig 2A and S2 Fig). Although they followed a similar expression trend during the mosquito reproductive cycle, the levels of their relative expression differ significantly (Fig 1). PYK catalyzes the final glycolytic irreversible step generating pyruvate and ATP [6]. Strikingly, the PYK gene expression was highly elevated during the mosquito gonadotrophic cycle, particularly during the PBM stage when its transcript increased over 200-fold, while PFK transcript increased more than 125 folds (Fig 2A). This suggests a dramatic acceleration of the glycolytic flux after blood feeding. Lactate dehydrogenase (LDH) is the enzyme that catalyzes conversion of pyruvate to lactate, and this reaction supplies NAD+ [7]. Notably, there was a 12-fold elevation in the LDH gene transcript level by 6h PBM, suggesting a drastic increase in the generation of lactate immediately after a blood meal (Fig 2A).
Using antibodies that recognize respective Aedes CM enzymes at the protein level (S3 Table), we performed western blot analyses of samples from 6h, 24h, 72h PE and 6h, 36h, 72h PBM developmental time points. Protein levels for the major glycogen-utilizing enzyme, GLY, and two key glycolytic enzymes, PGM and PYK, were high until 24h PE, after which there was a drop, at 72h PE. During the PBM period, all three proteins were in abundance at 36h (Fig 2B). HEX, the enzyme that catalyzes the first step in glycolysis, converting glucose to glucose-1-P, showed a weak accumulation at 72h PE in female mosquito fat bodies. However, this HEX protein could be detected at a much higher level at 36h PBM. Protein levels of all four tested enzymes declined during the late PBM phase (72h PBM) (Fig 2B). Overall, our western blots demonstrate that the protein levels of glycolysis and glycogen/sugar metabolic enzymes exhibit periodic changes throughout the mosquito reproductive cycle that correlate with the transcript abundance of their respective genes.
To find out whether sugar reserves correlated with fluctuating levels of CM enzymes at the gene and protein levels in the female mosquito fat body, we undertook thorough quantitative measurements of stored and circulating sugars during the female mosquito reproductive cycle.
In a newly eclosed female mosquito, the glycogen level was relatively low but increased significantly by 24h PE (about 150% of that of 6h PE) and was maintained at a similar level for the rest of the PE developmental phase (Fig 3A). A blood meal, however, triggers glycogen depletion and by 24h PBM its level dropped to about half that of the late PE phase mosquitoes. The glycogen level increased by 72h PBM, but was still lower than that of 72h PE mosquitoes (Fig 3A). In order to visualize the glycogen content in situ, we used Periodic acid/Schiff (PAS) staining of fixed female adult mosquito fat bodies. Glycogen content was at a detectable level at 6h PE. Consistent with our colorimetric measurements of glycogen, there was a significant increase in PAS positive signal in the 24h PE fat body. Using this staining method, however, the highest glycogen level was observed at 72h PE (Fig 3A). Correlating well with glycogen level measurements, PAS staining showed that glycogen content was less at 6h PBM than at 72h PE. Staining also revealed that the glycogen levels were moderately increased from 6h to 72h PBM in the fat body (Fig 3A).
We then measured the levels of circulating sugars using gas chromatography—mass spectrometry (GC-MS). Trehalose level increased during the PE phase, reaching its maximum at 72-78h PE (about 3-fold increase in comparison with that at 0-6h PE) (Fig 3B). A blood meal triggers depletion of trehalose and by 24h PBM its concentration dropped to about half that of late-stage PE mosquitoes. During late PBM, the level of trehalose increased and returned to original level by 72h PBM (Fig 3B). Although trehalose is the major form of the circulating sugar in insects, glucose and fructose function as additional circulating sugars found in the hemolymph [7,8]. During the late PE phase, there was approximately 10- and 20-fold increase in glucose and fructose levels, respectively (Fig 3B). Blood feeding resulted in a decrease in the levels of these two sugars until 36h PBM, after which it was restored back to PE levels by 72h PBM (Fig 3B).
Triacylglycerols (TAG) represent another important energy storage form in the mosquito fat body. During the PE phase, the change in TAG level was delayed compared with that of glycogen. The TAG level was relatively low from 6 to 24h PE, but increased by 72h PE (Fig 3C). During PBM phase, the TAG levels in the fat body dropped significantly at 6h PBM reaching its lowest level by 72h PBM (Fig 2C).
Adenosine triphosphate (ATP) serves as a major indicator of energy consumption by an organism [7]. To evaluate energy utilization in female mosquitoes throughout the reproductive cycle, we measured ATP levels using high performance liquid chromatography (HPLC). The ATP level was high in newly eclosed female mosquitoes at 6h PE, declining thereafter, and by 72-78h PE its level was only 50% of that of 6h-old mosquitoes (Fig 3D). However, the ATP level increased during the PBM phase, reaching a peak at 48h PBM, which was higher than that at 72h PE (Fig 3D).
To provide further insight into the CM dynamics in reproducing female mosquitoes, we used GC-MS to measure several intermediary metabolites (IMs) of glycolysis and the citrate cycle. Overall, this analysis revealed that IM profiles correlated with those of CM enzymes, exhibiting two pronounced waves at the PE and PBM phases, respectively (Fig 4). Glucose-6-phosphate represents the first key IM of the glycolytic pathway that also serves as a precursor for glycogen/sugar metabolism and pentose-phosphate pathways. During the PE phase, the level of glucose-6-phosphate was reduced 2-fold by 72-78h (Fig 4A). This metabolite also showed a significant drop in its level immediately after a blood meal, at 6h PBM, and its level remained low throughout the PBM phase, being elevated only by 72h PBM (Fig 4A). The level of the next glycolytic IM, fructose-6-phosphate, exhibited a 2-fold reduction by 72–78 h PE, but it was elevated at 6h PBM, maintaining its high level throughout the rest of the PBM phase (Fig 4A). There was a dramatic reduction in the level of pyruvate, the terminal product of glycolysis during the PE phase. However, more than a 100% increase of the pyruvate level was observed at 6h PBM, reflecting an increase in the glycolytic flux following the blood intake. The pyruvate level remained high until 36h PBM, declining thereafter (Fig 4A). Our transcript data analyses demonstrated a reduction in the mRNA level of the gene encoding LDH, the enzyme catalyzing transformation of pyruvate to lactate, at the end of PE period (Fig 1). Accordingly, GC-MS measurements of lactate showed a pronounced drop in its level late PE. Moreover, there was an elevation in the lactate level during the PBM phase corresponding to the rise in the expression of this gene (Fig 4A).
The citrate cycle is the key pathway used for energy production in all aerobic organisms. Pyruvate serves as an essential precursor for the citrate cycle, and its availability along with activity levels of citrate cycle enzymes determines the final outcome. The latter can be determined by measuring concentrations of citrate cycle IMs. The GC-MS analysis revealed considerable differences in PE and PBM profiles of citrate cycle IMs, which reflects contrasting energetic requirements of the female mosquito during these two phases of the reproductive cycle (Fig 4B). The level of citrate, the first IM of the citrate cycle, exhibited a significant reduction over the PE phase, while it was highly elevated at 6h PBM. Succinate and fumarate exhibited more moderate fluctuations. Malate, however, had PE and PBM profiles similar to those of citrate (Fig 4B).
To investigate the role of JH in regulation of CM during PE phase, we topically applied JH III onto newly eclosed female mosquitoes and investigated the effect of this treatment 20h later. The application of JHIII caused a premature drop in abundance of CM gene transcripts (S3A Fig). At the same time, there was a significant elevation in the levels of glycogen and glucose compared with control untreated mosquitoes (S3B Fig).
Our previous data indicated that the JH receptor Met plays a central role in regulating JH-mediated gene expression in the fat body of the PE female mosquito [5]. Met silencing has been shown to inhibit ovarian follicle growth as well as result in the reduction of the egg number [5,9]. We examined the transcriptome obtained from the fat body of Met RNAi-depleted females and analyzed the response of CM genes. The CM gene transcripts were enriched among upregulated gene cohorts of the iMet transcriptome (S4A Fig). The transcripts of genes belonging to glycogen/sugar metabolism and glycolysis were particularly upregulated, while those of the citrate cycle were elevated to a significantly lesser degree (Fig 5A and S4 Table). Next, we silenced Met by RNA interference (RNAi) in female mosquitoes (iMet) at 24h PE and analyzed transcript levels of CM genes 4 days later using qPCR. We measured mRNA levels of four glycogen/sugar metabolism genes—GLY, TPS, TPP, and TREA—in the Met-depleted background and found these genes to be considerably induced (Fig 5B and S4C Fig). Six glycolytic enzyme coding genes, including the rate limiting PYK and HEX, were significantly upregulated in the iMet mosquito fat body (Fig 5B and S4C Fig). We also tested the key rate-limiting enzyme PFK but found no effect of Met, consistent with microarray results (S4C Fig). Western blot analysis showed a substantial accumulation of enzymes for glycogen/sugar metabolism and glycolysis at the protein level in fat bodies of Met-silenced female mosquitoes (Fig 5C). These results demonstrate a dramatic effect of Met RNAi knockdown on CM gene and protein levels, suggesting that the JH receptor plays a critical role in CM regulation.
The glycogen levels were significantly reduced in Met-silenced female mosquitoes (Fig 6A and 6B). A dramatic depletion of glycogen reserves in fat bodies of Met-silenced female mosquitoes was confirmed by means of PAS staining (Fig 6A). Circulating hemolymph sugars—trehalose, fructose and glucose—significantly declined in abundance in Met-depleted female mosquitoes (Fig 6B). Like the sugar reserves, TAG levels also declined after Met depletion (Fig 6C). Met RNAi depletion resulted in elevated ATP levels, showing an increase in energy consumption in these mosquitoes (Fig 6D). We then measured levels of pyruvate and lactate, the metabolic end products of glycolysis, both of which were significantly elevated with Met depletion, indicating that Met affects the glycolytic flux. However, citrate, succinate and malate, IMs of the citrate cycle, showed no noticeable fluctuations in response to Met depletion (Fig 6E). Our data show that CM is severely compromised in Met-silenced female mosquitoes. This effect of Met RNAi silencing clearly demonstrates that Met functions as a major regulatory switch of CM during the PE phase of the gonadotrophic cycle.
20E and the Amino Acid (AA)/Target of Rapamycin pathway have been implicated in regulating vitellogenic events in female mosquitoes [10,11,12]. To test whether AAs and 20E affect CM gene expression in the female fat body, we used an in vitro tissue culture assay in which fat body tissue isolated from mosquitoes at 72h PE was incubated in the presence of AAs and/or 20E [11,13]. Incubation of the fat body in AA-containing medium elevated transcript abundance of PYK and GLY, while addition of 20E to this medium resulted in a further rise of their levels (S5A Fig). To investigate whether these regulatory factors were also involved in controlling CM metabolism during the PBM phase, female mosquitoes 72PE were injected with 20E and AAs. The expression of GLY and PYK were upregulated as a result of the simultaneous application of AAs and 20E; LDH was responsive to AAs but not 20E, while GYS to neither of these two regulators (S5B Fig). In agreement with in vitro experiments, AAs elevated LDH transcript abundance, but 20E had little effect. Thus, 20E and AAs play different roles in the regulation of CM genes. In in vivo experiments utilizing application of 20E and AAs, glycogen and glucose levels decreased when both AAs and 20E were given, mimicking the status of these sugars in PBM mosquitoes (S5C Fig).
20E is the principal hormone governing PBM reproductive events in female mosquitoes. EcR silencing has been reported in mosquitoes with EcR knockdown resulting in reduced ovarian follicular length [14], and egg numbers as compared to the controls (S6A Fig). Therefore, we investigated whether EcR plays a role in controlling CM. We silenced EcR using dsRNA to a common EcR region (iEcR) in female mosquitoes at 24h PE, blood fed them 4 days later, and analyzed transcript levels of CM genes at 36h PBM using qPCR (S6B Fig). Expression of TREA and TPP genes encoding enzymes of glycogen/sugar metabolism were transcriptionally suppressed at 36h PBM as a result of EcR silencing (S6C Fig). Representative glycolytic genes—HEX, PFK and PYK—were controlled by the EcR in a similar manner (Fig 7A and S6C Fig). GLY, PGM and GPI followed the same trend. In contrast, GYS and LDH were not affected by EcR RNAi silencing. This is in agreement with the lack of 20E effect on expression of these genes described above. This transcriptional alteration was also reflected in enzyme protein levels, although the effects were milder in the case of proteins (Fig 7B).
EcR dsRNA treatment resulted in an increase in the fat body glycogen 36h PBM as revealed by means of PAS staining (Fig 7C). An increase in circulating sugars was observed in EcR-depleted female mosquitoes; in particular, the levels of glucose and fructose were highly elevated, reflecting an inability of these mosquitoes to utilize sugars (Fig 7D). Both glucose and fructose levels increased by greater than 3-fold at 36h PBM as result of EcR knockdowns. EcR dsRNA treatment resulted in an increase in TAG levels and a drop in the ATP levels 36h PBM (Fig 7E and 7F). To examine whether EcR promotes CM, PBM, via altering the glycolytic flux, we measured the IM levels downstream of glycolysis. Consistent with their inability to catabolize sugar reserves, EcR-silenced mosquitoes showed accumulation of early intermediates of the glycolytic pathway—glucose-6-phosphate and fructose-6-phosphate (Fig 7G). There was also a considerable build-up of lactate (Fig 7G). The level of pyruvate declined slightly, while that of citrate decreased considerably in these mosquitoes. These results clearly pointed to the fact that EcR is a critical regulator of CM during the PBM phase of the gonadotrophic cycle in female mosquitoes.
PEPCK is an essential enzyme in maintaining glucose homeostasis and as such pays important role in response to stress and starvation [15,16,17]. The microarray and qPCR analysis has revealed that the level of the PEPCK gene transcript was high at the beginning of the PE phase, but it was dramatically reduced by 72h PE (Fig 1 and S7A Fig). In agreement with these data, the expression of the PEPCK gene was inhibited by the application of JH in vivo and activated by Met RNAi silencing, indicating negative regulation of this gene by the JH/Met in PE phase (S7B and S7C Fig). In contrast to most CM genes, activation of which reached maximum at 36h PBM, the PEPCK gene was highly upregulated immediately after a blood meal in Aedes females (Fig 1 and S7D Fig). Significantly, both in vivo and in vitro tissue culture experiments have shown that AAs play a key role in activating this gene expression (S7E and S7F Fig). However, these experiments have shown that 20E is not involved in regulation of this gene expression (S7E and S7F Fig). Furthermore, EcR RNAi silencing did not affect its transcript levels (S7G Fig).
The PPP consists of oxidative and non-oxidative branches [18]. In contrast to other CM pathways, the genes encoding PPP enzymes of both branches were transcriptionally active throughout the PE phase and downregulated during the PBM phase (Fig 1). In the PPP oxidative branch, glucose-6-phosphate is utilized for the synthesis of ribose-5-phosphate, with glucose-6-phosphate dehydrogenase being a rate-limiting enzyme [18]. In this respect, it was of particular interest that genes encoding glucose-6-phosphate dehydrogenase (G6PD) and ribose-5-phosphate isomerase A (RPIA) were sequentially activated during the PE phase (Fig 1). G6PD also reduces NADP+ to NADPH that is utilized in lipid biosynthesis [18]. In Aedes female mosquitoes, the expression of the gene encoding G6PD is regulated by Met (Fig 5A). G6PD RNAi depletion resulted in decreased TAG levels, suggesting that Met-dependent control of this enzyme contributes to fat metabolism in the mosquito fat body (Fig 8A).
Transketolase (TAL), which is the rate-limiting enzyme of the non-oxidative PPP branch [18], exhibited a higher expression level during the first 24h PE, while its expression was downregulated during the PBM phase (Fig 1). We used qPCR to examine the transcript abundance of genes encoding RPIA and TAL, representatives of the oxidative and non-oxidative PPP branches, respectively. This analysis confirmed the transcriptome data showing an elevation of transcript abundance of these two PPP genes at late PE phase and a decrease at 36h PBM (Fig 8B). RNAi depletion demonstrated that Met was an activator of expression of these genes, while EcR had no effect (Fig 8C). Moreover, in vitro fat body assay experiments confirmed the lack of 20E effect on TAL and RPIA expression (Fig 8D).
Throughout each gonadotrophic cycle, females of hematophagous mosquitoes undergo drastic physiological changes, shifting from nectar feeding and host seeking to blood utilization and rapid egg development. We show here that these changes are accompanied by CM reprogramming to support the dramatically different functional requirements of a reproducing female mosquito. To accommodate this reprogramming, the female mosquito fat body, which is the metabolic center, undergoes a particularly remarkable transformation. Our transcriptome and qPCR analyses have demonstrated that the expression of genes encoding CM enzymes in this tissue was synchronized with the two phases of the gonadotrophic cycle, responding to the varying energy requirements of the reproducing female mosquito. Protein levels of enzymes involved in glycolysis and glycogen/sugar metabolism exhibited periodic changes throughout the mosquito reproductive cycle that correlated with the transcript abundance of their respective genes. Levels of stored and circulating sugars revealed their periodic accumulation and depletion in response to changing energy requirements throughout. These sugar levels were concurrent with transcript levels of genes encoding glycogen/sugar metabolism enzymes. Metabolomics analysis provided further evidence that the CM dynamics were entirely different during the PE and PBM phases of the female mosquito gonadotrophic cycle. Moreover, these data corroborated with the existence of a link between levels of CM gene expression and IMs. In addition, our analysis has revealed that the timing and regulation of the PPP was different from other CM pathways, suggesting its pivotal role in metabolic homeostasis of the female mosquito. Overall, our analyses suggest that the temporal coordination of CM in female mosquitoes occurs to a large degree at the gene level.
We have demonstrated here that Met functions as a major regulatory switch that governs metabolic reprogramming during the PE phase of the female mosquito gonadotrophic cycle. Our data suggest that Met acts at the genomic level, affecting expression of CM genes, thus determining the PE CM reprogramming. The majority of genes involved in CM, including those of glycogen/sugar, glycolysis and the citrate cycle, were greatly elevated in Met-silenced female mosquitoes, while several genes encoding the PPP were downregulated. The JH receptor Met belongs to the bHLH-PAS family of heterodimeric transcription factors, proteins that respond to environmental or physiological signals and are involved in mediating multiple cell responses including metabolism and cancer [19,20]. The genomic action of Met has been established [21]—Met forms heterodimers with other bHLH-PAS factors in a JH-dependent manner and activates target genes via interaction with E-box motifs in their regulatory regions [22,23,24,25]. However, Met is also involved in the JH-mediated repression hierarchy. While the gene activation by Met appears to be direct, its repressive action requires intermediate factors [5]. In addition to a significant effect on CM enzymes at the transcript level, Met RNAi silencing caused elevation of glycolytic flux and depletion of stored and circulating sugars.
Ingestion of blood leads to dramatic events in a female mosquito. Our transcriptomics and metabolomics analyses have revealed that there is an immediate change in the CM status following blood feeding. The TREA transcript increase and the drop in the trehalose level at 6h PBM suggest an early onset of trehalose utilization for glycolysis. Likewise, there was a dramatic rise in the LDH transcript level as early as 3h PBM followed by a lactate spike. The citrate cycle IMs also exhibited early sharp increases at 6h PBM. This instant elevation of glycolysis to maintain high levels of glycolytic intermediates occurs prior to the rise of the 20E titer in the female mosquito, indicating that it is regulated by factors other than this hormone. Indeed, we found that the early PBM response is likely controlled by amino acids. In the in vitro fat body culture assay, the LDH gene transcript was elevated in response to amino acids, but downregulated by 20E. The role of the amino acid/TOR pathway in vitellogenic PBM events has been established [10]. The mosquito fat body serves as the nutrient sensor organ detecting signaling amino acids derived from a blood meal [1,26]. Here, we have uncovered the role of amino acids in regulating CM at the early PBM stage. PEPCK gene activation by AAs occurs at the onset of blood feeding, the time of the ingestion of a huge amount of food in a form of blood. This physiological state imposes an enormous energy requirement on a female mosquito that is needed for the rapid excretion of a large volume of fluid and digestion of a massive blood meal. It appears that a high elevation of the PEPCK gene expression that correlates with these events is essential for maintaining circulating sugar homeostasis and represents an adaptation of mosquito CM to hematophagy.
The PBM stage is the apex of the gonadotrophic cycle, when a female mosquito utilizes a huge blood meal and rapidly develops over a hundred eggs just within 48h. We show here that there is a stunningly high level of the CM activity during the middle of the PBM stage, particularly of glycolysis. The genes encoding the rate-limiting glycogen and glycolytic enzymes, such as GLY, PFK and PYK, were upregulated 40 to over 100-fold by 36h PBM. Apart from providing substrate for energy production, the major function of aerobic glycolysis is to maintain high levels of glycolytic intermediates to support anabolic reactions in rapidly dividing cells [6]. Our analysis of CM IMs showed that the glycolytic flux was extremely elevated at the PBM stage.
20E is the major hormone controlling events of the PBM stage of the female mosquito reproductive cycle, and its action is mediated by the heterodimer of EcR and the insect RXR homologue Ultraspiracle, both of which are members of the nuclear receptor superfamily [4]. Nuclear receptors are a specialized family of ligand-bound or unliganded transcription factors that play central roles in regulating development, growth and metabolism [27]. In female mosquitoes, the 20E regulatory hierarchy is responsible for the YPP gene expression in the fat body [28,29,30]. Our results further suggest that control of CM occurs mainly at the gene level, and EcR is an important regulator of these genes during dramatic increases in CM. In rapidly developing Drosophila melanogaster larvae, CM is temporally coordinated by the estrogen-related receptor (ERR) [8]. This nuclear receptor alters the expression of genes encoding metabolic pathway enzymes, thus playing the role of a metabolic switch. Whether ERR plays a similar role in the female mosquito and its mode of interaction with EcR in synchronizing CM during the PBM stage requires further study.
In summary, we have presented a comprehensive analysis of CM dynamics in the female mosquito during the reproductive cycle. We show that such metabolism is tightly correlated with the rapidly changing physiological conditions of this organism. Our transcriptomics and metabolomics studies have revealed the association of expression of genes encoding CM pathways and IMs. Our analyses have identified that Met is the key regulatory switch responsible for temporal coordination of CM during the PE phase of the female mosquito gonadotrophic cycle. We also show that 20E/EcR and amino acids play different roles in CM regulation. Further molecular analysis of these metabolic regulatory pathways may lead to the implementation of metabolism-based methods to prevent mosquito-borne disease transmission.
The mosquito A. aegypti Rockefeller strain was raised as described previously [11,31]. Adult mosquitoes were fed water and 10% sucrose solution continuously. All procedures for vertebrate animal use were approved by the Institute of Zoology Animal Care and Use Committee.
Sample sets from 12 independent mosquito populations were analyzed for every experimental condition. Six mosquitoes per sample point were washed in PBS buffer, frozen in liquid nitrogen, grounded in 400 μl pre-cooled 90% MeOH and then incubated for 1 h at -20°C [32]. Following centrifugation and debris removal, a second extraction step with 60% MeOH was performed. The supernatant was vacuum dried for 1 h and incubated with 40 μl O-methoxylamine hydrochloride (20 mg/ml saturated in pyridine) for 1 h at 37°C. Then, 50 μl MSTFA reagent was added to the samples, which were then incubated for 30 min at 37°C, with shaking, and finally diluted with 400 μl n-hexane and transferred to the auto sampler vials for the next step. GC-MS analysis was performed following a standard protocol using Agilent 7890 GC coupled with a 5975N series mass selective detector (MSD). The following temperature steps were used: initial temperature of 75°C for 1 min, 5°C /min ramp to 250°C for 5 min, 5°C/min ramp to 320°C for 3 min. A 1-μl sample was injected in split-less mode at 250°C with helium carrier gas flow set at 1 ml/min. A HP-5MS column with a 5-m-long guard column was used for the analysis. Chromatogram acquisition, peak de-convolution and library searches were performed using Agilent MSD Chemstation software. Metabolites were identified using authentic chemical standards analyzed on the same system.
Glycogen assays were performed as described previously [33,34]. SpectraMax Plus384 was used for detection. Six independent biological samples, with six adult female mosquitoes per sample, were used for each experimental condition. Following centrifugation, samples were transferred into 96-well plates, incubated with Free Glycerol Reagent (Sigma), and assayed using SpectraMax Plus384. For TAG measurements, six mosquitoes were homogenized in 100 μl PBST containing 0.5% Tween-20 and incubated at 70°C for 5 min. Then, the samples were incubated with Triglyceride Reagent (Sigma) and assayed colorimetrically. For ATP measurements, six mosquitoes were homogenized in extraction buffer (6 M guanidine-HCL, 100 mM Tris, 4 mM EDTA) and boiled for 5 min. After centrifugation, the supernatant was filtered via PTFE membrane for HPLC assays performed using an Agilent 1100 HPLC coupled with DVD detector, following a published protocol [35]. On the chromatogram, ATP peaks were identified by utilizing retention time of standards (Molecular Probes, 911734). Total glycogen, TAG, and ATP concentrations were normalized to endogenous protein level of the samples, determined using Bradford assays (BioRad, 500–0201).
For histochemical analysis of fat body glycogen content, staining and visualization were performed as previously described [36]. The abdomen was separated from the rest of the body and fixed in 4% paraformaldehyde at 4°C overnight. Each sample was then dehydrated with increasing concentrations of ethanol, embedded in paraffin, and sectioned into 3- to 5-μm slices. Abdominal fragments were stained according to the periodic acid Schiff (PAS) method (Sigma, 395B) and observed under a Nikon Ni-E microscope.
dsRNA synthesis was performed as previously described [5]. The bacterial luciferase gene was used to generate control iLuc dsRNA. A Nanoliter 2000 injector (World Precision Instrument) was used to introduce corresponding dsRNA into the thorax of cold-anesthetized mosquito females 24h PE. The specificity of gene knockdown was characterized by a 50–70% decrease in transcript abundance of target genes (S3B and S4A Figs). All primers used for making dsRNA are listed in S5 Table. To test the effect of JH, 0.5 μl of JH (10 μg/ml JH in acetone as solvent) or acetone was topically applied to newly eclosed female mosquito abdomens. The females were examined 20h post treatment as previously described [24]. For metabolite measurements, samples were collected 20h post JH treatment. To test the effect of 20E, 0.5 μl 10−6 M 20E was injected along with amino acids into 72h PE female mosquitoes. Mosquitoes were examined 20h post treatment. Experiments were performed in triplicates under the same condition.
Total RNA samples were prepared under three different conditions, and fat bodies were dissected from abdomens of 10–15 individual mosquitoes. qPCR reaction was performed on the MX3000P system (Stratagene, CA) using SYBR green PCR Master Mix (Tiangen, Beijing). Thermal cycling conditions were: 94°C, 5 s; 59°C, 20 s; and 72°C, 20 s. Quantitative measurements were performed in triplicate and normalized to the internal control S7 ribosomal protein mRNA for each sample. Primers used for qPCR are listed in S5 Table.
Eight mosquito fat bodies were homogenized in 100 μl of breaking buffer by pellet pestle, as described previously [37]. Aliquots of whole mosquito protein samples were resolved on 4–15% gradient SDS-polyacrylamide gels (Bio-Rad) and transferred to PVDF membranes (Invitrogen). After blocking, the membranes were incubated overnight with the primary antibody at 4°C (S3 Table). As loading control, an antibody against β-actin (Sigma) was used.
DEG datasets from PE and PBM time course microarray were utilized to reconstruct the expression profiles of the genes involved in metabolism. Complete linkage hierarchical clustering was performed using the hclust function in R [5]. Discrete clusters were obtained by cutting the resulting dendrogram with the cutree function using a visually determined height value. Orthologous groups and pathway information, based on the Kyoto Encyclopedia of Genes and Genomes (KEGG), were downloaded from the database [38] and used in this study. An enrichment analysis was used to detect the significance of alteration of each metabolic pathway, and p values were calculated based on hyper-geometric tests, as described previously [39]. In all other experiments, statistical significance was defined by a p value < 0.01, as evaluated using paired-end, two-tailed, student's t-tests (Graphpad 5.0). Comparisons were made between time points/ treatments and the controls and significant differences were indicated in the graphs. All quantitative data are reported as mean ± SD.
In vitro fat body culture experiments were performed as previously described [11,26]. Female mosquito abdominal walls with adhered fat body tissue were incubated in a culture medium under various conditions. In the culture medium lacking amino acids, an equal molar amount of mannitol was supplemented to compensate for changes in osmotic pressure [26]. 20E was added to the culture medium supplemented with a complete set of amino acids [26]. To mimic a natural rise in the 20E titer, the tissue was first incubated with 5 x 10−8 M of this hormone for 4h and then in the presence of 10−6 M for 4 h. Total RNA was then isolated and transcript abundance was analyzed using qPCR. The experiment was repeated three times under the same conditions.
|
10.1371/journal.pbio.2004470 | Full mutational mapping of titratable residues helps to identify proton-sensors involved in the control of channel gating in the Gloeobacter violaceus pentameric ligand-gated ion channel | The Gloeobacter violaceus ligand-gated ion channel (GLIC) has been extensively studied by X-ray crystallography and other biophysical techniques. This provided key insights into the general gating mechanism of pentameric ligand-gated ion channel (pLGIC) signal transduction. However, the GLIC is activated by lowering the pH and the location of its putative proton activation site(s) still remain(s) unknown. To this end, every Asp, Glu, and His residue was mutated individually or in combination and investigated by electrophysiology. In addition to the mutational analysis, key mutations were structurally resolved to address whether particular residues contribute to proton sensing, or alternatively to GLIC-gating, independently of the side chain protonation. The data show that multiple residues located below the orthosteric site, notably E26, D32, E35, and D122 in the lower part of the extracellular domain (ECD), along with E222, H235, E243, and H277 in the transmembrane domain (TMD), alter GLIC activation. D122 and H235 were found to also alter GLIC expression. E35 is identified as a key proton-sensing residue, whereby neutralization of its side chain carboxylate stabilizes the active state. Thus, proton activation occurs allosterically to the orthosteric site, at the level of multiple loci with a key contribution of the coupling interface between the ECD and TMD.
| Pentameric ligand-gated ion channels are an important class of receptors that are involved in many neurological diseases. They have been extensively studied but a full understanding of their mechanism of action has yet to be achieved. In an effort to bypass obstacles in the research of human receptors, bacterial versions have been used to characterize the family’s structure-function relationship. One key bacterial receptor, known as GLIC, has lead the way in structural resolution of various mechanistic states along the gating pathway, yet its activation by protons is significantly less understood than its human counterparts. To define the site(s) involved in proton gating, we systematically mutated all titratable residues near the pH50 of activation: Asp, Glu, and His. We determined that a previously established His residue in the transmembrane domain is structurally important but likely plays little or no role in proton gating. We instead found that proton activation is a complex multiple loci mechanism, with the key contribution stemming from the coupling interface between the extracellular and transmembrane domain, with E35 acting as a key proton-sensing residue.
| Pentameric ligand-gated ion channels (pLGICs) are key players of neuronal communication. They promote either cell depolarization or hyperpolarization with the passive permeation of ions through an intrinsic channel, whose opening is stabilized by the binding of specific neurotransmitters. pLGICs are ubiquitously expressed in virtually all neurons, contribute to central nervous system functions, including sensory and motor processing, central autonomous control, memory and attention, sleep and wakefulness, reward, pain, anxiety, emotions, and cognition [1]. As such, they are important drug targets. At least one member of each major subfamily of vertebrate pLGICs has been structurally resolved in the recent years: a serotonergic receptor (5-HT3A, [2]) and a nicotinic acetylcholine receptor ([nAChR] α4β2-nAChR, [3]) from the cationic receptors, and a GABAergic receptor (β3-GABAA, [4]) and two glycinergic receptors ([GlyRs] α1- and α3-GlyRs, [5,6]) from the anionic receptors. However, relating these 3D structures to the physiologically relevant allosteric states that mediate pLGIC activation and desensitization remains an open and debated question [1].
At present, the best structurally characterized pLGIC is the G. violaceus ligand-gated ion channel (GLIC), of prokaryotic origin. Its structure shows a highly conserved fold with the subsequently resolved structures in the pLGIC family. Each subunit consists of an extracellular domain (ECD), predominantly in a β-sandwich fold, and a transmembrane domain (TMD) composed of four helices, labeled M1–M4. The available structures of pLGICs support that the GLIC globally shares a common gating mechanism with its eukaryotic cousins, although molecular details differ [1].
Due to the relative ease for overexpression in Escherichia coli, as well as its biochemical robustness toward detergent-solubilization and mutations, the GLIC has been resolved in four distinct conformations. It is the first receptor to be resolved in an apparently open channel confirmation, as well as three various closed channel conformations [7–9]. In addition, the GLIC has been solved as a complex with a variety of allosteric modulators, such as barbiturates, bromoform, lidocaine, propofol, and xenon [10–14]. Membrane-inserted GLIC proteins have been studied by electron paramagnetic resonance spectroscopy and fluorescence-quenching experiments following site-directed labeling, relating some local conformational changes to the activation and desensitization transitions monitored by electrophysiology [8,15–18]. The GLIC has also been studied by computational methods including molecular dynamic simulations [19–23]. This combined set of data support that, upon activation, the subunits’ ECD regions move closer together, which precedes a concerted tilt of the pore-lining M2-α-helices to open the pore gate and activate the receptor.
Chimeras between the ECD of the GLIC and the TMD of the α1-GlyR were additionally shown to fold properly and to be functional, revealing compatibility between prokaryotic and eukaryotic domains [24,25]. However, the GLIC is a pH-gated channel, activated by lowering the pH, with a maximal activation at pH 4 and a pH50 around 5 [26]. This sharply contrasts with most eukaryotic pLGICs, which are activated by a neurotransmitter binding to a well-described cavity in an intersubunit interface of the ECD. Mammalian pLGICs are not directly activated upon pH changes, but the agonist-elicited responses are modulated by pH, notably for the α3β4-, α3β2-, and α4β2-nAChRs [27], α1-GlyR [28,29], and various GABAA receptors [30,31]. So far, a handful of invertebrate pLGICs were found to be directly activated by pH, a nAChR from Caenorhabditis elegans by low pH [32], and two insect GABAA receptors by high pH [33,34].
Fully understanding the molecular mechanism of GLIC signal transduction thus requires identifying the locus where protons act to activate the channel and which titratable groups are crucial in this process. The observation that a chimera composed of the ECD of the GLIC fused to the TMD of the α1-GlyR (GLICECD-GlyRTMD known as Lily [25]) or the Erwinia chrysanthemi ligand-gated ion channel (GLICECD-ELICTMD [35,36]) is activated by protons indicates that the major proton-sensing motifs are likely situated in the ECD. However, an inverse chimera, the ELICECD-GLICTMD, yields pH-gated currents when the gain-of-function mutation of I9ʹA is incorporated, suggesting either proton sensors too weak to activate in the ELIC ECD or their presence in the GLIC TMD [35,37]. Mutation of a His residue located in the middle of the TMD (His235) abolishes GLIC function, raising the possibility that it might be involved in pH sensing [38,39], but combination of this mutation with a gain of function mutation restores at least part of the pH-gated function [8,37]. Altogether, the location of the proton activation site(s) remain(s) essentially unknown thus far.
The present study provides an exhaustive mutational analysis of GLIC titratable residues, with pKa’s in the pH 6 to pH 4 range (namely Asp, Glu, and His), given the pH50 of GLIC activation, to search for the proton-sensing site(s). Each GLIC subunit contains 19 Asp, 16 Glu, and 3 His residues that were individually and/or collectively mutated. Mutation of an Asp/Glu/His residue may be expected to affect the GLIC function in two ways: (1) by altering direct proton sensing when the protonated form of this residue stabilizes the active state, as compared to the nonprotonated form, or (2) by altering the gating equilibrium independently of side chain protonation. In an effort to discriminate between these two possibilities, Asp/Glu residues were systematically mutated to Asn/Gln, replacing the carboxylate moiety by a nontitratable amide group, thereby tentatively mimicking a permanently protonated form, as well as to Ala, removing the titratable moiety. Selected mutants were also solved by X-ray crystallography to characterize the effect of the mutations on the local protein structure.
To identify the determinants of proton-modulation/gating, two-electrode voltage clamp electrophysiology using expression in Xenopus laevis oocytes was employed; activation was elicited by dropping the pH from neutral (pH 7.3–8) to lower values (minimum pH 3.7). All recordings showed a slow onset of the response, with no apparent desensitization during a 30–90 s pH application. Distinguishing the kinetics of activation and desensitization as compared to the wild-type (Wt) GLIC, proved difficult for the vast majority of mutants, therefore mutants were characterized on the basis of their pH50, defined as the pH eliciting half of the maximal current.
To minimize the influence of the intrinsic variability between oocyte batches, which show some variation in the Wt response to pH changes, mutants were also characterized by a ΔpH50, which corresponds to the variation of pH50 between each cell expressing a mutant, and the Wt cell(s) in the same batch of oocytes. Significance of the results in this report was determined as values larger than 0.5 pH units for the mean ΔpH50, and mean pH50, as compared to Wt, based on the standard deviation of 83 measurements of the Wt (with a standard deviation of 0.4 pH units). All values that fulfilled this criterion of significance also had a p value < 0.01 in the Student t test against the Wt. A pH50 could not be established for some mutants, often with minimal currents.
For clarity, the presentation of the results is organized according to five regions of a GLIC subunit: the apical top of the ECD, the Loop B and C on the principal (+) face of the interface, the basal ECD principal (+) and complementary (−) faces, and finally the TMD (Fig 1).
In the apical region, the residues either face the solvent (D13, E14, D55, E67, E69, D97), are located at the subunit interface (D49, E75, D136), or are located at the middle of the ECD β-sandwich (E104). Their mutation produces weak effects. D13N-E14Q, D49N-D55N, E67Q-E69Q, and E67Q-E75Q produced no significant change in pH50, and neither did the corresponding single mutants of D55N, E67Q, E67A, E69Q, and E75Q. The mutants D49N, E75A, D97N, D136N, and D136A display small increases in pH50, but do not meet the criterion for a significant effect. The effect of D49 replacement is corroborated in a recent study, whereas the same study showed D136A to have a decreased pH50 with an increased Hill-slope [36].
Finally, E104, which makes internal intrasubunit contacts, showed no significant change when mutated to Q. Mutation to A, meanwhile, produced a significant ΔpH50 and a noticeable, yet nonsignificant, decrease in pH50 (Fig 2).
The β9-β10-loop (Loop C) was investigated in conjunction with R133 (Loop B), both of which are located on the principal (+) intersubunit interface, a region that contributes to neurotransmitter binding in eukaryotic pLGICs. The mutation D185N at the base of Loop C did not have a significant effect (Fig 3). This position was also recently reported with Wt-like properties as both Asn and Ala mutations [36]. Removing all remaining titratable moieties in the sextuple mutant R133A-E177Q-D178N-R179Q-E181Q-K183Q produced a marked increase in pH50. In contrast, mutating only the three acidic residues (E177Q-D178N-E181Q) produced a marked decrease in pH50, whereas the individual Asp/Glu residue mutations show Wt-like responses with E181Q having a tendency to decrease the pH50. The mutations E177A and E181A were previously reported in a cinnamic acid study with pH50 values of 5.4 ± 0.1 and 5.6 ± 0.1, respectively, as compared to a Wt value of 5.2 ± 0.1 [40]. D178 was also recently reported as displaying a Wt phenotype for both the Asn and Ala mutations [36]. Evaluation of the basic residues shows R179 mutation increases the pH50, with R179Q producing a marked increase, whereas R179A only has a tendency to increase the pH50, showing that the removal of the side chain guanidinium is not solely responsible for the phenotype at this position. Meanwhile, replacing the remaining basic residue of Loop C, K183Q, has no effect. The single mutation R133A on Loop B also has no effect, which was also previously shown in [40]. Despite containing the triple mutation that results in a decrease in pH50, the mutant that removes all titratable residues from Loop C along with R133A has a similar but stronger change in pH50 as compared to the single R179Q mutant, yet their ΔpH50s vary by 0.5 units. This suggests that the main player in this region is R179, along with a complex network of other side chain interactions modulating GLIC activation at this level.
The complementary (−) face of the ECD contains eight Asp/Glu residues at the intersubunit interface and two solvent-exposed residues (D161 and E163) at the bottom of β9 near the pre-M1-π-helical loop [41]. The double mutants of pairs of proximal residues were initially tested: D86N-D88N, D145N-E147Q, D153N-D154N, and D161N-E163Q. The mutant of the vestibular pair on the β5 strand, D86N-D88N, yields a significant decrease in pH50 greater than 1 pH unit. The solvent-facing pair D161N-E163Q, and the pair near the bottom of the β8-β9-loop (Loop F) D153N-D154N, tend to both decrease the pH50, but below the threshold of significance. The pair D145N-E147Q, on Loop F, has no effect.
All residues were also tested individually, with most mutations having no significant effect, although a large majority tend to decrease the pH50 (notably D86N/A, D88N/A, and D91N). Although D91A was found to insignificantly increase the pH50 as compared to Wt, D91N and D91A were recently reported to have a slight decrease in pH50 values [36]. In contrast to the nonsignificant individual mutations, E26, which is found interacting with the bottom of the β4-β5 loop of the principal (+) face, shows a robust decrease of the pH50 when mutated to both Q and A.
A combination of E26Q with mutations of vestibular-facing residues that tend to decrease the pH50 produces an obvious decrease in pH50. The E26Q-D86N-D88N-D91N mutant exhibits strongly reduced maximal currents, which prevented reliable evaluation of the pH50 in most of the cells (Figs 1B and 4A). Of the 11 oocytes tested (over four injections), only three recordings had enough current to properly evaluate (S1 Table).
It is striking that among the 23 mutants tested in this region, most show a propensity to decrease the pH50 (Fig 4A).
The basal principal (+) face, below Loop C, contains E82 and H127, along with a cluster of Asp/Glu residues found close to the TMD: D31, D32, and E35 from Loop 2, as well as D115 and D122 from Loop 7.
Consistent with some of the previous studies of H127 [38], H127N and Q both produce Wt-like currents (Fig 4). A structure was resolved for H127N, which resulted in no appreciable difference from the open form of the GLIC (protein data bank identification codes [PDB IDs]: 4HFI/3EAM, S1 Fig).
The effects of D32 and D122, both of which are engaged in salt bridges with R192, have been previously studied in other pLGICs [42], as well as in the GLIC [22,43]. D32A and D122A were shown to be nonfunctional yet weakly expressed [43], whereas D32N shows marked decrease in pH50 and in maximal currents (Fig 4B, [22]). D122N showed no function, and subsequent expression studies showed no expression (Figs 4B and 5). D31, which points towards the vestibule, has no effect when mutated to D31N in this study, whereas a slight loss of function, insignificant with the criterion and to the Wt value of this study, was observed with both Ala and Asn mutations by Alqazzaz et al. [36].
Mutants of the solvent-exposed D115, further away from this triad on Loop 7, exhibit Wt-like properties. Yet, the D115N and D115A mutations show opposite phenotypes, with Asn having a tendency to increase the pH50 and Ala to decrease it. Of the mutants which had both Asn/Gln and Ala mutations tested, only the pair of E82Q and E82A follows the same pattern (Fig 4B).
Finally, mutation of E35 produced the strongest effect; therefore, a more extensive study of this position was performed. The mutations E35Q, E35A, and E35M (the GlyR-position equivalent) display a significant increase in pH50, whereas E35K shows a weaker nonsignificant effect, and E35H appears to show the inverse phenotype with a marked decrease in pH50. Interestingly, fitting E35H data with a single sigmoidal curve yielded poor fits, preventing reliable calculation of the pH50. Of the 13 recorded oocytes (over four injections), only two recordings could be fit properly (Figs 1B, 1C and 4B, S1 Table). The poor fit of E35H indicates a more complex mechanism at play, which could possibly be elucidated with a more in-depth study of mutation of this position. Altogether, these data show that the loss of charge at the E35 position is important to the gain-of-function effect found.
Three hydroxyl-containing residues that are situated near E35 and E82 on the complementary face (−) of the ECD were subsequently mutated: Y28, S29, and T158. Individually removing each hydroxyl group shows a tendency to increase the pH50, with Y28F (near E82) having a significant increase. The mutation of T158A did not have the same marked effect, although T158 is the residue positioned closest to E35, indicating that the global environmental change in charge at position 35 is more important than the direct residue–residue interactions.
Subsequently, the additivity of the Asp/Glu mutations was also tested; however, neither E82Q-D115N, E35Q-E82Q, nor E35Q-E82Q-D115N showed increased proton sensitivity as compared to E35Q. A combination of all other weak pH50-increasing mutations was performed with the addition of R179Q or E75Q-D97N-D136N to the triple mutant E35Q-E82Q-D115N, with neither of these mutants producing any shift greater than E35Q alone (Fig 4B).
The pH50 evaluation of this region effectively identified key residues involved in pH-sensing, but the results of the combined mutants demonstrate the complexity of the mechanism involved in GLIC-gating.
A structural analysis was performed on E26Q, E26A, E35Q, E35A, E67A, E75A, E82Q, E82A, D86A, D88N, D88A, H127N, E181A, and H277Q mutants, by solving their structure at pH 4. The respective PDB IDs and crystallographic statistics are listed in S2 Table. All structures were in the apparently open conformation. Each residue’s root-mean-squared deviation (RMSD) and Cα RMSD was evaluated in relation to the intrinsic variability between subunits and across the two Wt structures (3EAM and 4HFI) at pH 4.6 and pH 4, respectively, as a means to control for crystal variability. There are no significant differences, measured as 5-fold over Wt, seen in the backbone Cα residues of any of the resolved structures, with the exception of E35A and E82Q (Fig 6 and S1 Fig). The E82Q side chain takes on a different rotamer, causing a local change in the backbone, as well as a cascading rotamer/conformational change in Y28 and F156, the latter of which flips out towards the aqueous environment. Interestingly, E35A also produces the greatest structural deviations around Y28 along with neighboring residues. Of the few conformational changes seen, it is interesting to note that Y28F had the greatest functional effect of the nontitratable mutations tested.
It has been proposed that H235 is a key residue mediating proton sensing [38,39], but recent data show that the GLIC can still gate when H235 is mutated to nontitratable residues in combination with a strong gain-of-function mutation [8,37]. It has also previously been shown that GLIC expression in bacteria is highly sensitive to mutation at H235 [7]. Both of these findings are confirmed, firstly with the mutation H235A, which abolishes expression in oocytes (Fig 5), indicating a structural importance of the residue, and secondly, with the mutation H235Q, which does actually produce functional receptors, albeit with a strong decrease in pH50 and reduced maximal currents (Fig 7, S1 Table). H235Q does not allow for a charge or a change in protonation state at this position, and therefore one would expect a completely nonfunctional receptor if this residue were solely responsible for the proton-modulated gating of the GLIC. Rather, this finding confirms the structural importance of H235 that has been previously reported, along with the importance of a hydrogen-bonding network, between the M2 α-helix and the neighboring M3 α-helix, with the backbone carbonyl of I262 [7,38].
Among the other titratable residues of the TMD, neither a mutation of E272Q nor E282Q produced an overall significant effect, albeit a significant negative ΔpH50 was observed for E272Q, which was previously reported to be nonfunctional when mutated to A [44]. In contrast, the mutation of either E222 or E243 as Q or A, as well as H277Q, all produced significant loss-of-function effects, pointing to the key role of these residues in activation. E222 (E-2ʹ), at the beginning of the M2 α-helix, has been extensively studied as the key component of the selectivity filter, a charged constriction point in the lower part of the pore, for cationic pLGICs [45,46]. The E222A mutation has previously been resolved crystallographically and has no change in structure as compared to Wt at pH 4 [10]. Therefore, the strong loss-of-function mutation does not appear to affect conformation. E243 flanks the apical section of the TMD pore, making a pentameric ring at the end of the M2 α-helix. A mutation of E243C was shown to have a similar pH50 as Wt [18], whereas the mutation E243P is reported as nonfunctional and found in a locally closed conformation [7]. Mutation of H277, which lies in the adjacent M3 α-helix (nearby E222) with possible electrostatic interaction, has also previously been shown, using noncanonical amino acid substitution to decrease the pH50 [38].
The structure of H277Q was performed and was also found to have no apparent deviation from the apparently open pH 4 conformation, which further corroborates the several previously published works that show H277 does not appear to play a role in proton gating (S1 Fig).
All of the titratable Asp/Glu/His residues of the GLIC were mutated and evaluated for their impact on the receptor sensitivity to protons. This approach allows for the identification of key regions controlling gating.
Several mutations within Loop C markedly alter the pH50. The decrease in proton sensitivity following neutralization of the E177/D178/E181 cluster, which is completely negated and reversed with the neutralization of the basic R133, K183, and R179 residues, as well as the importance of R179 mutations alone, suggests a role in gating rather than in proton sensing. This idea is in line with the observation that replacement of the entire Loop C with other pLGIC Loop C sequences or a polyglycine segment is still compatible with a pH-gated channel [47]. Interestingly, this region was suggested to be the binding site for organic acids that act as negative allosteric modulators, including caffeic [40] and crotonic [48] acids. These data thus further document the key allosteric role of this region, which is homologous to the neurotransmitter binding site of eukaryotic receptors.
Mutants of E35 are unique in strongly increasing the proton sensitivity. Neutralization of the E35 side chain by mutation to A, Q, or M produce a similar increase in proton sensitivity, whereas the mutation of E35K exhibits the same behavior with half the increase. The E35H mutation, however, shows the inverse phenotype. This suggests that a charge at this level impairs activation, and a hydrophobic/polar side chain is preferred for stabilization of the active state. Interestingly, the structures of the Wt GLIC, E35Q, and E35A at pH 4 show that the side chain of residue 35 is located in a hydrophobic environment bordered by the side chains of P113, L114, F116, F156, T158, and P247. In the GLIC structure at pH 7, which is representative of the resting state, the carboxyl moiety of E35 is more solvent exposed toward the inner vestibule because the side chain of P247 is moved away following the key revolving motion of the M2–M3 loop (S2 Fig). Therefore, the resting state would be predicted to better accommodate a charged residue, which is consistent with the functional data. It is noteworthy that E35K, while charged, does not decrease the proton sensitivity. However, the length of the lysine side chain likely places the charged ammonium away from the E35 carboxylate locus. Altogether, the data support that E35 is a bona fide proton sensor, whose charged form stabilizes the resting conformation, whereupon protonation favors the active state. The functional proton gating of E35Q and E35A mutants additionally suggests that if E35 is a bona fide proton sensor, it can’t be the only one in the GLIC.
In the lower part of the ECD, the principal (+) face also carries two acidic residues, E82 and D115. E82Q and D115N display significantly higher proton sensitivity than E82A and D115A, respectively. Assuming that Gln and Asn faithfully mimic protonated side chains of Glu and Asp, this would suggest that residues E82 and D115 are also proton sensors. In this case, their protonated side chain would elicit stabilizing interactions specifically in the active state. To investigate this idea, the structures of E82Q and E82A were solved at pH 4. The E82A structure is similar to that of the Wt pH 4 structure, whereas that of E82Q shows a marked local reorganization of the Q82 moiety. Assuming E82 to be in a protonated state at pH 4, Gln is not a good mimic of a protonated Glu residue because Q82 is engaged in a completely different set of interactions with neighboring residues. Therefore, the E82Q mutant phenotype is likely due to local side chain reorganizations of Y28 and F156 rather than being a proton sensor itself.
Finally, D32 mutations significantly decreased the pH50. However, this residue is shown in the pH 4 structure to interact through a salt bridge with the already salt-bridged pair of R192 and D122. This interaction interlocking Loop 2, Loop 7, and the pre-M1 loop is often conserved throughout the pLGIC family [23], with notable exception of the ELIC and the α-glutamate-gated chloride channel from C. elegans. Thus, it is unlikely that it undergoes side chain protonation at pH 4, and likely contributes in gating rather than in proton sensing [22].
Mutation of Asp/Glu residues at the complementary (−) face, in general, decreased the proton sensitivity of the GLIC, with additive effects in many cases, especially for D88 and D86. However, the pattern of phenotypes does not allow discriminating whether those residues are involved in proton sensing or gating, because mutations to Asn/Gln also produced decreased proton sensitivity. The key mutants at E26, D86, and D88 were solved at pH 4, each showing a structure similar to that of the Wt (S1 Fig). E26 clearly is an important residue in GLIC function, but other techniques are required to elucidate its mechanistic role.
Mutations at both entrances of the ion channel, E243 (outer ring) and E222/H277 (inner rings), as well as at the middle of the TMD within each subunit four-helix bundle (H235), produced significant decreases in proton sensitivity, but again here, the pattern of phenotypes does not allow discriminating between proton sensing and gating. H235 was previously proposed to be the major proton-sensing residue for GLIC activation, but the mutation H235Q, which is no longer titratable, still produces a functional proton-sensitive receptor, questioning its role as the sole proton sensor. Clearly, H235 is a very important residue for functionality, although its protonated state would be hardly accommodated by the hydrophobic environment of the TMD in both known apparently open and closed forms.
Molecular dynamics simulations using a string method optimization made predictions of the change in protonated state and sensitivity of titratable residues throughout the gating mechanism of the GLIC [23]. The report identified E26 and E177 to have loss-of-function effects when mutated. E26 indeed results in a loss-of-function effect, whereas E177 seems to have no effect. Additionally, the prediction identified E35, E75, and E243 as gain-of-function locations if mutated, whereas E75 and E243, when mutated, have either no effect or the opposite effect. Therefore, these results suggest that molecular dynamics simulations can indeed help identify potentially interesting residues, as the most potent residues were identified, but that the predictions from the simulations still need further experimental testing as the actual functional results do not all agree with predictions of the molecular dynamics data.
Despite combined mutation attempts, channels displaying constitutive activity were not observed, nor was there a complete abatement of function. This shows that proton sensing is not mediated by a single Asp/Glu/His residue, but rather by several residues located on different parts of the protein. Individual residues in many places may contribute partially to proton-sensitive channel gating, unraveling a mode of proton-controlled activation quite different from that of classical agonists of pLGIC family receptors, which act at a well-characterized orthosteric site.
The results indicated that E35 is an important proton sensor, as well as the existence of a number of other proton-sensing candidates, notably E26 and the D86/D88 pair in the ECD, E243 at the top of the channel, and E222/H277 at the bottom of the channel. E35, E26, and E243 are all located nearby the ECD-TMD interface, suggesting that at least part, if not all, of the pH-elicited activation of the GLIC bypasses the orthosteric site. E222/H277 might also contribute to pH sensing, implying in this case the diffusion of protons from the extracellular compartment, possibly through the ion channel itself. The E222/H277 pair most likely confers the intracellular proton concentration sensitivity of the GLIC, which was reported from inside-out patch clamp experiments [49].
Complementary approaches that would be too cumbersome to perform on all the mutants evaluated in this study, such as reporting on the open probability, channel kinetics, and linking the efficiency of gating to the apparent affinities identified, are necessary to elucidate the specific effects of the key residues in the complex gating mechanism of the GLIC.
The recent studies employing chimeras between ELIC domains and GLIC domains validate the possibility of multiple proton-sensing sites. Interestingly, the pH50 of Lily, the GLICECD-ELICTMD mimicking Lily (GELIC), and E35Q/A mutants all closely hover around pH 6.5 [25,36]. Altering the TMD residues surrounding E35 could have the same effect as charge neutralization. However, another GLICECD-ELICTMD mutant (GE), which lacks the mutation Y119F and the C-terminal conversion of RGITLELIC to LFFGFGLIC, has a reported pH50 of 3.63 [35]. It is important to note that neither the study of Lily nor of GELIC tested below a pH of 5, and both found significantly reduced currents, whereas GE displayed robust currents at pH’s below 5. These discrepancies point to the intricate interactions between the ECD and TMD, which influence gating and possibly proton sensitivity. These studies combined with the current results also clearly show that the principal component of proton activation lies within the ECD and not the TMD, as previously assumed. The ELICECD-GLICTMD chimera is not susceptible to proton activation until further mutation by I9ʹA, which has previously been shown to destabilize the recovery time for the receptor to return to the resting state [26].
Residues other than Asp/Glu/His may also contribute to proton sensing, notably Arg/Lys side chains that can display pKa’s in the 4–6 range when located in very hydrophobic environments [50], or aromatic residues through cation-pi interaction with hydronium ions [51].
Among pH-gated ion channels, two were studied in molecular details. First, the bacterial potassium channel KcsA was found to be inhibited by a local network of ionic/H-bond interactions between two Glu, two Arg, and a single His residue. A disruption of this network upon protonation allows channel opening [52]. In this case, E to A mutation of the two key residues increased the sensitivity of the channel to protons. Such a “suppression of charge to activate the channel” mechanism on the GLIC is observed for the single E35 residue.
The pattern of phenotypes observed here is reminiscent of acid-sensing ion channels (ASICs). Indeed, ASICs have been extensively studied, but efforts to map the sites for proton binding have so far yielded inconclusive results because mutation of multiple individual Asp and Glu residues independently produces changes in proton sensitivity [53]. Simple mutation of basic Arg/His/Lys residues cannot mimic a protonated state either, therefore further limiting this approach to study ASICs.
The combined mutagenesis data support that, for several pH-sensitive Asp/Glu positions, notably E26Q, D32N, E82Q, D86N-D88N, D122N, E222Q, and E243Q, the change of charge does not contribute to activation. In these cases, the side chain carboxylic acid may engage in active state stabilizing interactions with neighboring residues or water molecules. Using the site-directed mutagenesis approach may only conclusively evaluate the perturbation of residue titratability, as Asn and Gln residues may in fact be poor mimics of protonated Asp and Glu residues. Either this is the case or the aforementioned residues must maintain their deprotonated state at pH 4 and provide stability to the active state as charged moieties, contradicting pKa predictions. Additionally, it is expected that a good mimic would maintain the same conformation as the protonated version of an Asp/Glu residue, and that this should be witnessed in a crystallographic structure. Yet the E82Q structure differs from Wt GLIC at pH 4 in which E82 would be presumably protonated. As previously mentioned, replacement of His residues may also only evaluate the removal of their titratability, whereas a simple mutagenesis cannot mimic the His protonated state. The functional results indicate either a sensitivity to protons or a structural importance for both H235 and H277, neither of which are crucial for proton-elicited gating of the GLIC.
Overall, the data suggest a complex network of H-bonds and polar interactions, with important positions below the orthosteric site, in the pH-sensitive channel opening mechanism of the GLIC. A specific importance of the ECD-TMD interface was identified with the position E35 acting as a key sensor next to the D32-R192-D122 triad involved in the signal transduction between domains [43].
Residue numbering follows that which was established by Bocquet et al. [54] for consistency. This numbering is shifted by one from other reports [35,49] that use the correct GLIC numbering.
X. laevis oocytes were obtained from the Centre de Ressources Biologiques–Rennes, France. Defolliculated oocytes were maintained at 4°C in a modified Barth’s saline solution (88 mM NaCl, 1 mM KCl, 1 mM MgSO4, 2.5 mM NaHCO3, 5 mM HEPES/Na pH 7.3) with 0.7 mM CaCl2. After intranucleus injection of approximately 30 nL cDNA (80 ng/μl specified clone cDNA with 20 ng/μl of GFP cDNA), using a compressed air microinjection system, the oocytes were transferred to 18°C. One to two days later they were evaluated for GFP expression, and subsequently maintained at 15°C. Recordings were made 1–5 d after injection using low-resistance (0.2–2 MΩ) electrodes filled with 3 M KCl, with a −40 mV holding potential. The standard solution superfusing the oocyte during recording at room temperature was 100 mM NaCl, 3 mM KCl, 1 mM MgCl2, 1 mM CaCl2, 10 mM MES at either pH 7.3, 7.5, or 8 using NaOH. In order to maintain the desired pH and maintain an equivalent Na+ concentration in all solutions, the stock solution was adjusted to the indicated pH using NaOH, and lower pH solutions were subsequently obtained using HCl and the addition of the stock solution.
Measurements were performed using pClamp 10.5 software, with a Digidata 1440A digitizer and GeneClamp 500 amplifier (Molecular Devices, LLC, Sunnyvale, CA), using an 8-valve (PS-8H) programmable gravity-driven pinch valve perfusion system (Bioscience Tools, San Diego, CA). pH-dependent responses were elicited by switching from pH 7.3–pH 8 to a series of pH values, with a minimal pH of 3.7, and a 0.5-log-unit increment from either pH 6.5, or pH 7.5 for elevated pH50 mutants. Perfusion times varied from 30 s to 90 s, with equivalent recorded wash periods in the holding buffer. pH series were performed either in an ascending order directly followed by a descent, or a descending order directly followed by an ascent, in order to remove bias. Evaluation of currents was done using Clampfit 10.5 (Molecular Devices, LLC), with Imax (μA) reported as the peak amplitude of negative going current with the holding current subtracted. The average of the two recorded peak values for a given pH was plotted in GraphPad Prism 4 (GraphPad Software, Inc, La Jolla, CA) against the pH and fitted with a nonlinear sigmoidal dose-response fit to obtain 1 (n-unit) value of pH50. The given number of injections and number of recorded oocytes per construct are listed in S1 Table, with generally 2–4 oocytes recorded per injection. Fits with an R2 < 0.9, a Hill-slope < 0.6 or > 4, or an absolute Imax < 0.9 μA were excluded from inclusion into mean values, and therefore not counted in the n-unit either. The Imax cutoff was chosen due to endogenous current at pH 4 or lower that appears on occasion. To be sure that this current does not greatly influence the pH50 calculation, a cutoff greater than 3-fold was chosen. The arbitrary cutoffs for Hill-slope and R2 were chosen to remain consistent in the removal of data that could not be properly fitted as a result of any circumstance. In order to minimize the influence of the intrinsic variability between oocyte batches, which show some variation in the Wt response to pH changes, mutants were also characterized by a ΔpH50. The ΔpH50 value corresponds to the variation of pH50 between each mutant expressing cell and the Wt cell(s) measured in the same batch of oocytes, using the same solutions of pH. The pH50 values and ΔpH50 are reported as mean ± standard deviation.
GLIC variants were purified as previously reported [54]. PET20b vectors carrying the GLIC constructs fused with an N-terminal maltose-binding protein (MBP) tag were used to transform E. coli C43 cells, cultured at 37°C in the 2YT medium containing 100 mg/ml ampicillin. At an optical density (OD) of 0.8, the cultures were cooled to 20°C and 0.1 mM IPTG was added for an overnight induction. All the purification steps were carried out at 4°C. Proteins were extracted from the cell membrane with a Tris-buffered saline solution (TBS, 300 mM NaCl, 20 mM Tris pH 7.6) containing 2% n-dodecyl-β-D-maltoside (DDM). Solubilized proteins were subsequently isolated by ultracentrifugation, loaded onto an amylose resin, and incubated for approximately 1 h. The resin was extensively washed using a TBS containing 0.1% DDM and subsequently with a TBS containing 0.02% DDM. Thrombin enzyme was added into the MBP-GLIC-bound resin and incubated overnight. The GLIC protein was eluted using a TBS containing 0.02% DDM and 20 mM maltose. A further purification step was carried out by size exclusion chromatography on a Superose 6 10/300 column (GE Healthcare, Little Chalfont, United Kingdom), which was equilibrated with a TBS containing 0.02% DDM. The fractions of the peak corresponding to the molecular weight of the GLIC pentamer were collected and concentrated to around 10 mg/ml for crystallization.
The concentrated protein was mixed at 1:1 volume ratio with a mother liquor solution containing 12%–14% PEG 4000, 400 mM NaSCN, 15% glycerol, 3% DMSO, and 0.1 M NaAcetate pH 4. The crystallization procedure was performed at 20°C using the hanging drop method. Microseeding was performed after an initial crystallization setup. Crystals appeared overnight and grew to full size in 2 wk. The crystals were flash frozen using liquid nitrogen. The diffraction data sets were collected either on the beamlines Proxima-1 of the SOLEIL Synchrotron or the European Synchrotron Radiation Facility (ESRF) ID29 and ID23A. The data sets were processed with xdsme [55] and further processed by CCP4 programs [56]. The structures were solved by molecular replacement in PHASER [57] using the GLIC (PDB ID: 4HFI) as the initial model. Further refinement was carried out using BUSTER refinement [58]. The quality of the structural models was checked by Molprobity web server [59]. All structures have been deposited in the Research Collaboratory for Structural Bioinformatics protein data bank (https://www.rcsb.org), respective deposition IDs and statistics for all crystal structures are listed in S2 Table.
RMSD and Cα distance calculations were performed by aligning a given structure (M) using the 2 subunit pair chains A+B, B+C, C+D, and D+E upon the chain pairs A+B, B+C, C+D, and D+E of a reference Wt structure at either pH 4 or 4.6 (Wt). The alignment of pair E+A was replaced with only an alignment of the E chain of M upon each chain of Wt, and each individual chain of M was aligned upon chain E of Wt as the Pymol structural alignment algorithm had difficulties doing an alignment with nonconsecutive chain pairs. The pairwise alignment method was chosen to include quaternary intersubunit interface interactions that a simple single chain alignment would ignore. The RMSD of each residue, for which alternate side chain conformations were removed, as well as the Cα atom distance, was calculated between the 25 pairwise alignments, and subsequently averaged to yield VRMSD(M-Wt) and VΔCα(M-Wt). The calculated intrinsic variation, VRMSD(Wt2-Wt1) and VΔCα(Wt2-Wt1), between the two Wt structures at pH 4.6 (Wt1) and 4.0 (Wt2) in which Wt1 and Wt2 are 3EAM and 4HFI, respectively, was used to obtain the reported “normalized” value (RMSD/RMSDctrl and ΔCα/ΔCαctrl) in which RMSD/RMSDctrl=VRMSD(M-Wt1)+VRMSD(M-Wt2)2*VRMSD(Wt2-Wt1) and ΔCα/ΔCαctrl=VΔCα(M-Wt1)+VΔCα(M-Wt2)2*VΔCα(Wt2-Wt1) for a given mutant, M, in Fig 6 and S1 Fig.
|
10.1371/journal.pgen.1003151 | A Synthetic Interaction Screen Identifies Factors Selectively Required for Proliferation and TERT Transcription in p53-Deficient Human Cancer Cells | Numerous genetic and epigenetic alterations render cancer cells selectively dependent on specific genes and regulatory pathways, and represent potential vulnerabilities that can be therapeutically exploited. Here we describe an RNA interference (RNAi)–based synthetic interaction screen to identify genes preferentially required for proliferation of p53-deficient (p53−) human cancer cells. We find that compared to p53-competent (p53+) human cancer cell lines, diverse p53− human cancer cell lines are preferentially sensitive to loss of the transcription factor ETV1 and the DNA damage kinase ATR. In p53− cells, RNAi–mediated knockdown of ETV1 or ATR results in decreased expression of the telomerase catalytic subunit TERT leading to growth arrest, which can be reversed by ectopic TERT expression. Chromatin immunoprecipitation analysis reveals that ETV1 binds to a region downstream of the TERT transcriptional start-site in p53− but not p53+ cells. We find that the role of ATR is to phosphorylate and thereby stabilize ETV1. Our collective results identify a regulatory pathway involving ETV1, ATR, and TERT that is preferentially important for proliferation of diverse p53− cancer cells.
| The conversion of a normal cell into a cancer cell involves activating genes that promote cancer growth (oncogenes) and/or inactivating genes that normally act to inhibit cancer growth (tumor suppressor genes). The tumor suppressor gene p53 is the most frequently mutated gene in human cancers, being inactivated in approximately half of all tumors. In addition, loss of p53 function is often associated with increased resistance to chemotherapy and/or poor survival. For these reasons, the selective destruction of p53-deficient (p53−) tumors has remained one of the most important goals and challenges of cancer therapy. One strategy for destroying p53− tumors is to inactivate genes that are preferentially required for the growth or survival of p53− cells. Here we carry out a large-scale genetic screen to identify a cellular pathway that is preferentially required for growth of p53− cancer cells.
| The p53 tumor suppressor (also called TP53; NP_000537.3) plays a pivotal role in regulating multiple cellular processes including cell cycle arrest, apoptosis, cell metabolism and senescence (reviewed in [1]). Activated p53 can either induce cell cycle arrest and inhibit cell growth or promote cell apoptosis depending on the type of stress and the cellular context. Mutations that perturb p53 function, typically in its DNA-binding domain, or disruptions of the p53 upstream or downstream regulatory networks, have been found in more than half of all cancer cases and are present in cancer-prone families with Li-Fraumeni syndrome (OMIM#151623) (reviewed in [2]). In addition, loss of p53 function is often associated with increased resistance to chemotherapy and/or poor survival (see, for example, [3]–[5]). For these reasons, the selective molecular targeting of p53-deficient (p53−) tumors has remained one of the most important goals and challenges of molecular oncology (reviewed in [6]).
One strategy for treating p53− tumors is to reestablish the growth-inhibitory functions of p53. The feasibility of this approach is supported by animal studies demonstrating that reactivation of wild type p53 leads to tumor regression [7]–[9]. Several experimental strategies have been used to restore p53 activity. For example, gene therapy involving viral vectors has been used to reintroduce p53 into p53− tumor cells [10]. Alternatively, for cancers that retain a wild type p53 gene, small molecule drugs have been identified that stabilize and activate p53 by interfering with its negative regulator MDM2 (NP_002383.2) (reviewed in [11]). Restoration of p53 function in cancers expressing only mutant p53 is even more challenging. However, small molecules that refold mutant p53 proteins, and thus reactivate p53 function, have been described (reviewed in [12]). Some of these approaches have progressed to clinical trials but to date none have been found to have clearly demonstrable clinical benefit [13].
An alternative approach to restoration of p53 function would be to target proteins that are preferentially required for proliferation or survival of p53− cells. Such targets can, in principle, be identified by synthetic lethal interaction screens, an idea first proposed by Hartwell et al. based upon studies in the budding yeast Saccharomyces cerevisiae [14]. The validity of this approach is supported by the realization that cancer cells are highly dependent upon or “addicted” to specific genes and regulatory pathways, confirming the existence of cancer cell-selective synthetic interaction targets (reviewed in [15], [16]). In addition, an important proof-of-principle is the demonstration that small molecule inhibitors of poly (ADP-ribose)-polymerase (NP_001609.2) are synthetically lethal with loss-of-function mutations in BRCA1 (NP_009225.1), BRCA2 (NP_000050.2) as well as other components of the homologous recombination pathway [17]–[19]. Here we carry out an RNA interference (RNAi)-based synthetic interaction screen to identify a regulatory pathway that is preferentially required for proliferation of p53− cancer cells.
To identify genes that are preferentially required for the viability and proliferation of p53− cancer cells, we designed a synthetic interaction screen, which is summarized in Figure 1A and briefly described below. The primary screen was carried out using a well-characterized isogenic pair of human HCT116 colorectal cancer cell lines, one harboring wild type p53 (p53+ HCT116) and the other bearing a homozygous p53 deletion (p53− HCT116) [20]. For these and all other cell lines used in this study, the presence or absence of functional p53 was confirmed by monitoring expression of the p53 target gene, p21 (also called CDKN1A; NP_510867.1) (Figure S1). A human shRNA library comprising ∼60,000 shRNAs directed against ∼27,000 genes [21] was packaged into lentivirus particles, pooled and used to infect in parallel the two HCT116 cell lines. Ten days later, genomic DNA from both cell lines was isolated, and shRNAs were PCR amplified and subjected to massively parallel sequencing; as a reference, the starting shRNA population in both cell lines (taken 40 hours post-infection) was also analyzed.
Statistical analysis of the four shRNA populations identified shRNAs targeting 103 genes (Table S1) whose abundance was significantly decreased in p53− HCT116 cells (≥4-fold) but not in p53+ HCT116 cells (≤2-fold) at 10 days post-infection relative to the earlier time point (Figure 1B). Such shRNAs are presumably synthetic with the p53 deletion, thus rendering p53− cells harboring these shRNAs inviable or growth impaired, and leading to their relative under-representation in the p53− HCT116 population.
To validate candidates isolated from the primary screen, each shRNA was analyzed in an independent colony formation assay. p53+ and p53− HCT116 cells were infected with a lentivirus expressing a single candidate shRNA, and 10 days later cells were puromycin selected, re-plated at low density, and monitored for colony formation. This secondary screen revealed 11 genes that, when knocked down, substantially decreased colony formation in p53− HCT116 cells compared to p53+ HCT116 cells (Figure 1C). ShRNAs targeting these 11 genes also preferentially decreased the ability of p53− HCT116 cells to proliferate in culture (Figure 1D and summarized in Table S2). Quantitative RT-PCR (qRT-PCR) confirmed in all cases that expression of the target gene was decreased in the knockdown cell line (Figure S2A).
To rule out the possibility that growth inhibition was due to an off-target effect of the shRNAs, for each of the 11 genes we derived a short interfering RNA (siRNA) whose sequence was unrelated to the original shRNA used in the experiments described above. Figure 1E shows that each siRNA also preferentially decreased proliferation of p53− HCT116 cells compared to p53+ HCT116 cells. In all cases, qRT-PCR analysis confirmed that the siRNA decreased expression of the target gene (Figure S2B).
p53− tumors from both the same and different types of cancers vary substantially with regard to additional genetic and epigenetic aberrations [22]. We were therefore interested in determining whether the 11 genes we identified were also preferentially required for the growth of other p53− cancer cells. To address this issue, we first analyzed an isogenic pair of human RKO colorectal cancer cell lines, one harboring wild type p53 (p53+ RKO cells) and the other bearing a homozygous p53 deletion (p53− RKO cells) (see Figure S1). ShRNAs to the 11 genes were introduced into the isogenic pair of RKO cell lines and proliferation was monitored. The results of Figure 2A indicate that five genes (ATR [NP_001175.2], ETV1 [NP_001156619.1], GFPT2 [NP_005101.1], NT5C3 [NP_001002009.1] and UMPS [NP_000364.1]) were preferentially required for growth of p53− RKO cells compared to p53+ RKO cells. By contrast, knockdown of the other six genes did not substantially inhibit growth of either p53− or p53+ RKO cells and were thus not further analyzed.
We next examined these five candidates in an unrelated isogenic pair of A549 human lung cancer cell lines. In this case, the parental p53+ A549 cells were rendered p53− by stable expression of a p53 dominant-negative mutant [23] (see Figure S1). The results of Figure 2B show that siRNAs against the five candidate genes (ATR, ETV1, GFPT2, NT5C3 and UMPS) preferentially inhibited growth of the p53− A549 cell line.
Finally, we analyzed the five candidate genes in a panel of four human lung cancer cell lines, two of which expressed wild type p53 (A549 and NCI-H460) and two of which were compromised for p53 function (NCI-H1299, which lacks p53, and NCI-H522, which expresses a p53 mutant) (see Figure S1). Of the five candidate genes, knockdown of two, ATR and ETV1, were the most consistent in preferentially inhibiting proliferation of p53− cell lines (Figure 2C) and were selected for further analysis. ATR encodes a checkpoint kinase involved in the DNA damage response [24], and ETV1 encodes a member of the ETS family of transcription factors [25].
We also tested whether knockdown of ATR and ETV1 would preferentially inhibit growth of p53− HCT116 tumors in a mouse xenograft model. p53+ or p53− HCT116 cells expressing an shRNA against ATR or ETV1, or a control non-silencing shRNA, were injected subcutaneously into opposite flanks of the same nude mouse, and tumor growth was monitored after four weeks. As expected, the control p53− HCT116 cells formed larger tumors than their p53+ counterparts (Figure 2D). Notably, knockdown of ATR or ETV1 markedly inhibited growth of p53− HCT116 tumors but did not have a significant effect on growth of p53+ HCT116 tumors.
We next sought to investigate the basis by which ETV1 and ATR were preferentially required for growth of p53− cells. A previous study has shown that ETV1 is a transcriptional activator of TERT (NP_001180305.1) [26], which encodes the catalytic subunit of telomerase and has a well-established role in the maintenance of cellular proliferation [27]. Therefore, in the first set of experiments we analyzed the effect of depleting ETV1 as well as ATR on TERT levels. Significantly, RNAi-mediated knockdown of ETV1 or ATR resulted in a substantial decrease in TERT protein (Figure 3A) and mRNA (Figure 3B) levels in p53− HCT116 cells but unexpectedly had only a modest effect on TERT levels in p53+ HCT116 cells. The effect of knockdown of both ETV1 and ATR in p53− HCT116 cells on cellular proliferation and TERT levels was similar to that observed with single knockdowns (Figure S3A). Pharmacological inactivation of ATR using two different chemical inhibitors, CGK773 [28] and ETP46464 [29], also resulted in decreased TERT levels in p53− but not p53+ HCT116 cells (Figure 3C). Inhibition of ATR was confirmed by monitoring phosphorylation of its target substrate, CHK1 (also known as CHEK1; NP_001107593.1) (Figure S4).
We also monitored senescence induction and performed cell cycle analysis in cells depleted of ETV1 or ATR. Figure 4A shows that in both p53+ and p53− HCT116 cells, RNAi-mediated knockdown of TERT substantially increased the number of cells that stained positively for senescence-associated β-galactosidase activity, indicative of senescence induction (see also Figure S5A). The level of senescence was higher in p53+ HCT116 TERT knockdown cells than in p53− HCT116 TERT knockdown cells, as expected, because p53-directed pathways contribute to senescence [1]. Significantly, Figure 4B shows that RNAi-mediated knockdown of ETV1 or ATR also induced senescence (see also Figure S5B). However, following knockdown of ETV1 or ATR, the induction of senescence was much greater in p53− HCT116 cells compared to p53+ HCT116 cells (Figure 4B and Figure S5C), consistent with the difference in TERT levels (see Figure 3A). In addition, knockdown of TERT increased the percentage of p53− HCT116 cells but not p53+ HCT116 cells in G2/M (Figure 4C and Figure S6A). Notably, a similar preferential increase in the percentage of p53− HCT116 cells in G2/M occurred following knockdown of ETV1 or ATR (Figure 4D and Figure S6B, S6C).
To determine whether decreased TERT levels were responsible for the preferential growth defect in p53− HCT116 cells depleted of ETV1 or ATR, we performed a “rescue” experiment. Proliferation was measured both by an Alamar Blue assay (Figure 4E) and by quantifying cell number (Figure S7A) following knockdown of ETV1 or ATR in p53+ and p53− HCT116 cell lines stably expressing either TERT or, as a control, green fluorescence protein (GFP) (Figure S7B). The results of Figure 4E and Figure S7A show that ectopic expression of TERT counteracted the large decrease in proliferation that occurred following knockdown of ETV1 or ATR in p53− HCT116 cells, as well as the modest proliferative decrease following ETV1 knockdown in p53+ HCT116 cells. Thus, the reduced TERT levels following depletion of ETV1 or ATR can largely explain the decreased proliferation of p53− HCT116 cells. Consistent with this conclusion, TERT knockdown had a similar effect on proliferation of p53− cell lines to that observed following knockdown of ETV1 or ATR (see Figure S3). Moreover, knockdown of both ETV1 and TERT, or ATR and TERT, decreased proliferation of p53− cell lines similarly to that observed with single knockdowns (Figure S3).
As described above, ETV1 has been previously shown to transcriptionally stimulate TERT expression [26]. However, the basis by which ATR promotes TERT expression is unknown. The similar results obtained with ATR and ETV1 in the TERT expression experiments of Figure 3 (and Figure S3) raised the possibility that the two proteins act in a common pathway. To address this possibility, we first asked whether ATR regulates ETV1 levels. The immunoblot results of Figure 5A show that in both p53+ and p53− HCT116 cells knockdown of ATR resulted in reduced ETV1 protein levels. The qRT-PCR results of Figure 5B showed that ATR depletion did not lead to reduced ETV1 mRNA levels, indicating that the ATR-mediated regulation of ETV1 occurs post-transcriptionally. Addition of an ATR chemical inhibitor also led to reduced ETV1 protein levels in both p53+ and p53− HCT116 cells (Figure 5C). Following inhibition of ATR activity, DNA damage did not result in ETV1 stabilization (Figure S8).
To test the generality of these results, we analyzed the effect of ATR pharmacological inhibition in several of the p53+ and p53− cell lines described above. Figure 5D shows that ATR inhibition reduced TERT protein levels in p53− RKO, NCI-H522 and NCI-H1299 cells but not in p53+ RKO, A549 and NCI-H460 cells Moreover, addition of an ATR inhibitor reduced ETV1 levels to varying extents in all cell lines. TERT protein levels were also reduced following knockdown of ATR or ETV1 in NCI-H522 cells and two additional human cancer cell lines that express mutant p53 (Figure S9A), as well as in HeLa cells, which lack p53 activity due to expression of the human papilloma virus E6 protein (Figure S9B). Thus, the results in these other p53+ and p53− cell lines are similar to those obtained in the isogenic pair of HCT116 cell lines used throughout this study.
Because ATR is a protein kinase, a likely mechanism for the ability of ATR to post-transcriptionally regulate ETV1 is through direct interaction and phosphorylation. Consistent with this possibility, ETV1 contains five potential ATR phosphorylation sites (Figure 6A). To test this idea, we ectopically expressed a FLAG-tagged ETV1 derivative (Figure S10) in p53+ and p53− HCT116 cells, and analyzed interaction between FLAG-ETV1 and ATR in a co-immunoprecipitation assay. The results of Figure 6B show that in both p53+ and p53− HCT116 cells, FLAG-ETV1 could be detected in the ATR immunoprecipitate (left) and, conversely, ATR could be detected in the FLAG immunoprecipitate (right), indicating ATR and ETV1 physically associate. To determine whether ETV1 was an ATR substrate, we immunoprecipitated FLAG-ETV1 from transfected p53+ and p53− HCT116 cell lysates and analyzed the immunoprecipitate by immunoblotting with an antibody that recognizes a phosphorylated serine followed by a glutamine [30], the product of ATR or ATM phosphorylation [31], [32]. The results of Figure 6C show that the immunoprecipitated FLAG-tagged ETV1 could be detected by the ATM/ATR phospho-specific antibody, suggestive of phosphorylation by ATR. Moreover, following treatment of cells with an ATR inhibitor, the immunoprecipitated FLAG-tagged ETV1 was no longer detected by the ATM/ATR phospho-specific antibody (Figure S11).
To confirm that ATR phosphorylates ETV1, we performed in vitro kinase experiments. We first tested whether ATR, in the presence of its positive effector ATRIP (NP_569055.1) [33], [34], could phosphorylate a glutathione-S-transferase (GST)-ETV1 (amino acids 1–290) fusion-protein that contained all five potential ATR phosphorylation sites. The results of Figure 6D show that ATR phosphorylated the GST-ETV1 fusion-protein but, as expected, not a control GST protein. To confirm and extend this result, we constructed and analyzed a series of GST-ETV1 fusion-proteins each containing a single potential ATR phosphorylation site. The results of Figure 6E show that only one of the five potential ATR phosphorylation sites (SQ2) was a substrate for ATR. Collectively, the results described above indicate that ATR phosphorylates ETV1 and stabilizes it from proteolytic degradation.
As discussed above, previous studies have shown that ETV1 is a transcriptional activator of TERT [26]. Therefore, we thought the most likely mechanism by which ETV1 promotes proliferation in p53− HCT116 cells is through direct binding to the TERT promoter and stimulation of TERT transcription. To test this possibility, we performed chromatin-immunoprecipitation (ChIP) experiments. The ChIP experiments of Figure 7A (left panel) show that in p53− HCT116 cells, ETV1 was bound to a region within intron 1, which has been previously reported to contain multiple ETV1 binding sites and is required for complete TERT transcriptional activity [26]. Remarkably, in p53+ HCT116 cells, whose proliferation is not dependent upon ETV1, there was no detectable binding of ETV1 to the same region of the TERT promoter. Notably, ectopic expression of wild type p53 in p53− HCT116 cells resulted in substantially decreased binding of ETV1 to the TERT promoter (Figure 7B, left). Conversely, ectopic expression of a p53 dominant-negative mutant in p53+ HCT116 cells resulted in substantially increased binding of ETV1 to the TERT promoter (Figure 7B, right).
In p53− HCT116 cells, binding of ETV1 to the TERT promoter was lost following pharmacological inhibition of ATR (Figure 7A and Figure S12A), which as shown above results in decreased ETV1 levels (see Figure 5C). Conversely, binding of ETV1 to the TERT promoter modestly increased following irradiation with ultraviolet light, which increases ATR activity (Figure S12B). ChIP experiments monitoring ATR occupancy revealed that ATR was bound to the same region of the TERT promoter as ETV1 (Figure 7C). Thus, in p53− HCT116 cells, ETV1 and ATR are both bound to the TERT promoter, which is consistent with our finding that the two proteins are physically associated (Figure 6B).
In conjunction with a previous study [26], the results presented above suggested that ETV1 is directly responsible for stimulating TERT expression and that ATR functions by phosphorylating and thereby stabilizing ETV1. A prediction of this model is that ectopic expression of ETV1 would bypass the requirement of ATR for proliferation of p53− HCT116 cells. The rescue experiment of Figure 7D shows that the decreased proliferation of p53− HCT116 cells following knockdown of ATR was counteracted by ectopic expression of ETV1 (Figure S13). Following knockdown of TERT, ectopic expression of ETV1 could no longer rescue proliferation of p53− HCT116 cells depleted of ATR (Figure S14A). In these experiments, ectopic expression of ETV1 had no effect on γ-H2AX foci formation, a marker of DNA damage [35] (Figure S14B). These results suggest that the growth arrest observed following loss of ATR is primarily due to decreased ETV1 levels.
In this report we have performed a large-scale shRNA screen to identify a regulatory pathway involving ETV1, ATR and TERT that is preferentially required for proliferation of diverse p53− cancer cells. We found that in p53− cells, TERT transcription is highly dependent upon ETV1, which functions as a direct transcriptional activator by binding to the TERT promoter downstream of the transcription start-site. In p53+ cells, ETV1, although present at comparable levels, is not required for TERT transcription and surprisingly is not bound to the same region of the TERT promoter. Notably, ectopic TERT expression restored normal proliferation in p53− cells depleted of ETV1 or ATR (Figure 4E and Figure S7A), indicating that the promotion of TERT expression is an important, but not necessarily the only, mechanism by which ETV1 and ATR maintain proliferation of p53− cells. Consistent with our results, a previous study reporting a requirement for ETV1 in TERT transcription [26] was primarily based upon experiments performed in 293T cells, which lack p53 activity due to expression of SV40 large T antigen.
The results described above suggest that p53+ cells express a transcription factor that functionally substitutes for ETV1, and that one or more proteins associated with the TERT promoter in p53+ cells prevent binding of ETV1. Several transcription factors, including SP1 (NP_612482.2), E2F1 (NP_005216.1) and MYC (NP_002458.2), have been previously shown to be associated with the human TERT promoter (reviewed in [36]). To ask whether these factors, or p53 itself, might contribute to the differential regulation of TERT we performed ChIP experiments in p53+ and p53− HCT116 cells. Consistent with previous studies, we found that E2F1 and MYC were associated with the TERT promoter; binding of E2F1 was modestly increased in p53− HCT116 cells (Figure S15A), whereas for MYC there was no difference in p53+ and p53− HCT116 cells (Figure S15B). In p53+ HCT116 cells there was increased binding of SP1 (Figure S15C) and, most notably, there was substantial binding of p53 to the TERT promoter (Figure S15D). Interestingly, a number of previous studies have reported physical and functional interactions between SP1 and p53 (see, for example, [37]–[41]). Our ChIP results reveal substantial differences between the composition of proteins associated with the TERT promoter in p53+ and p53− HCT116 cells, which may be related to the differential requirement for ETV1.
Interestingly, in contrast to human cancer cell lines, we found that ATR was not required for TERT expression in experimentally derived p53− MCF10A cells, an immortalized but non-transformed human cell line (Figure S16A). In addition, ATR was not required for TERT expression in p53− mouse embryo fibroblasts (Figure S16B), consistent with the lack of conservation between the mouse and human TERT promoter (data not shown). Thus, the requirement of ATR and ETV1 for TERT expression may be specific to human p53− cancer cell lines.
Several previous studies have reported results that are consistent with the synthetic interaction between p53 and ATR we have described here. For example, p53− cells have been found to be particularly sensitive to pharmacological inhibition of ATR (see, for example, [42]–[44]). In addition, mice expressing a hypomorphic allele of ATR have an aging phenotype that is exacerbated in the absence of p53 [45]. Significantly, mouse embryo fibroblasts containing this hypomorphic ATR allele have an elongated G2 phase following loss of p53, consistent with our cell cycle results (Figure 4D and Figure S6B, S6C). However, a preferential role for ETV1 in p53− cells and its cooperative function with ATR has not been previously described and underscores the power of unbiased, large-scale RNAi-based screens.
Our screening strategy did not emphasize reaching saturation but rather sought to follow-up, by directed experiments, a limited number of candidates isolated in the primary screen. For several reasons, we believe that our screen, like other large-scale shRNA screens (see, for example, [46]), did not achieve saturation. For example, a previous siRNA screen identified several factors, in particular the serine/threonine kinase receptor-associated protein UNRIP (also called STRAP; NP_009109.3), whose loss affected proliferation of p53− HCT116 cells more severely than p53+ HCT116 cells [47]. However, we did not isolate UNRIP in our primary screen and, conversely, ATR and ETV1 were not isolated in the previous siRNA screen, suggesting that neither screen was truly saturating. Reasons for a failure to reach saturation in this and other large-scale shRNA screens include suboptimal efficacy of some shRNAs [48], unequal representation of shRNAs in the primary screen, and an insufficient depth of deep sequencing. Thus, it is possible that additional factors that act in the ATR-ETV1-TERT pathway, or unrelated pathways preferentially required for proliferation of p53− cells, remain to be identified.
The decreased proliferation of p53− cell lines was first evident within a few days following knockdown of ETV1, ATR or TERT. It therefore seems likely that this reduced proliferation is not a result of replicative senescence due to telomere attrition, which would require many cell divisions. Senescence occurred at much later times (10–14 days) and may be a secondary effect of the proliferation block. We observed that knockdown of ETV1, ATR or TERT resulted in an increased percentage of cells in G2/M (Figure 4C, 4D and Figure S6). Although senescent cells are generally believed to arrest in G1, it has been found that senescent cells can also arrest in G2/M (see, for example, [49]).
A variety of previous studies have shown that TERT can promote proliferation by multiple mechanisms, several of which are unrelated to telomere length including inhibiting apoptosis [50], regulating cell signaling pathways and/or stimulating expression of diverse growth-promoting genes (see, for example, [51]–[54]). It seems likely that the decreased proliferation of p53− cells following depletion of ETV1, ATR or TERT involves one of these alternative mechanisms. We have found that p53− cells depleted of ETV1, ATR or TERT have multiple growth defects including increased levels of senescence (Figure 4A, 4B and Figure S5) and an altered cell cycle (Figure 4C, 4D and Figure S6). A further understanding of how TERT promotes proliferation of p53− cells is likely to identify new factors that are potential therapeutic targets.
Animal experiments were performed in accordance with the Institutional Animal Care and Use Committee (IACUC) guidelines.
Isogenic p53+ and p53− HCT116 and RKO cell lines [20] were provided by B. Vogelstein; A549, NCI-H460, NCI-H522, NCI-H1299 and HT29 cells were obtained from the National Cancer Institute; and DLD-1, HeLa and MCF10A cells were obtained from the American Type Culture Collection. The basis for the p53− status in each of the p53− cell lines is provided in Table S3. p53+ and p53− mouse embryonic fibroblasts were isolated from wild type and p53−/− C57BL/6 mice. All cells were grown according to the supplier's recommendations. Stable A549 and MCF10A cell lines expressing p53-DD, which harbors a deletion of 288 amino acids (Δ15-301; [23]) were generated by transfection with the plasmid pBABE-hygro-p53DD (Addgene; [55]) or the control vector, pBABE-hygro, and selection with hygromycin (150–200 µg/ml). Stable p53+ and p53− HCT116 cell lines expressing TERT were generated by transfection with the plasmid pWZL-Blast-Flag-HA-hTERT (Addgene; [56]) or control plasmid pWZL-Blast-GFP (Addgene; [57]), and selection with blasticidin (10 µg/ml). The ETV1 expression vector was generated by subcloning ETV1 cDNA (Open Biosystems) into pEF6-Blast-3xFlag to create pEF6-Blast-3xFlag-ETV1. The pEF6-Blast-3xFlag vector was generated by cloning a BsiWI-EcoRI double-stranded oligo coding for 3xFlag-tag (MDYKDHDGDYKDHDIDYKDDDDKEF) in Kpn1-EcoR1-digested pEF6/V5-HIS B (Invitrogen). Stable p53+ and p53− HCT116 cell lines expressing ETV1 were generated by transfection with pEF6-Blast-3xFlag-ETV1 or vector only and selection with blasticidin (10 µg/ml).
The Open Biosystems GIPZ lentiviral human shRNAmir library was obtained through the University of Massachusetts Medical School RNAi Core Facility. Twelve lentiviral pools, each comprising ∼5000 shRNA clones, were generated with titers of ∼2×106 pfu/ml. These lentiviral stocks were produced following co-transfection with the packaging mix into the 293T packaging cell line. To carry out the screen, p53+ and p53− HCT116 cells were plated at 1×106 cells per 100 mm plate, transduced the next day with one shRNA pool per plate at a multiplicity of infection (MOI) of 1, and grown in the absence of puromycin selection. Forty hours after transduction, 75% of cells were transduced (as evidenced by GFP fluorescence; the marker turboGFP is present in the pGIPZ vector). Each plate was divided into two populations: half of the cells were pooled and genomic DNA was extracted (referred to as “T0”), whereas the other half were transferred to 150 mm plates and passaged by 4-fold dilutions for 10 days, at which point the cells were pooled and the genomic DNA was extracted (referred to as “T10”).
To analyze the frequency of individual shRNAs in the four populations, 72 µg of genomic DNA was used as the substrate (split into 24 tubes) and PCR amplified (94°C for 1 min; 15 cycles of 94°C for 1 min, 58°C for 1 min, 72°C for 45 sec; 72°C for 10 min; and hold at 4°C) with primers GIPZF (5′-GAGTTTGTTTGAATGAGGCTTCAGTAC-3′) and GIPZHR (5′-CGCGTCCTAG GTAATACGAC-3′). The PCR product was gel purified, and 50 ng of DNA was used as the substrate for a second PCR amplification (94°C for 1 min; 15 cycles of 94°C for 1 min, 50°C for 1 min, 72°C for 45 sec; 72°C for 10 min; and hold at 4°C) using primers Forward Acu1 primer AMN (5′-CAACAGAAGGCTCCTGAAGGTATATTGCTGTTGAC-3′) and Reverse Acu1 primer AMN (5′-AAATTTAAACTGAAGTACATCTGTGGCTTCACTA-3′). Next, 1 µg of the PCR product was digested to completion with AcuI (New England Biolabs). The digested product was then ligated to the following pre-annealed adapters: L1ShSolexA (/5Bio/-ACACTC TTTCCCTACACGACGCTCTTCCGATCTCA) and L1ShSolexB (/5Phos/′-AGATCGGAAGA GCGTCGTGTAGGGAAAGAGTGT/3AmM, and L2ShSolexB (/5Phos/-AGATCGGAAGAGC TCGTATGCCGTCTTCTGCTTG/3Bio/) and L2ShSolexA (/5AmMC6/-CAAGCAGAAGACG GCATACGAGCTCTTCCGATCTAC). The product of the 3-way ligation was run on a 3% TAE agarose gel, visualized with ethidium bromide, purified and used as a substrate for a 15-cycle PCR reaction using Solexa-Illumina primers 1.1 and 2.1 and the cycling conditions recommended by the manufacturer.
The library was analyzed using the Solexa-Illumina GA Massively Parallel Deep Sequencer. Sequence information was extracted from the image files using the Solexa-Illumina Firecrest and Bustard applications. Prior to alignment of the sequence reads, a custom Perl script was used to identify the first six bases flanking the informative sequence in 5′ and the six bases flanking the informative sequence in 3′, starting at position 28. The core 21 bp sequences were extracted and mapped to the human reference genome sequence (hg18) using the Solexa-Illumina ELAND algorithm, allowing up to two mismatches to the reference sequence. No further analysis was performed on reads that did not contain the six bases of the 5′ sequence or the six bases of 3′ adapter sequence.
Sequences mapping to the same genomic location were binned and the count for each of the mapped genomic sequences was calculated for each of the four treatments. For each of the mapped genomic sequences, the Fisher Exact Test was applied to assess whether there was a differential depletion/enrichment of the shRNA sequences between T0 and T10 for both the p53− and p53+ HCT116 cell lines. The odds ratio and its 95% confidence interval were computed for each of the mapped genomic sequences using Fisher test function in R v2.8 based on conditional maximum likelihood estimation. To adjust for multiplicity, B–H method [58] was used. Those shRNAs with an adjusted p-value<0.01 and a decrease of at least four-fold at T10 compared with T0 in p53− HCT116 cells and no more than two-fold in p53+ HCT116 (or adjusted p-value≥0.01) were identified. The data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus [59] and are accessible through GEO Series accession number GSE15967 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE15967).
Lentiviral supernatants corresponding to individual shRNAs (listed in Table S4) were generated in 293T cells as described above. p53+ and p53− HCT116 cells were transduced with each lentiviral preparation at an MOI of 0.2–0.4, and grown for 10 days without puromycin selection, during which cells were passaged at a 1∶6 ratio every 4 days. Cells were then subjected to puromycin selection (1.5 µg/ml) for 5 days. For colony formation assays, cells were split at a 1∶200 ratio and plated in 6-well plates in the presence of 1.5 µg/ml puromycin. After 6–7 days, cells were fixed with 4% paraformaldehyde in phosphate buffered saline (PBS) at 4°C overnight and then stained with 0.1% crystal violet in PBS to visualize the colonies. At least two independent infections were performed for each shRNA; representative images are shown.
For shRNA-based experiments, cells were transduced with lentiviral supernatants at an MOI of 0.2–0.4, and subjected to puromycin selection (1.5 µg/ml) for 5 days. Cells were then passaged at a 1∶8 ratio every 3 days and cultured in growth medium containing puromycin. After 4 passages, cells were split at a 1∶6 ratio and seeded in a 12-well plate in RPMI medium without phenol red and supplemented with 5% fetal calf serum. After 18 h, the medium was replaced with 500 µl of medium containing 10% of an Alamar Blue solution (Invitrogen). After 2 h, 100 µl of the medium was used to measure fluorescence by excitation at 530 nm and emission at 590 nm. For siRNA-based experiments, siRNA duplexes were transfected into cells using Lipofectamine RNAiMax Transfection Reagent (Invitrogen) according to the manufacturer's instructions. Briefly, 1.2 µl of Lipofectamine was complexed with the siRNA (40 nM final concentration), and the solution was diluted with 100 µl of medium and applied to 2×104 cells in a volume of 500 µl culture medium per well in 24-well plates. The medium was changed after 24 h and proliferation assessed by Alamar Blue fluorescence after an additional 72 h. Sequences of siRNAs are listed in Table S4; the control LMNA siRNA sequence was previously described [60]. All experiments were performed at least 2–3 times in either duplicate or triplicate.
Four days post-siRNA-transfection, cells were trypsinized, resuspended in 0.5 ml growth medium, and stained with 0.5 ml 0.1% Trypan blue solution (HyClone Trypan blue, Thermo Fisher Scientific). Viable cells were counted using a Countess Automated Cell Counter (Life Technologies). Two independent transfections were carried out and analyzed in duplicate.
2×106 shRNA-transduced p53+ or p53− HCT116 cells were suspended in 100 µl of serum-free RPMI and injected subcutaneously into the opposite flanks of n = 9 (for non-silencing and ATR shRNAs) or n = 5 (for ETV1 shRNA) athymic Balb/c (nu/nu) mice (Taconic). Tumor dimensions were measured every week and tumor volume was calculated using the formula π/6×(length)×(width)2. A Mann-Whitney test was used to determine whether knockdown of ATR or ETV1 changes the tumor volume at week 4 compared to a non-silencing shRNA.
Cell extracts were prepared by lysis in modified RIPA buffer (0.05 M Tris-Cl [pH 8.0], 0.15 M NaCl, 1% Nonidet P-40, 0.5% desoxycholate, 0.1% SDS, 2 mM phenylmethylsulphonyl fluoride (PMSF), 20 µg/ml aprotinin, 1 mM Na3VO4 and 1 mM NaF) in the presence of a proteinase inhibitor cocktail (Roche). Blots were probed with α-TERT (Epitomics, 1531-1). α-ETV1 (Abcam, ab81086), α-Flag M2 (Sigma, F1804), α-phospho-CHK1(Ser317) (Cell Signaling Technology, 8191), α-p21 (BD Pharmingen, SX118), α-tubulin (Sigma, B5-1-2) or β-actin (Sigma, AC74). For ATR inhibition, cells were treated with 2–6 µM CGK733 (Calbichem) or 0.5–8 µM ETP46464 ([29]; kindly provided by O. Fernandez Capetillo) for 72 h prior to cell extract preparation; as a control, cells were treated with dimethyl sulfoxide (DMSO). For p53 functional assays (Figure S1), cells were treated with 25 µM etoposide (Sigma) or 10 µg/ml 5-fluorouracil (Sigma) for 24 h, and cell extracts were prepared as above. For RNAi experiments, experiments were performed at least 2–3 times in either duplicate or triplicate.
Total RNA was extracted using TRIzol Reagent (Invitrogen) and treated with Turbo DNA-free kit (Ambion Inc.). The same amount of total RNA (3 µg) for each sample was employed to produce templates for SYBR-green quantitative PCR analysis using SuperScript II Reverse transcriptase (Invitrogen). Target genes were amplified using specific primers and expression levels were normalized to that of GAPDH. Primer sequences are listed in Table S4. All experiments were performed at least 2–3 times in either duplicate or triplicate.
Assays were performed as described previously [61] with minor modifications. Briefly, 10–14 days following RNAi-mediated knockdown, cells were washed twice with PBS, then fixed using 3.7% paraformaldehyde for 5 min at room temperature. After three washes with PBS, cells were incubated with fresh staining solution (40 mM citric acid/Na2HPO4 pH 6.0, 150 mM NaCl, 2 mM MgCl2, 5 mM potassium ferricyanide, 5 mM potassium ferrocyanide, 1 mg/ml X-Gal) for 12–18 hr at 37°C (no CO2) and covered from light. Images were captured using a Spot TE-200 digital camera (SPOT Imaging Solutions). The number of blue cells in 10 fields (each containing 100–250 cells) was counted manually, and the percentage calculated. Two independent infections were performed for each knockdown.
Cells transduced with shRNAs were harvested by trypsinization, fixed in 80% ethanol and stored at −20°C overnight. Fixed cells were stained with propidium iodide buffer containing 50 µg/ml RNase (Sigma) and 50 µg/ml propidium iodide (Sigma) in PBS. Flow cytometry was performed by the UMass Medical School Core Flow Cytometry Lab using a FACScalibur flow cytometer (Becton Dickinson). Data were analyzed with FlowJo (Tree Star). All experiments were performed at least 2–3 times.
For Figure 6B, 5×107 p53+ or p53− HCT116 cells expressing Flag-ETV1 were rinsed twice with cold PBS, lysed in 1 ml IP lysis buffer (50 mM Tris-Cl pH 7.4, 250 mM NaCl, 5 mM EDTA, 0.2%Triton X-100, 0.5 mM DTT, 1× complete protease inhibitor [Roche], and phosphatase inhibitor cocktails 2 and 3 [Sigma, p5726 and p0044]) on ice. The lysate was cleared by centrifugation at 16,000 g for 30 min at 4°C. Whole cell lysate (2 mg per sample) was incubated with relevant antibodies (α-ATR [Abcam, ab2905] or control rabbit IgG [Abcam, ab37415] or α-Flag M2 [Sigma] or control mouse IgG [Santa Cruz, sc2343]) overnight at 4°C after being precleared with 50 µl Dynabeads-protein G (Invitrogen). Dynabeads Protein G (50 µl) were added to each lysate-antibody complex, incubated for 2 h, spun, and washed 5 times with IP lysis buffer. Protein complexes were eluted by boiling with Laemmli buffer. For Figure 6C, immunoprecipitations were carried out as described above with α-Flag M2 (Sigma), then immunoblotted with α-SQ2 ([30]; kindly provided by S. Elledge), or α-ETV1 (Abcam).
To create GST-ETV1 (amino acids 1–290), the corresponding portion of ETV1 was PCR amplified using pEF6-Blast-3xFlag-ETV1 as a template and cloned into pGEX-4T-3 (GE Healthcare). The construct was confirmed by sequencing. For the smaller GST-ETV1 fusion proteins, synthetic oligos corresponding to amino acids 9–23 (SQ1), 44–58 (SQ2), 97–111 (SQ3), 165–179 (SQ4) or 240–254 (SQ5) were annealed and cloned into pGEX-4T-3. In vitro kinase assays were performed as previously described [62] except that reaction volumes were quadrupled. 32P-labeled products were visualized by autoradiography.
ChIP assays were carried out as described previously [63], [64] with the following minor modifications. Briefly, 5×107 cells were first incubated with ethylene glycol bis(succinimidyl succinate) (EGS) for 30 min and then incubated with 1% formaldehyde for 10 min at room temperature before crosslinking was quenched by addition of 0.125 M glycine. Cells were collected by centrifugation and lysed in lysis buffer containing 50 mM Tris–HCl pH 8.0, 10 mM EDTA, 0.5% SDS, proteinase inhibitors (Roche) and phosphatase inhibitors (Sigma). The cell suspension was sonicated for 15 min total time with 30 seconds ON and 30 seconds OFF using Bioruptor (Diagenode). Sonicated chromatin was then incubated at 4°C overnight with 5 µg of the appropriate antibody: α-ATR (Abcam), α-ETV1 (Abcam), α-E2F1 (Santa Cruz), α-MYC (Cell Signaling Technology), α-p53 (Santa Cruz), α-SP1 (Abcam), and corresponding IgG control. Immunoprecipitated chromatin DNA was analyzed by real-time PCR using the following primers: TERT promoter (−3 kb) (for 5′-ACGATGGAGGCAGTCAGTCT-3′; rev 5′-T CCCCACACACTTCATGCTA-3′), TERT promoter (−300 bp) (for 5′-GTTCCCAGGGCCTCCA CATC-3′; rev 5′-GCGGAGAGAGGTCGAATCGG-3′), TERT intron 1 (0.4 kb) (for 5′-GAACC AGCGACATGCGGAGAGCA-3′; rev: 5′-AGCTCCTTCAGGCAGGACACCT-3′). Fold enrichment was calculated by comparing the amplification threshold (Ct) value of a given ChIP sample with that obtained in the IgG control at the same target locus. All experiments were performed at least 2–3 times in either duplicate or triplicate.
Quantification of γ-H2AX-positive cells was performed as previously described [65] with modifications. Briefly, cells were seeded onto 22-mm glass coverslips and 48 h later, the coverslips were washed in PBS, incubated in cytoskeleton buffer (10 mM piperazine-N,N′-bis[2-ethanesulfonic acid] [PIPES] pH 6.8, 100 mM NaCl, 300 mM sucrose, 3 mM MgCl2, 1 mM EGTA, 0.5% Triton X-100] for 5 min on ice. After several washes with ice-cold PBS, the cells were fixed in 4% paraformaldehyde for 20 min and permeabilized in 0.5% Triton X-100 solution for 15 min at room temperature. Cells were blocked with 2% BSA in PBS, incubated with primary antibody anti-phosphoH2AX (Ser139) (Millipore, JBW301) overnight at 4°C, washed three times with 1× PBS, and incubated with secondary antibody Cy3-conjugated sheep anti-mouse IgG (Sigma-Aldrich) for 1 h at room temperature. Cells were then washed, counterstained with 4′,6′-diamidino-2-phenylindole (DAPI), and mounted in 90% glycerol and 2% 1,4-diaza-bicyclo-(2,2,2)-octane (DABCO). Images were captured using a Zeiss AxioCam HRc camera, and 10 fields of cells were counted for each sample in duplicate.
|
10.1371/journal.pntd.0003816 | Differential Activation of Human Monocytes and Lymphocytes by Distinct Strains of Trypanosoma cruzi | Trypanosoma cruzi strains are currently classified into six discrete typing units (DTUs) named TcI to VI. It is known that these DTUs have different geographical distribution, as well as biological features. TcI and TcII are major DTUs found in patients from northern and southern Latin America, respectively. Our hypothesis is that upon infection of human peripheral blood cells, Y strain (Tc II) and Col cl1.7 (Tc I), cause distinct immunological changes, which might influence the clinical course of Chagas disease.
We evaluated the infectivity of CFSE-stained trypomastigotes of Col cl1.7 and Y strain in human monocytes for 15 and 72 hours, and determined the immunological profile of lymphocytes and monocytes exposed to the different isolates using multiparameter flow cytometry. Our results showed a similar percentage and intensity of monocyte infection by Y and Col cl1.7. We also observed an increased expression of CD80 and CD86 by monocytes infected with Col cl1.7, but not Y strain. IL-10 was significantly higher in monocytes infected with Col cl1.7, as compared to Y strain. Moreover, infection with Col cl1.7, but not Y strain, led to an increased expression of IL-17 by CD8+ T cells. On the other hand, we observed a positive correlation between the expression of TNF-alpha and granzyme A only after infection with Y strain.
Our study shows that while Col cl1.7 induces higher monocyte activation and, at the same time, production of IL-10, infection with Y strain leads to a lower monocyte activation but higher inflammatory profile. These results show that TcI and TcII have a distinct immunological impact on human cells during early infection, which might influence disease progression.
| Chagas disease remains a major public health problem in Latin America with over 13 million people infected. It is believed that the host immune response and genetic diversity of the parasite play an important role in the progression of Chagas disease, which presents a variety of clinical forms ranging from indeterminate to cardiac and digestive forms. Since parasite genetic diversity may influence the development of Chagas disease, our study aims to understand the immune response of human peripheral blood cells upon infection with two T. cruzi strains with different genetic backgrounds (Col cl1.7 – Tc I, and Y strain – TcII). Our study showed differences in the expression of cytokines and activation molecules between cells infected with strains from Tc I (Col cl1.7) and Tc II (Y strain). These data show the importance of parasite strain in the development of the host response early in infection, which may influence the clinical progression of Chagas disease.
| Human infection with the protozoan parasite, Trypanosoma cruzi, leads to Chagas disease, which presents as a spectrum of clinical forms, ranging from a relatively mild form (indeterminate), to a severe heart disease that affects approximately 30% of the infected individuals. Chagas disease is endemic to Latin America and the pathogenesis of Chagas heart disease is not clearly understood. A combination of host genetics, the host immune response and parasite factors seem to play important roles in the process [1] [2].
It has been demonstrated that patients with the indeterminate clinical form of Chagas disease display a predominantly modulatory immune environment, with higher production of the anti-inflammatory cytokine IL-10 [3] [4] [5] and IL-17 [6,7], which are produced by monocytes (IL-10) and T cell subsets (IL-10 and IL-17). On the other hand, the response observed in cardiac patients tends to be more inflammatory, with higher production of TNF-alpha and IFN-gamma, which are correlated with worse cardiac function [5,8].
Despite the clear polarity observed in the immune response of patients with different clinical forms, it is not possible to establish whether this is the primary cause of the development of distinct clinical forms. Genetic diversity of the parasite may have a great influence on different clinical outcomes [1]. Supporting this hypothesis, Vago et al. demonstrated that parasites isolated from heart or esophagus of Chagas patients display distinct genetic profiles [9].
T. cruzi strains are currently classified into six discrete typing units (DTU’s) named TcI to VI. It is known that these DTU’s have different biological and geographical features [10]. The DTU I is the most abundant of all T. cruzi DTU’s in the Americas and can be associated with sylvatic and domestic cycles. Despite its extensive distribution throughout the Americas, cases of Chagas disease caused by strains belonging to DTU I are concentrated in the north of South and Central America, with rare cases in the Southern Cone [11]. The DTU II is mostly associated with the domestic cycle and is mostly associated with chronic Chagas disease in South America [12–15]. After an outbreak of acute Chagas disease in Santa Catarina, Steindel et al. identified mixed TcI/TcII patterns in strains derived from Triatoma tibiamaculata, while strains isolated from patients were TcII [16]. Moreover, Pena et al. observed the selection of TcII strains after a mixed infection with TcI/TcII in murine and human macrophages [17].
Y strain of T. cruzi was isolated from a human host. Col cl1.7 was cloned from the Colombian strain, which was originally isolated from the blood of a chronic cardiac patient [18] and used in several studies since then [11,19]. These two strains represent the two major genetic groups of T. cruzi–Tc I (Col cl1.7) and Tc II (Y strain) [10] [19]. It was shown that Col and Y strain lead to different infection outcomes in experimental models. Duz et al. showed that dogs infected with Colombiana reach the parasitemia peak later than animals infected with Y strain, and that Y strain triggers a more intense immune response during the acute phase of infection in dogs in comparison with Colombiana strain [20]. Murine infection with Col cl1.7 and JG strain (which is Tc II), showed that animals infected with Col cl1.7 had a milder heart inflammation as compared to the JG strain [19].
Given the influence that the host immune response has on disease outcome, our goal was to determine whether infection by Y strain (Tc II) or Col cl1.7 (Tc I) had different effects on immunological characteristics of human monocytes and lymphocytes, which are key for establishing the immune response during infection. Our data showed that Col cl1.7 and Y strains lead to differential activation of monocytes and T cells, that are correlated with their profile of lower and higher virulence observed in animal models, respectively. Moreover, our data suggests mechanisms explaining how differences in parasite strains can lead to differences in human disease progression and outcome.
The donors included in our studies were non-chagasic healthy individuals (n = 9), as determined by negative specific serological test for Chagas disease. Individuals were from Belo Horizonte city, state of Minas Gerais, Brazil, with ages ranging between 23 and 34 years (average ± SD: 27±4.2). Five donors were males and 4 were females. We excluded from our study individuals with any chronic inflammatory disease, diabetes, heart and circulatory illnesses (including hypertension) or bacterial infections. All individuals included in this work were volunteers and provided written informed consent. This work was approved by the Ethical Committee of the Universidade Federal de Minas Gerais, under the protocol# ETIC077/06. Peripheral blood was collected from the donors by venipuncture.
Tissue culture derived trypomastigotes (TCT) of the Y strain and Col cl1.7 were isolated from infected monolayers of Vero cells. Vero cells were infected using five TCT/host cells and kept in RPMI enriched with 5% inactivated fetal calf serum (FCS), supplemented with antibiotics (penicillin at 500μ/mL and streptomycin at 0.5 mg/mL). After approximately 5 days, the TCT were collected from the supernatant, washed once by centrifugation with phosphate-buffered saline (PBS) pH 7.2 at 1000g for 10 min at 4°C and resuspended in RPMI to 6x107 TCT/mL. Parasites obtained in such manner were used for infecting adherent cells and peripheral blood cells from donors.
Adherent cells were used solely to confirm the infectivity of monocytes by the different strains using confocal microscopy. Peripheral blood mononuclear cells (PBMC) were purified as previously done by us [5]. Briefly, heparinized blood was diluted 1:1 with PBS and applied over a Ficoll gradient. The mixture was centrifuged for 40 min at 600g. PBMCs were collected at the interface between the plasma and the Ficoll. Cells were washed three times by centrifugation with PBS and resuspended in complete RPMI (RPMI supplemented 5% of human sera, antibiotics—penicillin at 500U/ml and streptomycin at 0.5 mg/ml—and 1mΜ of L-glutamine). To obtain adherent cells, 2x106 PBMC/well were plated on 13-mm round coverslips in complete RPMI and incubated for 3 hours at 37°C, 5% CO2. After incubation, non-adherent cells were removed by washing the wells with warm PBS and the adherent cells (monocytes) were used in infection experiments as described below. As previously determined by us, adherent cells obtained using this protocol are approximately 85% Cd11b+ or CD14+ [5].
The infection of monocytes (adherent cells) was performed as previously done by us [5]. Briefly, infection was performed over coverslips in duplicates. Parasites from Y or Col cl1.7 were added at a ratio of 10:1 TCT/monocytes and incubated for 3 hours. After the incubation period the monolayers were washed with PBS to remove extracellular parasites and re-incubated for 12 or 69 hours in complete RPMI, completing a total of 15 and 72 hours of culture. At the end of the culture time, cells were fixed by incubating the slides with 300ul of paraformaldehyde for 60 minutes at room temperature, washed three times with PBS and immunofluorescence was carried out by staining with 4’6’-diamino-2-phenylindole (DAPI). Briefly, coverslips containing the infected adherent cells were incubated with DAPI diluted 1:300 in PBS for 15 min at room temperature and mounting using Vectashield (Vector laboratories). Confocal analyses were performed using a Meta-510 Zeiss laser scanning confocal system running LSMix Software (Oberkochen, Germany) coupled to a Zeiss microscope using an oil immersion Plan-Apochromat objective (63X, 1.2 numerical aperture, Oberkochen, Germany).
Infection of whole blood was used for all experiments of surface molecule and cytokine expression analysis. For the infection of peripheral blood cells, trypomastigotes from Vero cultures, obtained as described above, were labeled with CFSE (carboxyfluorescein diacetate succinimidyl ester–Molecular Probes C1157) by using a protocol previously reported by us [5], with modifications. Briefly, 6.0 x 107 parasites were incubated with 5μM CFSE for 15 min at 37°C under 5% CO2. Labeled parasites were washed three times with cold PBS + 10% of inactivated fetal bovine serum by centrifugation at 1000g for 10 min at 4°C.
The infection was performed using 10 parasites/cell. Cells and parasites were incubated in suspension at 37°C in 5% CO2 for 3 hours with complete RPMI. After the incubation period the cells were washed by centrifugation with PBS at 600g for 10 min at 4°C to remove extracellular parasites. For the incubation of “15 hours” and “72 hours” we re-incubated the cultures for additional 12 and 69 hours respectively, after washing off the free parasites. Brefeldin A (1μg/ml) was added for the last four hours of infection in both groups (15 and 72 hours) to prevent protein secretion.
After 15 and 72 hours of incubation, the erythrocytes were lysed using RBC “Lysing buffer” (Bio Legend) at 20mL/1mL of peripheral blood. The tubes were incubated for 15 min at 20°C in the dark. After the incubation, cells were washed three times with PBS by centrifugation at 600g for 10 min at 4°C and resuspended in PBS to 107cells/ml. Cells were then immunostained and analyzed using multiparametric flow cytometry. 200.000 cells were incubated for 15 min at 4°C with different antibody combinations. Samples were washed three times in PBS-1% bovine serum albumin (BSA) and fixed by 20-min incubation with 2% formaldehyde solution. After removal of the fixation solution by centrifugation and washing once with PBS, we permeabilized the cells by incubation for 10 min with 0.5% saponin solution, centrifuged and incubated with antibodies to intracellular molecules for 30 min at 20°C. The antibodies to surface molecules used were: anti-CD4, anti-TLR-2 or anti-CD69 – labeled with PE; anti-CD14 – labeled with APC; anti-CD8, anti-HLA-DR, anti-CD80 – labeled with PE-Cy7; anti-CD86 – labeled with Pacific Blue and anti-CD4 labeled with APC-Cy7. For intracellular staining we used the following antibodies: anti-TNF-alpha, anti-IL-12/IL-23p40, anti-IL-10 and anti-Granzyme A. All antibodies were purchased from BioLegend, San Diego, CA, USA. After intracellular staining, cells were washed and resuspended in PBS and acquired using a FACSCanto II (Becton & Dickinson, San Jose, CA, USA). A total of 30,000 lymphocyte events were acquired and the parameters were analyzed in the monocyte or lymphocyte population. Lymphocyte analysis was done by gating the region occupied classically by lymphocytes in a size versus granularity plot, followed by gating in CD4+ or CD8+ cells. For monocytes, we first gated on CD14high cells in plot of size versus CD14 and further gated on CD14+CFSE+, CD14+CFSE- (Fig 1B). The analyses were performed using FlowJo 7.6.5 software (Tree Star Inc., Ashland, OR, USA).
We compared our results using One-Way Anova or Kruskal-Wallis test according to Kolmogorov-Smirnov normality test. All samples were submitted to rout test to identify outliers. Correlation analyses were made using Pearson’s correlation coefficient. All analyses were performed using Graph Pad Prism Software (La Jolla, CA, USA). Differences that returned p values equal or less than 0.05 were considered statistically significant from one another.
To determine the rate of infection with Y strain or Col cl1.7, we stained T. cruzi trypomastigotes with CFSE and infected peripheral blood cells from healthy donors, as described above. It is known that T. cruzi infects primarily monocytes [5], thus we analyzed the CD14+CFSE+ cells, which corresponds to the monocytes that were infected with trypomastigotes. Fig 1A shows that the efficiency of labeling parasites from Y or Col 1.7 strains with CFSE is similar and Fig 1B shows the gating strategy for CFSE+ cells (infected) as well as CFSE- cells (non-infected) that was used in our analysis of surface molecule and cytokine expression.
We observed a similar frequency of CD14+CFSE+ cells when the infection was performed with Y or Col cl1.7, after 15 or 72 hours of culture, comparing the different strains (Fig 1C). The intensity of infection on a cell per cell basis was also similar between the cells infected with the different strains (Fig 1D). Fig 1E shows representative analysis using fluorescent confocal microscopy, depicting infection of monocytes by parasites of either isolates stained with DAPI, confirming monocyte infection.
In order to access if there was a difference in monocyte activation after the infection with Y strain or Col cl1.7, we analyzed the expression of HLA-DR and TLR-2. HLA-DR is an important antigen-presenting molecule whose expression changes upon activation [21]. It is known that activation of TLR-2 is involved with activation of Rab-5, fusion with endosomes and phagocytosis of the trypomastigote form [22]. Since these changes occur early after activation, we evaluated the expression of HLA-DR and TLR-2 after 15 hours of infection. Our results show that monocytes infected with either T. cruzi strain (CD14+CFSE+ cells) express higher intensity of HLA-DR and TLR-2 compared with media control (Fig 2A and 2C). On the other hand, non-infected monocytes (CD14+CFSE- cells) did not show significant changes in the intensity of expression of these molecules as compared to media control (Fig 2B and 2D).
We investigated whether the expression of CD80 and CD86, monocyte ligands for co-stimulatory molecules, was modified after 72 hours of infection with Y strain or Col cl1.7. We also evaluated the expression of these molecules after 15 hours of infection and, although the results showed a similar trend as the ones observed in 72 hours, the frequencies and intensities were much lower (S1 Fig), indicating that the kinetics of expression of these molecules seems to be slower in response to T. cruzi infection. Thus, we chose to perform all analysis after 72 hours. Our results showed that CD14+CFSE+ monocytes infected by Col cl1.7 showed an increase in the intensity of expression of both co-stimulatory molecules, CD80 and CD86, as compared to media control (Fig 2E and 2G). Infection with Y strain did not have a significant effect on the expression of CD80 or CD86. Expression of CD80 or CD86 did not change in non-infected monocytes (CD14+CFSE-) (Fig 2F and 2H).
We questioned whether the monocyte activation triggered by infection had an influence on the expression of cytokines after 15 hours of infection with the Y strain or Col cl1.7. We observed a higher expression of TNF-alpha and IL-12/IL-23p40 in monocytes infected with the two strains compared to media control (Fig 3A and 3C, respectively). No changes were observed in non-infected monocytes (Fig 3B and 3D).
When we analyzed the expression of the anti-inflammatory cytokine IL-10 (Fig 3E and 3F), we observed an increase in infected and non-infected cells as compared to media, regardless of the strain. However, we observed a higher percentage of IL-10+ cells when cultures were infected by Col cl.1.7, compared to infection by the Y strain (Fig 3E).
We then evaluated the intensity of molecule expression by infected (CD14+CFSE+) and non-infected (CD14+CFSE-) monocytes exposed to either Y strain or Col cl1.7. Interestingly, it was observed that monocytes infected by Y strain or Col cl1.7 (CD14+CFSE+) had a more intense expression of HLA-DR, TLR2, CD80, CD86, IL-12/IL-12p40, TNF-alpha and IL-10 as compared to monocytes in the same cultures that were not infected by the parasite (CD14+CFSE-) (Table 1). These data suggests that direct contact and infection by the parasite is important to induce the phenotypic and functional changes seen in the infected cultures and that some changes in molecule expression in non-infected cells may be due to bystander effects.
Despite the low frequency of infection by Col 1.7 and Y strains in CD4 and CD8 lymphocytes (supporting material, S2 Fig), these cells have a major role in orchestrating the immune response against the parasite [2] and their activation depends on the interaction with infected monocytes. In order to determine if exposure to monocytes infected with Y strain and Col cl1.7 led to differences in lymphocyte activation, we analyzed the expression of the activation molecule CD69 and cytokines in lymphocytes after 72 hours of infection with both lineages. Our data showed that infection with either lineage led to a higher expression of the activation molecule CD69 by CD4+ and CD8+ T lymphocytes as compared to media control (Fig 4A and 4B).
We then evaluate the expression of the pro-inflammatory cytokine TNF-alpha. This cytokine is known to be important in the control of T. cruzi during the early stages of infection acting synergistically with IFN-gamma and activating monocytes to produce nitric oxide [23–26]. However, the persistence of a TNF-alpha-rich, pro-inflammatory environment, is associated with pathology during the chronic phase [2]. While no change was observed in the frequency of CD4+TNF-alpha+ T cells after 72 hours of infection with either T. cruzi strain (Fig 4C), our data showed an increase in the frequency of CD8+TNF-alpha+ T lymphocytes (Fig 4D).
We observed a higher expression of IL-17 in CD4+ T lymphocytes after infection with both strains compared to media control (Fig 4E). However, the increase expression of IL-17 in CD8+ T lymphocytes was observed only after infection with Col cl1.7 (Fig 4F).
IL-10 has a crucial role in orchestrating the immune response during T. cruzi infection, modulating the immune response in chronic disease [2–4]. The infection with either strain did not alter the percentage of expression of IL-10 by CD4+ T lymphocytes (Fig 4G).
Similar to TNF-alpha expression by CD4+ T cells after 72 hours, infection with either strain did not induce an increase expression of granzyme A (Fig 4I). However, we observed a higher expression of granzyme A by CD8+ T lymphocytes after infection with either strain compared to media control (Fig 4J).
We performed correlation analysis to determine whether the increase in expression of Granzyme A by CD8 T lymphocytes was associated with the expression of TNF-alpha by these cells. Our data showed a positive correlation between the expression of TNF-alpha and the expression of Granzyme A. Interestingly, this correlation was only observed after the infection with Y strain and not in media or infection with Col cl1.7 (Fig 5).
A triad of factors involving host genetics, immune competence of the affected population, and genetic diversity of the parasite influence the outcome of Chagas disease [1]. Our goal was to evaluate the effects of infection with either the Y (Tc II) or Col cl1.7 (Tc I) strains on immunological characteristics of human peripheral blood cells. These two strains were isolated from humans and represent the two major genetic groups of T. cruzi–Tc I (Col cl1.7) and Tc II (Y strain) [10] [19].
Our data showed that the frequency of infected cells was similar when comparing Y strain and Col cl1.7. The intensity of infection was measured by CFSE (Fig 1D) and also by DAPI staining (Fig 1E) and showed that the intensity of infection on a cell-per-cell basis did not change when comparing infection with Col cl 1.7 and Y strains. We next questioned if the similar rate of infection with both Y strain and Col cl1.7 could differently activate monocytes. We analyzed the expression of activation-related molecules (HLA-DR, TLR-2, CD80 and CD86) in infected (CFSE+) and non-infected (CFSE-) monocytes, from cultures exposed to the different strains. This separation allows us to determine the effects of direct contact with the parasite versus a bystander effect in the expression of the molecules. We observed that infection with both strains led to a higher intensity of HLA-DR and TLR-2 expression by infected monocytes (CD14+CFSE+), as compared to media controls; these increases of intensity did not occur in non-infected monocytes (CD14+CFSE-) from the same cultures. It is known that T. cruzi activates TLR-2 through GPI-anchored mucin-like glycoproteins (tGPI-mucin) and this activation triggers an inflammatory response [27]. Moreover, it has been shown that TLR-2 and TLR-9 are important for parasite control in the early phases of infection [28]. It is also known that TLR-2 is involved with the internalization of T. cruzi in murine macrophages via activation of Rab5 [22]. Thus, the increase in the intensity of TLR-2 only in infected monocytes supports the hypothesis that activation of TLR-2 is important in T. cruzi phagocytosis [22]. We also investigated the expression of the ligands for co-stimulatory molecules, CD80 and CD86. These co-stimulatory molecules provide the second signal necessary for T cell activation [29] and their expression has also been associated with activation of monocytes after T. cruzi infection [30]. Surprisingly, while infection with Y strain and Col cl1.7 both led to an increase in the intensity of expression of HLA-DR and TLR-2, the two isolates affected differently the expression of CD80 and CD86. Infection with Col cl1.7, but not with Y strain, led to an increase in the expression of CD80 and CD86. This suggests a higher activation of monocytes after 72 hours of infection when infected by Col cl1.7, as compared to Y strain.
Several studies have reported the importance of macrophage activation in experimental infection with T. cruzi. In the acute phase, after interaction with T. cruzi, macrophages produce inflammatory cytokines such as IL-12 and TNF-alpha, which activate the production of IFN-gamma by NK cells. The IFN-gamma produced, together with TNF-alpha, activates macrophages to produce oxygen derivatives to eliminate the parasite [23–26]. On the other hand, the anti-inflammatory cytokine IL-10 appears to be detrimental in the early infection. Experimental data suggest that the expression of anti-inflammatory cytokines inhibit IFN-gamma, decreasing the trypanocidal activity of macrophages [31,32]. However, the maintenance of a phenotype with high expression of inflammatory cytokines compared to the expression of anti-inflammatory cytokines is associated with progression to the cardiac clinical form during the chronic phase of human disease [2]. Our data showed an increase in the percentage of TNF-alpha+ monocytes after the infection with Y strain and Col cl1.7. This increase was observed only in infected monocytes, and the intensity of expression of TNF-alpha was also greater in infected monocytes compared to bystander cells.
The percentage of monocytes expressing IL-10 was higher after the infection by the two strains in infected and non-infected monocytes. However, an even higher percentage in IL-10+ monocytes was observed in cultures infected with Col cl1.7, as compared to Y strain. It is known that the expression of IL-10 is crucial for the orchestration of the immune response during chronic Chagas disease. A higher frequency of monocytes producing IL-10 can be found in indeterminate as compared to cardiac patients [5]. Interestingly, in this study, the expression of IL-10 in monocytes was the only parameter that allowed for distinguishing between the two strains, where we observed a higher IL-10 expression after infection with Col cl1.7 compared to infection with Y strain. This result shows that although monocytes infected with both strains show similar activation, the production of cytokines differ between them. The higher expression of IL-10 by monocytes infected with Col cl1.7 suggests that this isolate is able to induce a more balanced immune response, as compared to Y strain. This is consistent with the fact that Col induces a more mild infection than does Y strain in experimental models [19] [20]. This is an important finding, as it suggests that strains of parasite that are able to induce a more balanced response, with TNF-alpha and IL-10 production, may lead to a better infection outcome. The fact that a balanced immune response may be beneficial is supported by the findings that showed that while cardiac patients have a predominantly inflammatory profile (higher expression of TNF-alpha and IFN-gamma), indeterminate patients, despite producing inflammatory cytokines, have a more anti-inflammatory profile (due to high IL-10 expression) [33] [5].
It is known that IL-12/IL-23p40 triggers the expression of IFN-gamma, which is important for the control of T. cruzi in experimental models [34]. We observed an increased expression of IL-12/IL-23p40 only in infected monocytes after infection with both strains, indicating activation, and suggesting that the production of this cytokine early on may be critical for parasite control, regardless of the strain.
Our data demonstrated a higher activation of infected monocytes as compared to non-infected cells in the same cultures, suggesting that a direct contact with the parasite is required for better activation. This was confirmed by the fact that cells infected by either of the two strains have more intense expression of HLA-DR, TLR-2, CD80, CD86, as well as cytokines IL-12/IL23p40, TNF-alpha and IL-10 when compared with bystander cells. Carvalho and colleagues in 2008 observed that dendritic cells (DCs) infected with Leishmania braziliensis were the main cell type responsible for the expression of TNF-alpha, while bystander DCs have enhanced HLA-DR expression [35]. Our results do not corroborate these findings, showing that monocytes infected with two strains of T. cruzi exhibit greater activation and are also the main producers of cytokines compared to bystander cells. Thus, T. cruzi-infected monocytes appear to be responsible for the activation of lymphocytes and cytokine expression, while bystander cells appear to be less responsive.
The presence of activated T cells has been observed in the endomyocardial tissue and peripheral blood of patients infected with T. cruzi in acute and chronic phase of Chagas disease [36] [37] [38]. Given the observed activation of monocytes and their importance in T cell activation, we analyzed the activation state and the expression of cytokines and granzyme A by CD4+ and CD8+ T cells. We first showed that CD4+ and CD8+ T lymphocytes display higher expression of CD69 after infection with Y strain and Col cl1.7, showing that both strains induce T cell activation. We next analyzed the expression of cytokines to determine if the activation was also correlated with cytokine production. No difference in IL-10 expression was seen after infection with either T. cruzi strains in CD4+ T cells, as compared to media control. However, expression of TNF-alpha by CD8+ T lymphocytes was increased following infection with both strains. The same happened with the expression of Granzyme A. It is not surprising that CD8+ T cells seem to be more functionally responsive to the phenotypic and functional changes than CD4+ T cells, considering that the stimulation was performed with live trypomastigotes, which favors activation via class I molecules, since T. cruzi is an intracellular parasite [39]. Despite that, activation via class II is also possible, since soluble proteins of the parasite, like TS proteins, are able to activate CD4+ T lymphocytes, which corroborates with the higher expression of CD69 also observed in CD4+ cells [40].
A protective role of IL-17 in experimental T. cruzi infection was recently reported [41] [42]. In chronic chagasic patients, high levels of IL-17 are related to better clinical prognosis [6,7]. Here we observed a higher expression of IL-17 in CD4+ T cells after the infection with both strains. Interestingly, only the infection with Col cl1.7 led to a higher expression of IL-17 in CD8+ T lymphocytes. Erdmann et al. observed that IL-17 stimulates macrophages to phagocyte trypomastigotes, trapping T. cruzi in endosomal/lysosomal compartments and enhancing exposure time to antimicrobial effectors of the macrophages that subsequently led to eradication of parasites [43]. The fact that IL-10 and IL-17 both have been associated with the indeterminate form [5] [7] suggests that the induction of IL-17 by Col cl1.7 may also be associated with the more mild infection observed in experimental models, as compared to Y strain.
Expression of TNF-alpha and Granzyme A also increase only in CD8+ T cells. The increased expression of IL-17, TNF-alpha and Granzyme A in CD8+ T cells was an interesting finding, since several data have shown an important role for CD8+ T cells in murine models, as well as in human disease [44] [45]. Interestingly, it has been shown that Granzyme A knockout mice are more susceptible to infection by T. cruzi [46], suggesting a role for this molecule in control of parasitemia. Moreover, expression of granzyme is high in CD8+ T cells found in the inflammatory infiltrate of patients with severe chronic chagasic cardiomyopathy, which is also rich in TNF-alpha+ cells [36]. We then asked whether the expression of the cytotoxic molecule Granzyme A by CD8 T cells was related to the expression of TNF-alpha. We observed a positive correlation between the expression of TNF-alpha and the expression of Granzyme A only after the infection with Y strain.
Taken together, our results show that infection of human cells with the Col cl1.7 leads to a higher expression of CD80 and CD86, as well as of IL-17, favoring monocyte activation. In addition, the observed higher expression of IL-10 by cells infected with Col cl1.7 might be important to avoid tissue pathology in the acute infection, favoring host survival. These data are consistent with the results observed in experimental models, in which Col parental strain and its clones displays low virulence and are able to induce chronification of infection [19,20]. On the other hand, Y strain led to a more inflammatory profile with high TNF and Granzyme expression, which might be associated with pathology. This is also consistent with the high virulence data observed in experimental models [20]. An important data is that TcI and TcII are found more frequently in north and south of Latin America, respectively, and pathology associated with Chagas disease is more frequent in southern than northern Latin America [1]. While a clear association between TcII and greater pathology was not directly performed, this suggests that TcII is at least more frequent in an area with more pathology than TcI.
Transcritome analysis of myoblast cell line infected with different strains of T. cruzi belonging to Tc I and Tc II, showed that different strains lead to several changes in gene transcription, and that these changes were significantly different amongst strains [47]. Interestingly, the Y strain, also used in our study, led to the least changes in the myoblasts transcriptome profile [47]. The results presented here show, for the first time, the mechanisms by which Y and Col cl1.7 strains influence differently the host’s immune response, while clearly showing the importance of the parasite strain in shaping the host response early on, which might influence disease outcome at later times. Thus, analysis of these parameters in individuals with acute infection of Chagas disease might bring valuable information for patient follow up and care.
|
10.1371/journal.ppat.1006405 | Elevated catalase expression in a fungal pathogen is a double-edged sword of iron | Most fungal pathogens of humans display robust protective oxidative stress responses that contribute to their pathogenicity. The induction of enzymes that detoxify reactive oxygen species (ROS) is an essential component of these responses. We showed previously that ectopic expression of the heme-containing catalase enzyme in Candida albicans enhances resistance to oxidative stress, combinatorial oxidative plus cationic stress, and phagocytic killing. Clearly ectopic catalase expression confers fitness advantages in the presence of stress, and therefore in this study we tested whether it enhances fitness in the absence of stress. We addressed this using a set of congenic barcoded C. albicans strains that include doxycycline-conditional tetON-CAT1 expressors. We show that high basal catalase levels, rather than CAT1 induction following stress imposition, reduce ROS accumulation and cell death, thereby promoting resistance to acute peroxide or combinatorial stress. This conclusion is reinforced by our analyses of phenotypically diverse clinical isolates and the impact of stochastic variation in catalase expression upon stress resistance in genetically homogeneous C. albicans populations. Accordingly, cat1Δ cells are more sensitive to neutrophil killing. However, we find that catalase inactivation does not attenuate C. albicans virulence in mouse or invertebrate models of systemic candidiasis. Furthermore, our direct comparisons of fitness in vitro using isogenic barcoded CAT1, cat1Δ and tetON-CAT1 strains show that, while ectopic catalase expression confers a fitness advantage during peroxide stress, it confers a fitness defect in the absence of stress. This fitness defect is suppressed by iron supplementation. Also high basal catalase levels induce key iron assimilatory functions (CFL5, FET3, FRP1, FTR1). We conclude that while high basal catalase levels enhance peroxide stress resistance, they place pressure on iron homeostasis through an elevated cellular demand for iron, thereby reducing the fitness of C. albicans in iron-limiting tissues within the host.
| The pathogenic yeast Candida albicans faces multiple challenges within its human host. These include the need to protect itself against the toxic oxidants used by the host to kill invading microbes, and the need to scavenge iron, an essential micronutrient that is limiting in certain tissues. The iron-containing enzyme, catalase, detoxifies hydrogen peroxide, thereby playing a major role in protecting C. albicans against reactive oxygen species and neutrophil killing. Indeed, we show that high basal catalase expression increases the resistance of this yeast to oxidative and combinatorial (oxidative plus cationic) stresses. Yet, rather than enhancing the virulence of C. albicans as had been predicted, high basal catalase expression decreases fungal colonisation in certain iron-limiting tissues. Furthermore, we demonstrate that catalase inactivation does not significantly perturb the virulence of C. albicans in models of systemic infection. We also show that ectopic catalase expression increases the demand for iron in C. albicans, thereby reducing the fitness of this pathogen in the absence of stress under iron-limiting conditions. Therefore, high basal catalase expression is a double-edged sword: it enhances the fitness of C. albicans in the presence of stress, but reduces fitness in the absence of stress. This explains why catalase overexpression reduces rather than enhances virulence.
| Of the circa three million fungal species that are thought to inhabit our planet [1], only a relatively small number have been reported to cause infections in humans. (About 400 species are described in the Atlas of Clinical Fungi [2].) Nevertheless, there is an increasing awareness that these fungal pathogens impose a major burden on human health worldwide [3]. These clinically important fungi generally share common features that promote colonization of their human host, such as the thermotolerance that permits growth at body temperatures. These common features include relatively robust stress responses, which mitigate against the stresses imposed by host immune defences [e.g. 4–6]. They also include the ability to scavenge essential micronutrients, such as iron, from their host [7–10].
Iron is an essential micronutrient that is required for the functionality of key ferroproteins and haem proteins. However, excess iron is toxic because ferrous ions promote the Fenton reaction which produces highly toxic hydroxyl radicals [11], and therefore host and pathogen alike must tightly regulate their acquisition, storage and release of iron. Consequently, the levels of free ion are vanishingly low in some host niches [12]. Furthermore, following infection the host activates the process of nutritional immunity in an effort to limit iron availability for the invading microbe [10,12]. Fungal pathogens respond to this iron limitation by down-regulating genes encoding iron-containing proteins and upregulating efficient iron scavenging mechanisms [13–17]. In Candida albicans this response includes the induction of genes encoding ferric reductases (e.g. CFL5, FRP1), high affinity iron permeases (e.g. FTR1) and proteins involved in iron assimilation (e.g. FET3) [15]. This response allows the fungus to counter the changes in iron homeostasis within the host that are triggered by systemic candidiasis [10].
Fungal pathogens activate oxidative stress responses when they come in contact with the host [18–22], and these responses promote resistance to phagocytic attack and fungal virulence [5,23–26]. In an attempt to clear invading fungal pathogens, host neutrophils and macrophages phagocytose the fungal cells and subject them to a battery of reactive oxygen species (ROS) that damage proteins, DNA and lipids, and can induce programmed cell death [27]. The impact of ROS is augmented when combined with a cationic stress, and this synergistic impact of combinatorial oxidative and cationic stresses is thought to contribute to the potency of human neutrophils [28,29]. C. albicans cells respond to oxidative stress by inducing functions that detoxify the ROS, repair the oxidative damage, synthesize antioxidants and restore redox homeostasis. This includes the induction of genes encoding catalase (CAT1), superoxide dismutases (SOD), glutathione peroxidases (GPX) and components of the glutathione/glutaredoxin (GSH1, TTR1) and thioredoxin (TSA1, TRX1, TRR1) systems [6,30–32]. In particular, CAT1 mRNA levels are strongly induced by oxidative stress [30,33]. However, C. albicans cells are unable to activate a normal transcriptional response to oxidative stress when subjected to combinatorial oxidative plus cationic stress or acute peroxide stress, and this contributes to the lethality of these types of stress [28,29].
Catalase (Cat1) plays a major role in protecting C. albicans against peroxide stress [28,29]. This iron-requiring enzyme, which has been well-characterised structurally [34], belongs to a superfamily of heme peroxidases and catalases that are conserved across bacteria, plants, fungi and animals [35]. Catalase catalyses the conversion of hydrogen peroxide (H2O2) to water. C. albicans cells rapidly detoxify extracellular H2O2 following exposure to an acute peroxide stress, and this detoxification is mainly dependent on catalase (CAT1) [28].
We showed previously that ectopic expression of catalase using the ACT1 promoter (ACT1p-CAT1) protected C. albicans from acute oxidative and combinatorial stresses [28]. More recently, Jesus Pla’s group has confirmed that catalase overexpression protects C. albicans against peroxide stress [36]. They also demonstrated that high catalase levels provide protection against antifungal drugs. These observations raise an interesting conundrum: if catalase overexpression confers effects that might be expected to promote host colonisation, why has C. albicans not evolved to express high basal levels of catalase? We address this in this study. We show that while high basal catalase levels enhance the fitness of C. albicans in the presence of oxidative and combinatorial stresses, these high catalase levels reduce fitness in the absence of stress. We also reveal the molecular basis for this fitness defect. Our observations suggest a partial explanation for the lack of emergence of catalase overexpression during the evolution of this major fungal pathogen. We also show that, in contrast to the prevailing view [23], the virulence of C. albicans is not compromised by catalase inactivation.
We demonstrated previously that ectopic expression of catalase from an ACT1 promoter-CAT1 fusion (ACT1p-CAT1) reproducibly protected C. albicans against acute peroxide stress (5 mM H2O2) and combinatorial stress (5 mM H2O2 plus 1 M NaCl) [28]. Subsequently we noted that the stress resistance of ACT1p-CAT1 cells declined over time (S1 Fig). Therefore, we constructed new C. albicans strains in which catalase expression is regulated by the doxycycline conditional tetON promoter [37–39]. Control strains were made by transforming congenic wild-type (CAT1) and catalase null strains (cat1Δ) with empty tetON vectors. Test strains were made by integrating a tetON-CAT1 plasmid into the genome of the cat1Δ null mutant. We refer to these strains, which all have the same genetic background (Materials and Methods; S1 Table), as wild-type (CAT1), null (cat1Δ) and tetON-CAT1 strains, respectively. Three isolates were generated for each strain type. For each strain type, the isolates displayed similar stress phenotypes (below).
First we tested Cat1 expression levels in wild-type (CAT1), null (cat1Δ) and tetON-CAT1 cells. Catalase levels were induced in response to oxidative stress in wild-type (CAT1) cells, and were undetectable in cat1Δ cells (Fig 1). Catalase levels in these strains were not affected by doxycycline addition. In contrast, catalase levels were strongly induced by doxycycline in tetON-CAT1 cells (red bars, Fig 1). Significantly, wild-type cells express significant basal levels of catalase in the absence of stress (Fig 1), as we reported previously [33]. Catalase levels in doxycycline-treated tetON-CAT1 cells were higher than these basal levels (Fig 1).
We then compared the stress resistance of wild-type, null and tetON-CAT1 cells (Fig 2A & S1 Fig). As expected [23,28,40], wild-type (CAT1) cells displayed modest resistance to an oxidative stress (H2O2) and a combinatorial stress (H2O2 plus NaCl), whereas the null mutant (cat1Δ) was sensitive to both types of stress. These phenotypes were not affected by the presence or absence of doxycycline (Fig 2A). In the absence of doxycycline, the tetON-CAT1 strains were sensitive to both oxidative and combinatorial stress, reflecting their null cat1Δ background. When these strains were pre-grown with doxycycline, they displayed enhanced oxidative and combinatorial stress resistance (Fig 2A). This reinforces our earlier conclusion [28] that elevated basal CAT1 expression levels protect C. albicans cells against a sudden and acute oxidative or combinatorial stress.
Interestingly, the tetON-CAT1 strains were sensitive to both oxidative and combinatorial stress when pre-grown in the absence of doxycycline and the inducer was only provided when the stress was imposed (Fig 2A). We then measured the impact of individual and combinatorial oxidative (H2O2) and cationic (NaCl) stresses upon cell death by cytometric analysis of propidium iodide (PI) stained doxycycline-grown cell populations (Fig 2B). Relative to the control wild-type strain, cat1Δ null cells were more sensitive, and tetON-CAT1 cells were more resistant to these oxidative and combinatorial stresses. Compared to the control wild type cells, doxycycline-treated tetON-CAT1 cells displayed 9-fold less cell death following exposure to the oxidative stress, and 2.5-fold less death after the combinatorial stress (Fig 2B).
This correlated with a reduction in internal ROS accumulation following stress imposition by tetON-CAT1 cells relative to the wild-type and cat1Δ cells (Fig 2C). The accumulation of intracellular ROS was 2.6-fold lower in doxycycline-treated tetON-CAT1 cells after the peroxide stress, and 1.5-fold lower following the combinatorial stress, compared to the wild type control (Fig 2C). Taken together, our data indicate that cells with low catalase levels at the point of stress imposition are more sensitive to peroxide than cells with high catalase levels. This suggests if catalase levels are low at the point of stress imposition, the dynamics of catalase induction are too slow to permit the normally rapid clearance of peroxide [28] and to prevent ROS-mediated cell death [27]. The data indicate that C. albicans cells require high basal levels of catalase at the time of stress imposition if they are to survive an acute oxidative or combinatorial stress.
C. albicans clinical isolates display a high degree of natural variation [41,42]. We exploited this to select strains that display relatively low levels of oxidative stress resistance. A diverse range of C. albicans clinical isolates (65 in total) from different epidemiological clades and from different patient colonisation sites were subjected to a robotic screen in which they were plated on YPD containing different peroxide concentrations (Fig 3A). All of the isolates tested displayed a degree of resistance to this stress, showing some growth at 3.2 mM H2O2. However, some isolates were more resistant to peroxide, displaying robust growth at 6.4 mM H2O2, whereas sensitive strains were unable to grow at this H2O2 concentration. We selected a subset of four sensitive isolates and three resistant isolates (which included SC5314, the clinical isolate from which most laboratory strains are derived), and compared the basal CAT1 expression levels in these isolates to a standard laboratory strain (CAI4 containing CIp10 (URA3)). Basal CAT1 mRNA levels were lower in the oxidative stress sensitive isolates tested (Fig 3B), and furthermore, the basal levels of the enzyme were also lower in these isolates (Fig 3C). These data are consistent with the idea that elevated basal catalase levels promote oxidative stress resistance in C. albicans.
Next we examined how a subset of cells within an apparently homogeneous population of C. albicans cells can survive an acute oxidative stress [28,36,43]. Based on the above observations, we reasoned that this might be partly explained by stochastic variation in basal catalase levels between individual cells in such a population. To test for potential population heterogeneity in basal catalase levels we generated a strain in which both CAT1 alleles were tagged with GFP (CAT1-GFP/ CAT1-GFP) to express a Cat1-GFP fusion protein. Western blotting revealed a Cat1-GFP protein of the expected mass in these cells (approximately 80 kDa: Fig 4A), and the GFP fluorescence was located in punctate spots (Fig 4B), consistent with the peroxisomal localisation of catalase in C. albicans [44]. We then compared the oxidative stress resistance of the CAT1-GFP strain with congenic control wild-type (CAT1/CAT1), heterozygous (CAT1/ cat1Δ) and null (cat1Δ/cat1Δ) strains. The CAT1-GFP strain was as resistant to oxidative stress as the wild-type control (Fig 4C), indicating that the CAT1-GFP alleles are functional.
We then examined the basal levels of GFP fluorescence in unstressed populations of exponentially growing C. albicans CAT1-GFP cells by flow cytometry. As predicted, these genetically homogeneous cell populations displayed heterogeneity with respect to their Cat1-GFP expression levels (Fig 4D & S2 Fig). Using flow cytometry, we selected cells of similar size, sorted cells that display relatively low levels of Cat1-GFP from those expressing high levels (Fig 4D & S2 Fig), and then plated them onto media containing a range of H2O2 concentrations. Cells expressing relatively high levels of Cat1-GFP were more resistant to peroxide stress (Fig 4E). When an analogous experiment was performed with cells expressing a control gene (ACT1-GFP), stochastic differences in ACT1-GFP expression did not affect oxidative stress resistance (S3 Fig). These observations reinforce our conclusion that high basal levels of catalase promote oxidative stress resistance. Furthermore, this confirms that C. albicans cell populations display stochastic variation in their basal CAT1 expression levels, and that this contributes to the survival of a subset of C. albicans cells following an acute oxidative stress.
We tested whether high basal catalase levels affect the ability of C. albicans to colonise different tissues during systemic infection. At first we reasoned that the elevated oxidative stress resistance conferred by high basal catalase levels (above) might enhance host colonisation. To test this we compared directly the three isolates for wild-type (CAT1), null (cat1Δ) and tetON-CAT1 strains (nine in total) using a barcode sequencing (barseq) strategy. The C. albicans strains were pre-grown separately in the presence or absence of doxycycline. Approximately equal amounts of the nine doxycycline-treated strains were mixed and used to induce disseminated candidiasis in doxycycline-treated mice. In parallel, the nine untreated control C. albicans strains were mixed and used to infect untreated mice. Mice from each group were culled after four days, and the fungal cells harvested from their kidneys, livers, spleens and brains. Barseq was then performed on genomic DNA from these fungal populations to determine the relative proportion of each C. albicans strain in each tissue. We observed significant differences between the wild-type (CAT1) and tetON-CAT1 strains in their ability to colonise certain tissues (Fig 5). Doxycycline-treated tetON-CAT1 cells were less able to colonise the kidney and brain than the control untreated tetON-CAT1 cells, but this was not the case in the liver and spleen. This effect was observed for tetON-CAT1-1 cells, but not for the other two tetON-CAT1 isolates (4 and 10: S1 Table). This correlated with a reduction in catalase levels in these isolates (S4A Fig) and a corresponding loss of phenotype (S4B Fig). Therefore, like ACT1p-CAT1 cells (above; S1 Fig), isolates 4 and 10 appeared to have lost their phenotype over time. Taken together, our data indicate that, contrary to our initial prediction, high basal catalase expression levels appear to compromise, rather than enhance, the ability of C. albicans to colonise certain tissues.
To our surprise, we did not observe any significant differences between the wild-type (CAT1) and null (cat1Δ) strains in their ability to colonise the host (Fig 5). All of the wild-type and null isolates displayed similar levels of colonisation. This indicated that cells lacking catalase can infect the host—a conclusion that contrasts with the prevailing view that C. albicans cat1Δ null cells display attenuated virulence [23,40]. We reasoned that cat1Δ cells might be able to colonise host tissues if they are co-infected with CAT1 and tetON-CAT1 cells. For example, cat1Δ null cells might be rescued via a “cheater” or “bystander” effect [45,46], whereby catalase expressing cells protect the null mutant against local peroxide stress.
We tested this by comparing the virulence of our wild-type (CAT1) and null (cat1Δ) strains separately in the three-day murine model of systemic candidiasis [47]. We observed no significant difference between the wild-type or mutant strains with respect to their fungal burdens in the kidneys, and the strains induced similar levels of weight loss in mice, yielding similar outcome scores that displayed no significant difference (Fig 6A). This observation reinforced the idea that inactivating CAT1 does not attenuate the virulence of C. albicans.
Wysong and co-workers observed a virulence defect for cat1Δ cells using a long-term mouse model of systemic candidiasis [23]. Therefore, it seemed possible that our short-term and their long-term model of systemic infection might yield differing outcomes for C. albicans cat1Δ cells. To test this we re-examined the virulence of our wild-type (CAT1) and null (cat1Δ) strains in mice over 14 days. (We were unable to access the strains used by Wysong and co-workers [23]. Hence we could not perform a direct comparison with their mutant.) No major difference in the virulence of wild-type and cat1Δ cells was observed using a long term infection model (p = 0.074: Fig 6B). We also compared our wild-type (CAT1), null (cat1Δ) and tetON-CAT1 strains in Galleria mellonella, observing no significant difference in their virulence in this invertebrate model of systemic candidiasis (p = 0.68: Fig 6C).
The cat1Δ mutant generated by Wysong and co-workers had the URA3 marker inserted at the cat1 locus (cat1::URA3) [23]. In contrast, in our cat1Δ mutant URA3 was reintroduced at the RPS1 locus using the CIp10 plasmid backbone [48]. After the study of Wysong and co-workers was published [23], URA3 position effects were found to influence C. albicans virulence, and reinsertion of URA3 at RPS1 using CIp10 was shown to overcome these effects [49]. We conclude that CAT1 inactivation does not significantly attenuate the virulence of C. albicans.
It has been reported that catalase null mutants do not display significantly higher sensitivities to neutrophil killing [5]. Once again, these experiments were performed with a cat1Δ null mutant in which URA3 was integrated at the CAT1 locus (cta1Δ::loxP-URA3-loxP: [5]). Therefore, in light of our findings (above), we retested neutrophil killing using our new cat1Δ strain in which URA3 is integrated at the RPS1 locus. We tested the strains separately to exclude potential cheater effects [45,46]. We observed that, following exposure to human neutrophils, our new cat1Δ strain displayed significantly reduced survival compared to the congenic wild-type control (Fig 7). This strengthens the observation of Miramon and co-workers, who reported a slight difference between cat1Δ and CAT1 cells that was not statistically significant [5]. Furthermore, we also observed a statistically significant difference in neutrophil killing between tetON-CAT1 cells that were pre-grown in the presence or absence of doxycycline (Fig 7). These data indicate that catalase promotes the resistance of C. albicans against neutrophil attack. We note that elevated basal levels of catalase did not enhance the resistance of C. albicans to neutrophil killing in our hands (Fig 7: compare wild type and doxycycline-treated tetON-CAT1 cells).
C. albicans cat1Δ cells are clearly sensitive to oxidative stress (Fig 2). However, in mixed populations they could conceivably be rescued by catalase expressing cells. Therefore, we tested whether cat1Δ cells act as cheaters by examining their fitness in mixed cultures alongside wild-type (CAT1) and tetON-CAT1 cells. The three barcoded for the null mutant, wild-type and tetON-CAT1 strains were pre-grown separately in the presence of doxycycline, mixed in approximately equal proportions, and then used to inoculate YPD cultures containing doxycycline. A parallel mixture of untreated barcoded strains was also prepared, and this untreated mixture used to inoculate YPD cultures without doxycycline. The relative fitness of each strain was then compared in the presence or absence of oxidative stress (5 mM H2O2), by comparing the relative abundance of each barcode over time in each culture by barseq. With one notable exception (discussed below), the three isolates for each strain type displayed similar behaviours (Fig 8).
In the absence of doxycycline and stress, the relative abundance of the wild-type (CAT1), null (cat1Δ) and tetON-CAT1 strains did not change significantly over the twelve hour period examined (Fig 8A). In contrast, in the absence of doxycycline but in the presence of stress, the abundance of cat1Δ and tetON-CAT1 cells rapidly declined in the population and these strains were rapidly outcompeted by the wild-type CAT1 strains (Fig 8B). The comparable behaviour for the cat1Δ the tetON-CAT1 cells under these conditions was entirely consistent with the negligible catalase levels in tetON-CAT1 cells without doxycycline induction (Fig 1). These data strongly reinforce the view that catalase is vital for peroxide stress resistance in C. albicans [5,23,33,36,40]. Our data also show that cat1Δ cells do not act as cheaters: they are not rescued by catalase expressing cells under peroxide stress conditions (Fig 8B).
In the presence of doxycycline in the presence of stress, the tetON-CAT1 cells rapidly outcompeted the null (cat1Δ) cells (Fig 8D). This again highlighted the peroxide sensitivity of cat1Δ cells. Significantly, the tetON-CAT1 cells also out-competed wild-type (CAT1) cells (Fig 8D), confirming directly that ectopic catalase expression enhances oxidative stress resistance (Fig 2) [28,36]. Therefore, elevated basal catalase levels increase the fitness of C. albicans cells in the presence of peroxide stress.
Interestingly, in the presence of doxycycline but in the absence of peroxide stress, there was a decrease in the abundance of tetON-CAT1-01 cells in the population over the twelve hour time-course, relative to the wild-type (CAT1) and null (cat1Δ) cells (Fig 8C). This suggested that ectopic CAT1 expression might render C. albicans cells less fit in the absence of stress.
Doxycycline-treated C. albicans tetON-CAT1 cells appeared to display a fitness defect in the absence of stress (Fig 8C). We tested this further by examining biomass formation on YPD (final OD600) (Fig 9A). All of the strains displayed similar growth in the absence of doxycycline, and the wild-type (CAT1) controls remained unaffected by doxycycline. However, the growth of tetON-CAT1 cells decreased in the presence of doxycycline, reinforcing the view that elevated catalase levels reduce fitness in the absence of stress.
Catalase is a ferroprotein [34] expressed at relatively high basal levels in C. albicans (approximately 1.5 x 105 molecules per cell [33]). In bacteria, catalase overexpression has been reported to affect the requirement for iron [50]. Therefore, we reasoned that the fitness defect conferred by high basal catalase levels in C. albicans might be mediated by an elevated cellular demand for iron. Hence, we tested whether iron supplementation can suppress this fitness defect. Growth of tetON-CAT1 cells was measured in YPD containing doxycycline supplemented with different concentrations of ferric ions (Fig 9B). These data indicate that the fitness defect caused by ectopic catalase expression can be completely suppressed by iron supplementation. This suppression was due to the improved growth of doxycycline-treated tetON-CAT1 cells in the presence of iron (S5A Fig). We also showed that iron supplementation suppresses the reduced fitness of doxycycline-treated tetON-CAT1 cells in direct competition experiments with wild type (CAT1) cells (S5B Fig).
These observations suggested that high basal catalase expression increases the cellular demand for iron in C. albicans. To test this further we examined the impact of ectopic catalase expression upon key genes involved in iron assimilation and homeostasis: CFL5 (encoding a ferric reductase that is induced in low iron), FET3 (encoding a copper oxidase that is required for growth in low iron), FRP1 (encoding a ferric reductase that is induced by iron chelation) and FTR1 (encoding a high-affinity iron permease that is required for growth in low iron). All of these genes are targets of the iron-responsive transcriptional activator Sef1 [15]. CFL5, FET3, FRP1 and FTR1 transcript levels were measured relative to the ACT1 mRNA internal control in tetON-CAT1 cells grown in the presence and absence of doxycycline. Their levels were then normalised against those in doxycycline-treated wild type (CAT1) cells to exclude any potential effects of this treatment on these transcripts [51]. All four iron-responsive transcripts were strongly induced following tetON-CAT1 induction (Fig 9C). Taken together, the data indicate that high basal catalase levels increase the requirement for iron in C. albicans.
This study has important implications for the impact of the key peroxide detoxifying enzyme, catalase, upon the stress resistance and virulence of the major fungal pathogen, C. albicans. Firstly, our analyses of new cat1Δ null mutants, in which potential URA3 position effects have been circumvented [49], have reinforced the view that catalase is essential for normal levels of oxidative and combinatorial stress resistance in C. albicans (Figs 2 & 8). They also show that catalase contributes to the resistance of this pathogenic fungus against neutrophil killing (Fig 7). However, our most surprising finding was that, in contrast to the generally held view [23,40], catalase is not essential for the virulence of C. albicans, at least in models of disseminated candidiasis. This unexpected finding is supported by virulence assays in both short term and long term murine models of systemic infection, and in an accepted invertebrate model of systemic infection (Fig 6). This view is further reinforced by our in vivo competition experiments, in which the cat1Δ null mutant competed effectively against wild-type and catalase overexpressing strains for colonisation of the kidney, liver, spleen and brain (Fig 5). We suggest that the attenuated virulence of the cat1Δ mutants reported previously [23,40] might be explained by URA3 position effects in these strains [49].
Why might catalase be important for oxidative stress resistance and yet apparently not required for systemic infection? The sensitivity of cat1Δ cells to neutrophil killing (Fig 7) does indicate that protection against peroxide is required in certain contexts in vivo. Therefore, this lack of cat1Δ virulence defect probably reflects the multifactorial nature of virulence phenotypes, as well as the nature of the systemic infection models often used to examine virulence. In these models sufficient fungal doses are applied to overcome immediate clearance by circulating phagocytes [47]. Furthermore, few of the fungal cells colonising the kidney appear to be exposed to oxidative stress [31].
Secondly, our data indicate that high basal levels of catalase promote the resistance of C. albicans to peroxide and combinatorial stress (Fig 2). These data reaffirm previous reports that elevated catalase expression promotes peroxide resistance [28,36]. Significantly, our data indicate that this phenotype is dependent on high basal levels of catalase at the point of stress imposition, rather than CAT1 induction in response to stress. Three independent observations support this view. (A) tetON-CAT1 cells are only protected against peroxide or combinatorial stress if these cells are pre-treated with doxycycline, not if doxycycline is only provided at the same time as the stress (Fig 2). (B) Clinical isolates that are relatively resistant to oxidative stress tend to express catalase at relatively high levels (Fig 3). (C) Unstressed C. albicans cell populations display heterogeneity in Cat1-GFP levels, and those cells that express more Cat1-GFP are less susceptible to killing by oxidative stress (Fig 4). Hydrogen peroxide is normally rapidly detoxified by wild-type C. albicans cells (within 60 minutes) in a catalase-dependent fashion [28]. Elevated basal levels of catalase presumably enhance cellular protection by accelerating the clearance of this reactive oxygen species. The heterogeneity in catalase expression within C. albicans populations, which might arise via stochastic differences between cells [52–54], appears to account, in large part, for the ability of a subset of C. albicans cells to survive an acute oxidative stress. This would appear to represent the first example in C. albicans of the kind of “bet-hedging” strategies that have been observed in bacterial and S. cerevisiae populations [55,56]. Furthermore, these observations are entirely consistent with the well-established observation that an adaptive response to a small dose of a particular stress can transiently endow yeasts with resistance to a subsequent acute dose of the same stress by inducing the expression of key stress protective functions. This observation has been reported for heat shock, osmotic and oxidative stress in S. cerevisiae for example [57,58], and has been extended to other yeasts including C. albicans [43,59,60].
Thirdly, our data provide key insights into the impact of catalase levels on the virulence of C. albicans. In our hands, direct competition assays suggested that elevated catalase levels might affect C. albicans colonisation of the kidney and brain (Fig 5). This is consistent with a parallel study which reported that catalase overexpression attenuates the virulence of C. albicans [36]. Roman and co-workers described this as “a most unexpected result” given that catalase overexpression enhances oxidative stress resistance. They speculate that this might have arisen via some alteration in fitness, which they were unable to detect in vitro, but which might interfere with activation of the Hog1 and Mpk1 MAP kinases [36]. In this study we show clearly in direct competition assays that elevated basal catalase levels attenuate the fitness of C. albicans in the absence of stress (Fig 8). We conclude that catalase overexpression confers a selective disadvantage in C. albicans in the absence of stress.
Fourthly, we have identified a major cause of this fitness defect. High basal catalase levels increase the cellular requirement for iron in C. albicans. We present two key observations that support this. (i) The fitness defect is suppressed by iron supplementation (Fig 9B and S4 Fig). This effect, which has also been observed in bacteria [50], is probably mediated by the depletion of intracellular iron through high level expression of an abundant heme-requiring enzyme. (ii) Ectopic catalase expression induces the expression of iron-responsive genes that play key roles in iron scavenging and homeostasis: e.g. CFL5, FET3, FRP1 and FTR1 (Fig 9C). Therefore, the demand for iron and catalase expression are intimately linked in C. albicans. Both modulate the accumulation of intracellular ROS. Iron stimulates CAT1 expression in C. albicans [16,61]. This increase in catalase affects iron demand and homeostasis (Fig 9B & 9C) and also enhances the detoxification of hydrogen peroxide, thereby decreasing the production of highly toxic hydroxyl radicals via the iron-dependent Fenton reaction [11]. Parallels exist in S. cerevisiae, where heterogeneity in superoxide dismutase (SOD1) gene expression affects the fitness of individual cells in the presence of copper [62].
The impact of catalase levels on the requirement for iron is likely to have a profound effect on C. albicans pathogenicity because iron homeostasis is tightly regulated during infection [10,15] and efficient iron assimilation is essential for colonisation of iron limiting niches in the mammalian host [7]. It would appear significant, therefore, that we observed reduced colonisation for catalase overexpressing cells in the kidney and brain, but not in the iron-rich liver and spleen (Fig 5).
In conclusion, elevated basal catalase levels appear to be a double-edged sword whereby they protect C. albicans against oxidative and combinatorial stresses imposed by the host while increasing the pathogen’s demand for an essential, but limiting micronutrient in the host. This double-edged sword would appear to account for the apparently counterintuitive observation that catalase overexpression in C. albicans decreases host colonisation in some tissues [36]. It also helps to explain why C. albicans has not evolved to express the high levels of catalase that would protect it from phagocytic killing [28,36].
The strains used in this study are listed in S1 Table. C. albicans was routinely grown at 30°C, 200 rpm in YPD (2% dextrose, 2% mycological peptone, 1% yeast extract) containing 20 μg/ml doxycycline (Dox) when required. On the day of an experiment, overnight cultures were diluted into fresh YPD to an OD600 of 0.2, and incubated at 30°C at 200 rpm until they reached an OD600 of 0.8, whereupon they were subjected to the appropriate treatment and analysed. Plates were incubated for 48 h at 30°C.
Osmotic stress was applied with 1 M NaCl and oxidative stress was applied with H2O2 at the specified concentration. Combinatorial stress was imposed using 1 M NaCl plus 5 mM H2O2 as described previously [28,63].
Robotic plating was performed using a Singer RoToR robot (Singer Instruments, Watchet, UK). Fitness was assayed by monitoring growth in microtitre plates at OD600 every 20 min for 48 h, and data from independent triplicate experiments were analysed.
The CAT1 locus was deleted from the C. albicans strain CEC2908 using the Clox system as previously described [64] (S2 Table), thereby generating the homozygous cat1Δ null mutant Ca2037 (S1 Table). Using published procedures [38], the C. albicans CAT1 ORF was then cloned into barcoded CIp10-PTET-GTw plasmids and these plasmids were integrated at the RPS1 locus in C. albicans Ca2037 (S1 Table) to generate the strains Ca2038, Ca2040, Ca2041, Ca2043, Ca2044, and Ca2046 (S1 Table). Empty barcoded CIp10-PTET-GTw plasmids were transformed into C. albicans CEC2908 to create strains Ca2084, Ca2085 and Ca2087 (S1 Table). Empty barcoded CIp10-PTET-GTw plasmids were also transformed into C. albicans Ca2037 to generate strains Ca2089, Ca2092 and Ca2130 (S1 Table). This created an isogenic set of nine barcoded wild-type (CAT1), null (cat1Δ) and tetON-CAT1 strains. Their 25 bp barcodes are described in S3 Table.
The CAT1-GFP/CAT1-GFP strain Ca2213 (S1 Table) was constructed by PCR amplifying CAT1-GFP-URA3 and CAT1-GFP-HIS1 cassettes (S2 Table) [65] and integrating these sequentially at the 3’-end of the CAT1 alleles in C. albicans RM1000 (S3 Table).
To quantify the relative concentration of each barcoded strain in mixed populations of tetON strains, genomic DNA was prepared from the populations by phenol: chloroform extraction method [66]. A 60 bp region carrying the barcodes (S3 Table) was amplified with common primers (S2 Table) using the KAPA HiFi HotStart ReadyMix PCR Kit (KAPA Biosystems, London, UK) and ethanol precipitated. These purified amplicons, which contained the Illumina overhang, were then indexed with Illumina Nextera XT v2 indices (Illumina, Inc., San Diego, CA, USA). Briefly, the dual indexed Illumina libraries were prepared with 5 μl of DNA, 5 μl each of i5 and i7 index primer, 25 μl KAPA HiFi HotStart ReadyMix, and 10 μl of PCR grade water and PCR amplified (95°C for 3 min; 8 cycles of 95°C for 30 sec, 55°C for 30 sec and 72°C for 30 sec; 72°C for 5 min; and a final hold at 4°C) on a Life Technologies Veriti thermal cycler (Thermo Fisher Scientific, Waltham, MA, USA). The libraries were purified and size selected using a double size selection with SPRIselect (Beckman Coulter, Brea, CA, USA) with a SPRIselect to sample ratio of 0.85x followed by 1.0x. Libraries were quantified using the Thermo Fisher Scientific Quant-iT dsDNA High Sensitivity Assay and the fluorescence measured on a BMG Labtech FLUOstar Omega microplate reader (BMG Labtech GmbH, Ortenberg, DE). The quality and size (bp) of the libraries were analysed on an Agilent 2200 TapeStation with High Sensitivity D1000 ScreenTapes (Agilent Technologies, Santa Clara, CA, USA). The libraries were pooled in equimolar amounts and sequenced on an Illumina MiSeq Sequencing System using MiSeq v3 chemistry with 76 bp paired-end reads. Base calling and fastq output files were generated with RTA v1.18.54 software on the MiSeq instrument.
To analyse the barseq data, a wrapper script was coded over the open source BBDuk tool (BBMap suite version 35.43 [67]). The wrapper visits each sample directory and runs the 3rd-party bbduk.sh script over each of the compressed read 1 and read 2 FASTQ files, generating corresponding FASTQ output files for the “matched” and “not-matched” reads for each barcode. The wrapper then computes the total number of reads for each barcode and its abundance relative to the total number of barcode reads. The barseq data are presented as the relative abundance of a barcode normalised to its starting concentration in the population. Means and standard deviations from three replicate measurements are presented.
RNA was extracted from C. albicans cells using the Zymo Research YesStar RNA Kit (Cambridge Bioscience, Cambridge, UK). cDNA was prepared using SuperScript II reverse transcriptase from Invitrogen (Fisher Scientific, Loughborough, UK), and qRT-PCR was performed with a Roche Light Cycler 480 II using the primers described in S2 Table. Transcript levels were measured in triplicate, expressed relative to the internal ACT1 mRNA control [28], and then normalised against the levels in doxycycline-treated wild type (CAT1) cells to exclude potential effects of doxycycline on these transcripts.
C. albicans cells grown in YPD containing 0 or 20 μg/μl Dox were subjected to no stress or one hour of 5 mM H2O2, protein extracts prepared, and catalase activities measured using the BioAssay Systems EnzyChrom catalase assay kit (Universal Biologicals Ltd., Cambridge, UK), according to the manufacturer’s instructions [28]. Assays were performed in triplicate.
CAT1-GFP and ACT1-GFP expression in C. albicans cell populations was examined and cell subsets isolated using the BD Influx cell sorter. Heterogeneity in C. albicans cell size was first analysed (Forward Scatter (FSC), Side Scatter (SSC)) and cells of similar size selected (S2 & S3 Figs). Cells were then sorted on the basis of their GFP expression level (S2B Fig). Cells (n = 200) that expressed GFP at relatively low levels and 200 cells expressing GFP at high levels were plated onto YPD containing various concentrations of H2O2. These two populations sorted were separated to 99% purity. Control experiments were performed to confirm cell viability by propidium iodide staining (2 μg/ml). Data were analysed using BD FACS software and Flowjo software version 10.0.8.
CAT1-GFP cells were visualized using a DeltaVision Core microscope (Applied Precision, Issaquah, WA). Western blotting was performed as described previously [68].
Cell viability was assayed by measuring colony forming units (CFU) on YPD plates and by propidium iodide (PI) staining and flow cytometry on a BD LSR II, as described previously [28,63].
Intracellular ROS accumulation was measured by staining the cells with 20 μM dihydroethidium for one hour in darkness, at 30°C and 200 rpm, and then analysed using a BD LSR II flow cytometer. Data were analysed using Flowjo software version 10.0.8.
Blood from healthy donors was obtained according to the protocol approved by the University of Aberdeen College Ethics Review Board (Application number—CERB/2012/11/676). Polymorphonuclear (PMN) cells, or neutrophils, were isolated from this blood using Histopaque-1119 and Histopaque-1077 (Sigma Aldrich) as described previously [28]. C. albicans cells pre-grown with 20 μg/ml Dox were incubated with PMNs (1:10 ratio of yeasts to neutrophils) for 2 h in RPMI 1640 containing 10% heat inactivated foetal bovine serum. After incubation the PMNs were treated with 0.25% sodium docecyl sulphate and DNase I and yeast survival determined by assaying CFU. Data from eight healthy donors are presented with their means and standard deviation.
The virulence of C. albicans wild type and cat1Δ cells were measured in a short term murine model of systemic candidiasis [47]. Strains were pre-grown in YPD and injected intravenously (4 x 104 CFU/g body weight) into the lateral tail vein of 6–10 week old female BALB/c mice (Envigo, UK). Mice were randomly assigned to cages (n = 6 per group) and inocula were randomly assigned to cages. Infections were allowed to proceed for 4 days whereupon the mice were humanely culled by cervical dislocation and fungal burdens (CFU/g) determined in the kidneys. Fungal burden and weight loss were used as measures of virulence [47].
The virulence of C. albicans wild type and cat1Δ strains were also tested in a longer term mouse infection model. Again, C. albicans cells were injected into the tail veins of 6–10 week old female BALB/c mice (3 x 104 CFU/g body weight). Once again, the mice were randomly assigned to cages (n = 8 per group) and inocula were assigned randomly to cages. The mice were monitored and weighed daily, and were humanely culled when they had lost 20% of their body weight and death recorded as having occurred on the following day. Experiments were continued for a maximum of 14 days, when all surviving mice were culled and analysed. The data are presented as Kaplan-Meier survival curves (log rank tests).
To directly compare the colonisation of C. albicans tetON strains in the mouse model of systemic candidiasis, the strains were pre-grown in YPD containing 0 or 20 μg/ml doxycycline and injected into the tail vein of 6–10 week old female BALB/c mice (4 x104 CFU/g body weight: n = 6 mice per group). Mice were gavaged with 100 μl of 0 or 40 mg/ml doxycycline. Infections were allowed to proceed for up to 4 days. Mice were culled, their kidneys, spleen, liver and brain removed and homogenized in 500 μl saline, and the entire sample from each organ plated onto YPD. The fungal colonies from each individual organ were then pooled, and genomic DNA prepared for barseq (above).
The virulence of C. albicans strains was also evaluated using the invertebrate Galleria mellonella infection model [69]. For each C. albicans strain, 105 cells were injected into 20 Galleria larvae (6th instar: BioSystems Technology, Exeter, UK). Sterile PBS was injected into control larvae. Survival was monitored for 5 days at 37°C, represented using Kaplan-Meier curves, and analysed using log rank tests.
All animal experiments were conducted in compliance with United Kingdom Home Office licenses for research on animals, and were approved by the University of Aberdeen Ethical Review Committee (project license number PPL 70/8583). Animal experiments were minimised, and all animal experimentation was performed using approaches that minimised animal suffering and maximised our concordance with EU Directive 2010/63/EU.
Power analyses based on data generated in previous experiments were applied to estimate the minimum number of animals per group required to achieve statistically robust differences (P <0.05). The power analyses to determine group size for the short term systemic infection model were based on the variation in fungal burdens between animals, whereas those for the long term model were based upon mean survival times. Animals were monitored at least twice daily for signs of distress, which was minimised by expert handling. Euthanasia was performed humanely by cervical dislocation when animals showed signs of progressive illness (e.g. ruffled coat, hunched posture, unwillingness to move and 20% loss of initial body weight). During these studies there were no unexpected deaths. Analgesia and anaesthesia were not required in this study.
Statistical analyses were performed in GraphPad Prism 5 and IBM SPSS Statistics (v24.0.0). Two tailed Mann-Whitney U analysis was used to test the statistical difference between two sets of data with a non-parametric distribution. Associations between growth parameters, such as doubling time, lag phase or propidium iodide staining, were determined by one-way and two-way ANOVA and Dunnett post-hoc t-tests. Unstressed samples were used as controls and the values of other samples were compared against these controls. The following p-values were considered: * p < 0.05; ** p <0.01; *** p < 0.001; **** p < 0.0001.
|
10.1371/journal.pntd.0001801 | Severe Hemorrhagic Fever in Strain 13/N Guinea Pigs Infected with Lujo Virus | Lujo virus (LUJV) is a novel member of the Arenaviridae family that was first identified in 2008 after an outbreak of severe hemorrhagic fever (HF). In what was a small but rapidly progressing outbreak, this previously unknown virus was transmitted from the critically ill index patient to 4 attending healthcare workers. Four persons died during this outbreak, for a total case fatality of 80% (4/5). The suspected rodent source of the initial exposure to LUJV remains a mystery. Because of the ease of transmission, high case fatality, and novel nature of LUJV, we sought to establish an animal model of LUJV HF. Initial attempts in mice failed, but infection of inbred strain 13/N guinea pigs resulted in lethal disease. A total of 41 adult strain 13/N guinea pigs were infected with either wild-type LUJV or a full-length recombinant LUJV. Results demonstrated that strain 13/N guinea pigs provide an excellent model of severe and lethal LUJV HF that closely resembles what is known of the human disease. All infected animals experienced consistent weight loss (3–5% per day) and clinical illness characterized by ocular discharge, ruffled fur, hunched posture, and lethargy. Uniform lethality occurred by 11–16 days post-infection. All animals developed disseminated LUJV infection in various organs (liver, spleen, lung, and kidney), and leukopenia, lymphopenia, thrombocytopenia, coagulopathy, and elevated transaminase levels. Serial euthanasia studies revealed a temporal pattern of virus dissemination and increasing severity of disease, primarily targeting the liver, spleen, lungs, and lower gastrointestinal tract. Establishing an animal LUJV model is an important first step towards understanding the high pathogenicity of LUJV and developing vaccines and antiviral therapeutic drugs for this highly transmissible and lethal emerging pathogen.
| The pathogenic arenaviruses are a diverse group of human pathogens capable of causing a wide range of human illness ranging from encephalitis to severe hemorrhagic fever throughout the New and Old World. In 2008, a previously unknown virus (now named Lujo virus) caused a high case fatality outbreak (80%) in southern Africa. Limited data available from these patients indicated that LUJV HF was characterized by thrombocytopenia, elevated liver transaminases, coagulopathy, viral antigen in multiple tissues, neurological symptoms in some cases, and eventual death. The source of exposure of the index patient remains unknown. Due to the unusually high lethality and rapid human to human spread, we sought to develop an animal model of Lujo hemorrhagic fever. We report here that after infection with Lujo virus, Strain 13/N guinea pigs develop a hemorrhagic fever syndrome similar to the disease observed in human patients. This animal model of severe Lujo hemorrhagic fever is a critical first step to increase our understanding of this highly pathogenic virus, and to develop anti-viral therapeutics or experimental vaccines for this new and unique threat to human health.
| Beginning in the 1930s, novel pathogenic arenaviruses have been increasingly recognized as emerging threats to human health [1], [2], [3], [4], [5], [6], [7], [8], [9], [10]. During the 1960s and 1970s, several previously unknown arenaviruses emerged as a significant public health threats and causes of a severe and often fatal human hemorrhagic fever (HF) syndrome. In 2008, Lujo virus (LUJV), a novel member of the family Arenaviridae, was first identified after an outbreak of severe HF in southern Africa [11]. During this outbreak, the index patient was transported by air from Lusaka, Zambia, to a private hospital in Johannesburg, South Africa, thus giving the virus its name, Lu-Jo.
The index patient died from the infection approximately 12 days after the onset of the presumed first symptoms, and 2 days after hospitalization in Johannesburg. During transport and hospitalization of the index patient, a total of 4 health care workers (3 nurses and 1 janitor) were infected with LUJV. After a period of 10–13 days of progressively severe illness, 3 of these individuals died, resulting in a total case fatality of 80% (4/5). Limited data available from these patients indicated that LUJV HF was characterized by thrombocytopenia, elevated liver transaminases, coagulopathy, viral antigen in multiple tissues, neurological symptoms in some cases, and eventual death. While the outbreak was small, the ease with which LUJV spread among the primary, secondary, and tertiary contacts with the index patient, and the lack of a defined etiology, caused significant alarm. The viral cause of the outbreak was identified as a novel arenavirus only after the last case fatality [12]. The suspected source of exposure of the index patient to LUJV (presumably a rodent) remains unknown.
The arenaviruses are a large and genetically diverse group of over 30 viruses broadly divided into New World and Old World serogroups. They are exclusively rodent-borne, except Tacaribe virus, which was isolated from a bat [13], [14]. Phylogenetically, LUJV is distinct from both the Old World and New World arenavirus lineages, and is the sole member of a distinct branch more closely related to the known Old World arenaviruses [12]. Although many arenaviruses are not pathogenic, a large number can cause a spectrum of human disease ranging from neurological symptoms in pediatric or immunocompromised patients (e.g., Lymphocytic choriomeningitis virus, LCMV) [15], [16], [17], [18], [19], [20], to hemorrhagic syndromes with high case fatalities (e.g., Lassa virus (LASV), Junin virus (JUNV), and Machupo, Guanarito, Chapare, and Sabia viruses) [21], [22], [23].
All arenaviruses are enveloped particles containing bi-segmented, single-stranded, ambi-sense RNA genomes encoding a total of 4 genes [13]. The large (L) genome segment (∼7.2 kb) contains the viral RNA-dependent RNA polymerase and the multi-functional Z-protein, an important matrix protein that is also responsible for virus budding. The small (S) genome segment (3.4 kb) encodes the viral nucleoprotein (NP) and the glycoprotein precursor (GPC), which is post-translationally processed into the G1 and G2 structural proteins. Both NP and Z proteins function as viral virulence factors antagonizing host cell interferon responses by a variety of mechanisms [21], [24], [25], [26].
Although many experimental vaccine candidates and antiviral drugs are under development [27], [28], [29], [30], currently the only available vaccine for any pathogenic arenavirus is the live-attenuated Candid1 vaccine for JUNV, which is restricted for use only in high-risk individuals such as laboratorians and those living in endemic areas [31], [32]. The antiviral drug ribavirin has shown some efficacy in treating LASV and some other arenaviruses, but its side effects limit use to high-risk exposures and severe cases [33], [34]. Due to the unique characteristics of the LUJV outbreak and the highly novel genomic nature of LUJV, we sought to establish an animal model capable of developing HF similar to that observed in the 4 fatal human cases. Establishing a robust animal model is necessary for further investigating LUJV pathogenesis, and to provide a system to test potential vaccine candidates and antiviral therapeutic drugs. Initial experiments with LUJV infection in 2-day-old newborn and 14-day-old weanling mice failed to provide a lethal model. This was highly surprising given the near uniform lethality of pathogenic New World or Old World arenaviruses in newborn or weanling mice, respectively [12], and highlights another unique feature of LUJV.
We next attempted to develop a LUJV HF model in guinea pigs (Cavia porcellus), which have been used since the 1960s as reliable models for a variety of pathogenic New World and Old World arenaviruses [35], [36], [37], [38], [39], [40], [41], [42], [43], [44]. The inbred strain 13/N guinea pig is highly susceptible to LASV infection; the animals develop severe, progressive, and ultimately fatal disease 15–21 days post-infection (PI), showing many pathological changes that closely mimic Lassa fever in humans [37]. Given their susceptibility to Old World arenaviruses, we began experiments in strain 13/N guinea pigs to study LUJV virulence and pathogenesis. Here, we report successfully establishing a lethal model of LUJV HF in strain 13/N guinea pigs infected with either authentic wild-type or a recombinant full-length reverse genetics-derived LUJV. This robust and highly uniform animal model will permit further detailed investigations into the molecular determinants of LUJV pathogenesis, and provide an in vivo system for testing novel anti-viral therapeutics and vaccines against this highly pathogenic and unique arenavirus.
All work with infectious virus or infected animals was conducted at the Centers for Disease Control and Prevention (CDC, Atlanta, Georgia, USA), in a biosafety level 4 laboratory. All laboratorians and animal handlers adhered to international biosafety practices appropriate for biosafety level 4, strictly following infection control practices to prevent cross-contamination between individual animals. All animals were individually housed in an isolator-caging system (Thoren Caging, Inc., Hazleton, PA, USA) with a HEPA-filtered inlet and exhaust air supply.
All procedures and experiments described herein were approved by the CDC Institutional Animal Care and Use Committee (IACUC) and conducted in strict accordance with the Guide for the Care and Use of Laboratory Animals [45]. All animals were housed in a climate-controlled laboratory with a 12 h day/12 h night cycle. The CDC is an Association for Assessment and Accreditation of Laboratory Animal Care International (AAALAC) fully accredited research facility. No human patient derived clinical materials were used in the completion of these studies.
A total of 8 litters of pregnant outbred mice were obtained from a commercial vendor (Charles River Laboratories, Wilmington, MA, USA). All mice were housed as individual family units, and supplied a commercially available mouse chow and water ad libitum. The cage environment was enriched with large amounts of soft bedding, shredded paper, and cotton nestlets. After infection, each animal was observed at least once per day, and its health assessed and scored by experienced CDC veterinarians or animal health technicians. Animals were humanely euthanized with isoflurane vapors once clinical illness scores (including, but not limited to, neurological signs, piloerection, ocular discharge, weight loss, changes in mentation, ataxia, dehydration, or dyspnea) indicated that the animal was in distress or in the terminal stages of disease.
A total of 47 strain 13/N guinea pigs (healthy adult males and females aged 1.0–1.5 years) were obtained from an established breeding colony located at the University of Iowa (Ames, IA, USA). All animals were housed individually on deep soft bedding and given food (commercial guinea pig chow, alfalfa cubes, and fresh green parsley), and water supplemented with guinea pig appropriate vitamins ad libitum, following standard laboratory animal husbandry protocols for guinea pigs. After infection, each animal was observed at least twice per day, and its health assessed and scored by experienced CDC veterinarians or animal health technicians. Animals were humanely euthanized with isoflurane vapors and sodium pentobarbital (Schering-Plough, Kenilworth, NJ, USA) at either predetermined times PI, or once clinical illness scores (including, but not limited to, piloerection, ocular discharge, weight loss, changes in mentation, ataxia, dehydration, dyspnea, or hypothermia) indicated that the animal was in the terminal stages of disease.
Wild-type LUJV (wtLUJV) from the Centers for Disease Control and Prevention Viral Special Pathogens Branch reference collection was passaged in VERO-E6 cells five times before use. A full-length recombinant LUJV (recLUJV) was derived from cDNA plasmids using T7-driven reverse genetics as reported in Bergeron et al. (manuscript in review). To differentiate recLUJV from wtLUJV, a few silent (non-coding) nucleotide changes were introduced into the full-length L segment clones of recLUJV (also described in Bergeron et al., manuscript in review). For mouse experiments, wtLASV-Josiah and wtJUNV-XJ13 were prepared and utilized as described in [46], [47]. Prior to use, all virus stocks were titrated and full-length genomic sequences were verified using standard techniques as described in [46], [47].
Specimens of liver, lung, spleen, kidney, whole blood, urine, and/or pleural effusion or abdominal fluid (if present) were collected sterilely at 2, 5, 7, 9, 12, and 14 days PI, and from moribund animals that reached experimental end-points. For RNA extraction, approximately 100 mg specimens of tissues were stored in RNA extraction buffer (Tripure, Roche Diagnostics, Indianapolis, IN, USA) at −80°C until homogenization in a high-throughput tissue grinder (Genogrinder2000, BT&C Inc., Lebanon, NJ, USA). An equal volume of molecular grade chloroform was added to each specimen homogenate and vortexed. After a 10 minute spin at >10,000 rpm in a microcentrifuge, the supernatant was collected and an equal volume of 70% ethanol was added. The supernatant and ethanol was used for total RNA extraction (RNAeasy 96 platform, Qiagen, Valencia, CA, USA) following the manufacturer's recommended protocols.
Briefly, LUJV RNA was detected using qRT-PCR with primers and probe with internal (Zen, Integrated DNA Technologies) and 3′ Iowa Black-FQ quencher moieties specific for the NP gene (forward primer: 5′-CTCACACCCACAGGAAAT-3′; reverse primer: 5′-GGCCATACATCTCTTCCAGA-3′; probe: 5′-6FAM-ACCCTACAC/Zen/CTCCACAGAACGAAAG-IowaBlackQuencherFQ-3′). For each viral genome detection reaction, 1 uL of total RNA was added to a one-step qRT-PCR (Invitrogen), where the first stand was synthesized using Superscript III at 50°C for 15 min, denatured at 94°C for 2 min, and amplified for 40 cycles of 94°C for 15 s and 60°C for 1 min (ABI 7500, Life Sciences, Grand Island, NY, USA). LUJV RNA genome equivalents in infected blood, fluid, or tissue specimens were quantitated using a standard curve generated by serial dilutions of a known-titer stock virus spiked into normal whole guinea pig blood. The results of all qRT-PCR tests were normalized to endogenous rodent-specific controls (glyceraldehyde 3-phosphate dehydrogenase (GAPDH), Invitrogen) following the manufacturer's recommended protocols to account for sample-to-sample variation in RNA extraction efficiency.
Guinea pig whole blood was collected by intracardiac techniques into either EDTA-coated or heparin-coated vacutainer tubes. Complete blood counts (CBC) were obtained using the Hematrue blood analyzer (HESKA, Loveland, CO, USA). Blood chemistry profiles were obtained from heparinized samples using either the Piccolo point of care chemistry analyzer (Abaxis, Union City, CA, USA) or the Hitachi P-module analyzer (Hitachi Hi-Tech, Tokyo, Japan).
Liver, spleen, lung, and kidney tissues were collected 2, 5, 7, 9, 12, and 14 days PI (serial euthanasia groups), and from moribund animals that reached experimental end-points of terminal disease. Specimen RNA was treated with DNase I (Qiagen) followed by RNA cleanup utilizing the RNeasy Mini columns and wash buffers (Qiagen) per manufacturer's recommendations. Total RNA was quantified after DNase I treatment and cleanup using a NanoDrop spectrophotometer (Thermo Scientific, Wilmington, DE, USA). Previously reported gene specific primers were used to detect interleukin (IL)-1b, IL-2, IL-8, IL-12p40, tumor necrosis factor alpha (TNFa), transforming growth factor beta (TGFb), regulated upon activation normal T-cell expressed and secreted (RANTES), interferon gamma (IFNg), monocyte chemotactic protein (MCP)-1, inducible nitric oxide synthetase (iNOS), and GAPDH [48], [49]. Generally, 50 ng of RNA was used for each individual qRT-PCR; however, for samples with low RNA concentrations, a minimum of 5 ng was used. Invitrogen's SuperScript III Platinum SYBR green one-step qRT-PCR kit was used for 25 uL total volume reactions, with final reaction concentrations of 1× reaction mix, 0.2 M primers, 0.5 uL enzyme mix, and 50 ng RNA. Identical thermocycling profiles were utilized for all assays; 55°C for 10 min; 95°C for 5 min; amplification for 40 cycles of 95°C for 15 s and 58°C for 30 s; 40°C for 1 min; and a dissociation curve (CFX96 Touch, Bio-Rad, Hercules, CA, USA). A gene-specific real-time assay was developed for IL-10 (GenBank accession JN020146) using the GenScript real-time PCR primer design tool, IL10 F 5′-CACAGGATCAGCTGGACAAC–3′, IL10 R 5′-GGGCATCACCTCCACTAGAT-3′, and IL10 Probe 5′(FAM)-CCTGGGTTGCCAAGCCTTGTC-(BHQ1)3′. The Invitrogen SuperScript III Platinum one-step qRT-PCR kit was used for 25 µL total volume reactions following the manufacturer's protocol and the following thermocycling profile; 55°C for 10 min, 95°C for 2 min, amplification for 40 cycles of 95°C for 15 s and 60°C for 30 s. Guinea pig GAPDH was used as the internal control calibrator. Bio-Rad CFX manager software v2.1 was used to analyze the cycling threshold (CT) values and melt curves for each reaction, and results were analyzed using the comparative CT method as described by Schmittgen and Livak [50]. The fold change of each serial euthanasia group was compared to mean fold change of the sham-infected control guinea pigs (N = 6), and the standard error of the mean was calculated for each experimental group. Due to technical issues during RNA extraction, the day 5 PI spleen data was generated from only 1 infected animal.
At the time of collection, tissue specimens were fixed in 10% neutral buffered formalin and gamma-irradiated (2.0×106 RAD) prior to sectioning into 4 um-thick slices and staining with hematoxylin and eosin following routine histology protocols.
All analyses were completed using the PRISM v5.0 program (Graphpad, LaJolla, CA, USA). Potentially significant differences between wtLUJV and recLUJV groups were evaluated using a student's t-test. In subsequent analyses, wtLUJV and recLUJV data were combined for all day 9 and terminal group analyses. For the complete blood counts, clinical chemistry, and gene regulation data, significant differences between LUJV-infected and sham-infected animals at each time point were analyzed using a one-way analysis of variance (ANOVA) with Dunnett's adjustment for multiple comparisons (*p<0.05; **p<0.01, ***p<0.001).
Mice were monitored for 28 days PI with 500 FFU of wtLUJV, wtLASV-Josiah, wtJUNV-XJ13, or inoculation with DMEM as a negative control. As expected, wtJUNV-XJ13 caused uniform neurological signs and lethality by 15 days PI in 2-day-old, but not 14-day-old mice. In contrast and as expected, wtLASV infection resulted in near uniform (90%) lethality in 14-day-old weanling mice, but was non-lethal in 2-day-old newborn mice. Infection was confirmed by the detection of anti-Lujo virus specific antibodies at 28 days post-infection in 3 surviving animals. Surprisingly, wtLUJV did not cause any signs of clinical illness or lethality in either 2-day-old or 14-day-old mice regardless of the dose (up to 2.0×103 FFU) or inoculation route (intracranial, subcutaneous, or intraperitoneal) (Fig. S1 and data not shown).
The pathogenic arenaviruses are genetically diverse, globally distributed, and capable of causing human illness ranging from encephalitis to severe and often lethal HF. Divided into New World and Old World lineages, these rodent-borne viruses are most often transmitted to humans by direct contact or aerosol exposure to infectious rodent excreta, or, in some cases, via a chain of human-to-human transmissions within hospital settings. The overall public health impact can range from only a few cases (e.g., Sabia or Chapare viruses) to over 100,000 cases per year (e.g., LASV) [22], [51]. LUJV is the most recently discovered pathogenic arenavirus, identified in 2008 after a high fatality (80%) cluster of cases among primary, secondary, and tertiary contacts with the index patient [52].
A number of factors mark LUJV as a unique arenavirus. Although the data from the 2008 outbreak are limited, the high case fatality was striking compared with most other arenavirus outbreaks that typically are associated with case fatalities of 10–40% [53], [54]. Phylogenetically, LUJV is distinct from all previously identified arenaviruses, forming a unique lineage more closely related to the Old World than New World arenaviruses, but with over 40% divergence from LASV at the nucleotide level [12]. The unique genomic sequence and high antigenic diversity compared to other Old World and New World arenaviruses greatly complicated its initial diagnosis by molecular (RT-PCR) or serology (IgM/IgG) techniques [11]. Unfortunately, due to these and other factors, the virus later named LUJV was only identified as the specific cause of the 2008 outbreak several weeks after the last fatal case of Lujo HF.
To assess the in vivo characteristics of LUJV, we began with the traditional newborn and weanling outbred mouse models of arenavirus infection [12]. These experiments further demonstrated the unique virulence properties of LUJV compared to New World (JUNV-XJ13) and Old World (LASV-Josiah) prototype arenaviruses. Both JUNV and LASV are highly virulent in newborn and weanling mice, respectively, causing lethality in an age-dependent pattern [12], [46], [47], [55]. In previous experiments, JUNV dosages as low as 1.0×101 FFU caused uniform lethality after intracranial inoculation into 2-day-old mice (data not shown). Similar dosages of LASV were lethal in 14-day-old mice (data not shown). However, LUJV was non-lethal in mice regardless of the route of inoculation (intracranial, subcutaneous, or intra-peritoneal), mouse age at inoculation (2 or 14 days old), or viral dose (ranging up to the maximum dose tested, 2.0×103 FFU; Fig. S1, and data not shown).
In marked contrast with the mouse results, LUJV caused severe, rapidly progressive, and uniformly lethal and hemorrhagic disease in strain 13/N guinea pigs. In this model, we observed an apparent incubation period of 5 to 6 days from the time of inoculation to the first clinical signs of illness (fever and weight loss). Over the next 24 to 48 h, the animals began to display signs of progressive illness (bilateral ocular discharge, continued fever, weight loss, and dehydration) until they were found dead or humanely euthanized when moribund. By day 5 PI, significant hematological changes began to occur, including hypoproteinemia, thrombocytopenia, and lymphopenia (Fig. 3). Interestingly, although we saw consistent elevation in key serum transaminase enzyme (aspartate transferase, alkaline phosphatase, and alanine transferase) activities, these were neither dramatic nor statistically significant and are typical of the generally poor release of tissue transaminases in guinea pigs in the face of tissue damage or necrosis. Subjective comparisons of clinical illness severity, and gross and histologic pathology between guinea pigs infected with LUJV or LASV suggests that LUJV caused a more profound illness (greater and more rapid weight loss, and frank hemorrhage and congestion in gastrointestinal organs, bladder, lymph nodes, and abdominal cavity), tissue damage (hepatic and myocardial necrosis), and hallmarks that may be consistent with disseminated intravascular coagulation (DIC) than LASV in preliminary experiments completed in our laboratory and as reported in [37] (Figs. 5 and 6, and data not shown).
Previous studies of human pathogenic New World (JUNV, Guanarito (GTOV), Machupo (MACV)) and Old World (LASV) arenaviruses using rodents and other small animal models failed to demonstrate clear consistent signs of HF (reviewed in [56]). LUJV infection of guinea pigs, however, resulted in severe infarction, fibrin deposition, and hemorrhage in multiple organs, suggesting DIC. The progressive reductions in platelet numbers (9, 12, and 14 days PI) are consistent with the consumption of platelets at sites of localized virus inflammation and tissue destruction, or, in the most extreme cases, may be consistent with DIC at later time points PI. Anecdotally, the animal with severe frank hemorrhage into the abdominal cavity developed profound thrombocytopenia (99×103 uL−1). Unfortunately, no assays to determine coagulation factor parameters (i.e., activated partial thromboplastin time (APTT) or prothrombin time (PT)) were completed to directly assess the influence of these coagulation pathways on the observed coagulopathy. It is clear that further work is essential to definitively characterize the underlying mechanisms of these observations before firm conclusions can be drawn. For these types of studies this guinea pig model may provide a potentially useful alternative to non-human primate models for studying basic pathogenesis of a bona fide human pathogenic arenavirus causing severe coagulopathy and HF.
The rapid rise and magnitude of LUJV-specific antigen deposition, histologic pathology, and RNA titers in tissues, blood, urine, and abdominal fluid were surprising. Within 48 h of infection, the virus already disseminated to the liver, spleen, kidneys, and lungs, and was rapidly replicating up to 1.8×105 TCID50 eq/g of liver. Interestingly, this high-titer replication continued for another 48–72 h before the onset of illness, indicated by increased body temperature and weight loss 5–6 days PI. Tissue viral loads remained extremely high throughout the course of the disease, with death occurring 11–16 days PI.
Like other arenaviruses, LUJV broadly modulates host immune responses during infection. The molecular basis of the seemingly high LUJV virulence in humans has not been characterized. However, recent work using reverse genetics-derived recombinant viruses indicates that LUJV has unique promoter elements that influence the expression of the viral NP and glycoprotein, and an unusually long intergenic region sequence on the viral L segment, which influences expression of the Z protein (Bergeron et al., in-review). Both LASV and lymphocytic choriomeningitis virus NP and Z proteins have immunomodulatory properties and function as potent antagonists of host cell antiviral responses ([21], [24], [25], [26] and reviewed in [57]). These and other yet unrecognized molecular motifs may augment the apparently enhanced virulence of LUJV by increasing viral replication or interfering with immunoregulatory mechanisms within the host.
The ability of LUJV to influence the immune response is illustrated in guinea pigs by the finding that, despite the very high viral loads by day 5 PI, pro-inflammatory cytokine/chemokine genes, such as IL-1b and RANTES, were downregulated 2–4-fold early during infection (Fig. 7). In contrast, as early as day 5 PI, potent mediators of macrophage and neutrophil activation and inflammation (IL-8, MCP-1) and pro-inflammatory molecules (IL-12p40, and IFNg) were transcriptionally upregulated (10 to >100-fold) compared to mock inoculated animals, especially in liver and kidney tissues (Fig. 7, and Table S1). Potentially important and robust induction of the broad immunomodulatory mediator IL-10 was also detected in lungs by day 9 PI, and may signal attempts to limit and control pro-inflammatory activity, allowing for further virus replication.
Overall, the pattern of gene induction and histological evidence from our study allow for the speculation that, in early stages of infection, the animals mounted a pro-inflammatory and innate immune response (presumably involving macrophages, neutrophils, NK-cells (i.e., Kurloff cells in the guinea pig), and/or NKT-cells) in multiple tissues. This was followed by a predominantly Th1 response dominated by IFNg, MCP-1, and IL-12p40, which likely stimulated activation and enhanced function presumably of NK cells, CD4+ TH1 cells, and/or CD8+ cytotoxic lymphocytes, resulting in a strong bias towards cell mediated immunity. Since LUJV is not highly cytopathic in cell culture, these responses may have been more deleterious than helpful to the host, due to immune cell-mediated destruction of vital organs. This hypothesis is consistent with recent work describing enhanced LASV pathogenesis due to deleterious T-cell mediated activation and stimulation of monocytes/macrophages, leading to tissue destruction in humanized HHD mice [58]. Regardless of the immune mechanisms stimulated by LUJV infection, these responses were insufficient to control virus replication and dissemination throughout the host, and to prevent eventual lethality.
Similarly to LASV infection in humans and non-human primate animal models, LUJV does not appear, at least at the mRNA transcriptional level, to elucidate an end-stage cytokine storm as seen in fatal hemorrhagic cases of infection with Ebola, Marburg, or Rift Valley fever viruses [59], [60], [61], [62]. Although IFNu-producing cells (presumably natural killer cells, NKT-cells, and/or macrophages) are clearly activated, the increase in mRNA encoding the TNFa, iNOS, and IL-1b genes was not significant even in terminal cases. Among the limited genes analyzed in this study, we speculate that complete dysregulation of the host immune response is not responsible for the dramatic vascular permeability changes and coagulopathy observed in the guinea pig model of LUJV HF. Although our results are suggestive, more definitive studies of the guinea pig immune response are necessary before drawing distinct conclusions regarding the roles of inflammatory mediators, immune effector cells, and viral virulence factors in the pathogenesis of LUJV HF in this animal model.
The dramatic severity of the clinical illness, short survival times, high tissue viral loads, and the mRNA gene expression patterns in visceral organs further highlight the unique pathogenic characteristics of LUJV compared with other studies of LASV or JUNV [37], [44]. Our in vivo data, taken together with recent insights into the unique genomic elements regulating viral replication in cell culture, stress the importance of further work to elucidate the possibly novel pathogenic mechanisms employed by LUJV to cause HF in both humans and guinea pigs. Direct comparisons of the precise pathogenic mechanisms utilized by LUJV and other pathogenic arenaviruses are underway, and may reveal insights into the apparent enhanced virulence of LUJV in humans and guinea pigs, and provide broader understanding of arenavirus HF. Regardless of the exact mechanisms, LUJV is clearly highly pathogenic, easily transmitted in health care settings, and a potential health threat in southern Africa. Establishing a robust and reliable animal model of severe and lethal LUJV HF is a critical first step for further investigations to increase our understanding of LUJV and for developing anti-viral therapeutics or experimental vaccines for this new and unique threat to human health.
|
10.1371/journal.pntd.0005792 | Adding tsetse control to medical activities contributes to decreasing transmission of sleeping sickness in the Mandoul focus (Chad) | Gambian sleeping sickness or HAT (human African trypanosomiasis) is a neglected tropical disease caused by Trypanosoma brucei gambiense transmitted by riverine species of tsetse. A global programme aims to eliminate the disease as a public health problem by 2020 and stop transmission by 2030. In the South of Chad, the Mandoul area is a persistent focus of Gambian sleeping sickness where around 100 HAT cases were still diagnosed and treated annually until 2013. Pre-2014, control of HAT relied solely on case detection and treatment, which lead to a gradual decrease in the number of cases of HAT due to annual screening of the population.
Because of the persistence of transmission and detection of new cases, we assessed whether the addition of vector control to case detection and treatment could further reduce transmission and consequently, reduce annual incidence of HAT in Mandoul. In particular, we investigated the impact of deploying ‘tiny targets’ which attract and kill tsetse. Before tsetse control commenced, a census of the human population was conducted and their settlements mapped. A pre-intervention survey of tsetse distribution and abundance was implemented in November 2013 and 2600 targets were deployed in the riverine habitats of tsetse in early 2014, 2015 and 2016. Impact on tsetse and on the incidence of sleeping sickness was assessed through nine tsetse monitoring surveys and four medical surveys of the human population in 2014 and 2015. Mathematical modelling was used to assess the relative impact of tsetse control on incidence compared to active and passive screening.
The census indicated that a population of 38674 inhabitants lived in the vicinity of the Mandoul focus. Within this focus in November 2013, the vector is Glossina fuscipes fuscipes and the mean catch of tsetse from traps was 0.7 flies/trap/day (range, 0–26). The catch of tsetse from 44 sentinel biconical traps declined after target deployment with only five tsetse being caught in nine surveys giving a mean catch of 0.005 tsetse/trap/day. Modelling indicates that 70.4% (95% CI: 51–95%) of the reduction in reported cases between 2013 and 2015 can be attributed to vector control with the rest due to medical intervention. Similarly tiny targets are estimated to have reduced new infections dramatically with 62.8% (95% CI: 59–66%) of the reduction due to tsetse control, and 8.5% (95% 8–9%) to enhanced passive detection. Model predictions anticipate that elimination as a public health problem could be achieved by 2018 in this focus if vector control and screening continue at the present level and, furthermore, there may have been virtually no transmission since 2015.
This work shows that tiny targets reduced the numbers of tsetse in this focus in Chad, which may have interrupted transmission and the combination of tsetse control to medical detection and treatment has played a major role in reducing in HAT incidence in 2014 and 2015.
| A global programme aims to eliminate Gambian sleeping sickness (Human African Trypanosomiasis, HAT) as a public health problem by 2020. Gambian HAT is a neglected tropical disease caused by trypanosomes spread by tsetse flies and its control has relied largely on detection and treatment of human cases. In the Mandoul focus of southern Chad, regular screening of the human population (~39000 people) between 2002 and 2013 resulted in the detection and treatment of ~100 cases/year. We examined whether even better control might be achieved through the addition of vector control to medical screening. In February 2014, 2600 insecticide-treated targets (‘Tiny Targets’) were deployed in areas where tsetse were present; tsetse are attracted to the targets and on contacting it they pick up a lethal dose of insecticide. Monitoring of the tsetse population, using a network of 44 traps operated regularly between November 2013 and October 2016, showed that the mean daily catch of tsetse declined by 99.99%, from 0.7 tsetse/trap before targets were deployed to 0.005 tsetse/trap. The number of HAT cases detected by a programme of active screening also declined during this period. Mathematical modelling of the number of HAT cases reported during the period 2000–2015 suggests that 70% of the decline in cases during 2014–2015 was due to vector control. The model also suggests that the combination of these interventions may have interrupted transmission and may to lead to the elimination of sleeping sickness in the Mandoul focus by 2020. The results from Mandoul provide further empirical and theoretical evidence that the global elimination of Gambian HAT can be achieved through the integrated use of (i) case detection and treatment and (ii) vector control.
| Human African Trypanosomiasis (HAT), also called sleeping sickness, is an endemic neglected tropical disease found in sub-Saharan Africa, caused by subspecies of Trypanosoma brucei transmitted by tsetse flies (Glossina). There is no vaccine against this lethal disease, and treatment is difficult with 1–2 weeks hospitalisation for drug treatment [1]. Most (>97%) cases of HAT are caused by T. brucei gambiense. The WHO roadmap aims to eliminate Gambian HAT as a public health problem by 2020 and reach full elimination globally by 2030 [2].
Mass screening of populations followed by diagnosis and treatment of cases has been the main method used to control Gambian HAT. These mass-screening programmes, combined with passive surveillance, have saved lives and have led to a dramatic reduction of the annual reported incidence of Gambian HAT, from > 37,000 cases/year in 1998 to <3,000 cases/year in 2015 [1, WHO Global Health Observatory data repository: http://apps.who.int/gho/data/node.main.A1636?lang=en)]. Nonetheless some foci persist despite mass screening of populations [3, 4].
The recent development and successful application of cost-effective methods—particularly so-called ‘tiny targets’ [5, 6]—offers the exciting prospect of a method that can reduce densities of tsetse in HAT foci and contribute, in association with medical interventions, to the interruption of transmission [7, 8].
An example of a persistent focus is the case of Mandoul in Southern Chad. Here, notwithstanding the progress achieved during several years of active "screen and treat" programmes, and passive surveillance, often with high coverage, the number of HAT cases reported greatly exceeded the threshold for elimination as a public health problem. In 2014, we implemented a tsetse control programme to determine whether reducing the numbers of tsetse could reduce the incidence of HAT. The entomological intervention itself was conducted using insecticide-impregnated targets that kill tsetse, whilst monitoring of vector densities was performed using traps that catch and retain the flies.
The Ministry of Health of the Republic of Chad approved the study protocols and gave administrative authorizations for the activities performed by the National Control Programme against HAT (PNLTHA: Programme National de Lutte contre la Trypanosomiase Humaine Africaine). Mass screening and treatment of HAT patients were performed according to the national Chadian HAT procedures as recommended by the World Health Organization. Information on HAT, on tsetse and tsetse control using targets and the study objectives was provided directly to individual households and more broadly via radio broadcasts (Ngor, Arab, French languages) and discussion groups organized with the Bodo district health authorities, village administrations and religious groups. Consent to participate was oral and was not mandatory. All activities were made according to the national rules of the Ministry of Health in Chad, through the national program against HAT (PM, coordinator of the NCP is one of the co-authors of the paper). These national rules do not involve written consent.
All data of the paper from participants were recorded as numbers, so anonymized, from the national program.
No human biological samples other than those required for HAT diagnosis were taken from participants in the course of this study.
The study was performed in the Mandoul focus (ca. 8.12°N, 17.11°E) located in Southern Chad, close to the border with the Central African Republic (CAR). The study area covered part of Bodo, Beboto, Koldaga, Dilingala and Bekourou cantons and is about 840 km2. In this area, the mean elevation is ~400 meters and annual rainfall is between 1000 and 1200 mm/year, with two seasons: a wet season from June to October and a dry season from November to May. The landscape is characterized by woody savannah with gallery forest along the rivers. The natural vegetation is degraded in parts through agriculture activities. Within the focus lies a swamp. People pass through the swamp in the course of their daily activities. The local population comprise Ngor who are sedentary mixed crop-livestock farmers and also Arab and M’bororo (pastoralist livestock keepers). The numbers of the latter vary according to season and the availability of water and grazing in the region. Activities that bring the population into areas where tsetse are present include the cultivation of crops (sorghum, sesame, sweet potatoes), fishing, collection of wood and beekeeping. The Mandoul area is a historical focus of sleeping sickness. It was included in the “secteur de prophylaxie numéro 3” created in 1919 and was visited by the medical teams led by Gaston Muraz in 1928 [9, 10]. Two species of tsetse were historically recorded in the area: Glossina fuscipes fuscipes and Glossina morsitans submorsitans [11, 12].
A census of the human population was conducted in 2013 to quantify the number of people and their geographical distribution. The number of persons living in households was recorded and the location recorded using a Global Positioning System (Garmin 64s). For each household, the name of the head person and the number of inhabitants were recorded. The main roads, tracks and paths leading between settlements and towards the swamp were mapped, as well as the routes used to cross the swamp.
Prior to the beginning of the tsetse control campaign, entomological baseline data were collected in November 2013 using 108 biconical traps deployed for 48 hours. Along the banks of the Mandoul River, traps were set in pairs, with each trap being at least 100 m apart and pairs of traps deployed at intervals of ~2 km. In addition, traps were also arranged in a transect running from the riverbank to the savannah where possible. Traps were also deployed at several jetties which were likely to be sites of human-tsetse contact.
Tiny targets comprising 0.25m × 0.25m blue polyester flanked by 0.25m × 0.25m black polyethylene netting [6] impregnated with deltamethrin at 300mg/m2 were used to kill but not monitor tsetse. Targets were obtained commercially from Vestergaard-Frandsen (Lausanne, Switzerland). They were deployed in the forest gallery along the Mandoul River, in places frequented by people identified during the population census and where tsetse were caught during the pre-intervention survey. Tsetse were detected over a limited range along the Mandoul but targets were deployed up to 4 km beyond where tsetse were caught. The first deployment was done in January—February 2014, and all targets were replaced at the same period in 2015 and 2016. Targets were suspended from tree branches at 10–20 cm above the ground, using string, or erected with wooden sticks obtained locally (Fig 1).
Entomological monitoring was done using 44 geo-referenced sentinel traps which were selected according to both geographical coverage of the area and tsetse presence in the traps during the T0 survey. These sentinel traps were deployed for 48 h continuously at every monitoring survey to allow comparison between surveys unless stated otherwise.
Nine monitoring surveys (T1-T9) were implemented to assess changes in the catch of tsetse within the study area: the first one was undertaken in April-May 2014, 2–3 months after the deployment of targets, and thereafter during the months of March, May and October of the two following years (2015–2016). During the fourth and fifth surveys (T4, T5) respectively, additional traps and sticky traps (standard traps with the blue attractive part wrapped with transparent adhesive film; [13]) were deployed at sites where relatively high numbers of tsetse were caught during the T0 survey, in order to increase probability of catching tsetse. For the same purpose, traps were operated continuously for six consecutive days instead of the usual two during the sixth survey (T6). All surveys conducted in 2016 (T7-T9) reverted to the standard protocol with traps being operated for 48 h.
Data on HAT cases were obtained through the WHO HAT Atlas for 2000–2014 [4, 14] and also through the PNLTHA of the Republic of Chad for 2015. These epidemiological data come from both passive surveillance and active screening activities. Passive detection relies on self-presentation of symptomatic people to medical facilities. HAT cases were detected and treated according to the WHO algorithm [15], slightly modified [16] in order to allow examination of up 1500 people per day by active screening. These modifications essentially involved splitting of the medical team to examine more people. First, all screened people were tested with CATT (card agglutination test for trypanosomiasis) on whole blood, the negatives were released, and serum was taken from the positives. People positive for the CATT test done on serum up to 1/4 dilution but negative at 1/8 were released and considered as serological suspects. Parasitological examination consisted of capillary tube centrifugation (CTC), and/or direct examination of lymph nodes aspirates.
All the positive individuals to at least 1/8 dilution of the CATT test, but parasitologically negative (meaning no trypanosome detected) were considered as serological cases. Serological cases, as well as parasitological ones, where trypanosomes were observed directly through microscopic examination of blood and/or cerebrospinal fluid, were then treated with pentamidine for infections considered to be first stage, and with Nifurtimox and Eflornithine combination (NECT) for those in the second stage. Stage diagnosis was done using lumbar puncture and cerebrospinal fluid examination. In 2015 surveillance was enhanced in the Mandoul region by improving the capacity for passive screening through equipping health facilities with rapid diagnostic tests (RDTs) and upgrading confirmation and treatment centres [17].
Comparison of catches of tsetse from traps during the different monitoring periods were analysed using a collection of generalised linear mixed-effects models (GLMEs), with Poisson error distributions and log link functions [18]. Trap sites (n = 44) were grouped into 15 different spatial locations. A baseline fixed effect was associated with each spatial location. Climatic predictors considered were precipitation and temperature and the study design was included via an indicator for tiny target (intervention) deployment and the duration of monitoring trap deployment. Trap level random effects and overdispersion were considered in the models. The best model was selected to be the one with the lowest corrected Akaike Information Criterion (AICc) and also we considered others within 2 AICc as competing hypotheses [19, 20]. All statistical analyses were carried out using the MATLAB 2015a fitglme package.
In order to determine the unobservable impact of vector control upon the incidence of new infections, a previously developed mathematical model [21, 22] for Gambian HAT was modified. The mechanical model tracks human hosts in their various stages of HAT infection including stage 1 and stage 2 disease, as well as infection prevalence in the tsetse population (Fig 2 and Eq 2.1 in S1 Methodology). The model was fitted to the available epidemiological human case data using Metropolis Hastings MCMC to estimate the unknown parameters specific to this region, which include tsetse-human ratio, passive detection rate of stage 1 cases and underreporting. Model outputs of cases were stratified by their detection method (active/passive) in order to match to the data. Since infection prevalence in tsetse was unknown this could not be used for fitting, but the model can be used to provide an estimate of tsetse prevalence based on knowledge of the tsetse-host transmission cycle and fitting to the human data (Fig 3 in S1 Methodology). Case data from 2000–2013 was used to fit the model, while 2014 and 2015 were predicted based on known active screening levels continued passive surveillance and tsetse population reduction. These years were therefore used as validation years to test the predictive ability of the model.
In 2015, passive surveillance was expanded by equipping health facilities with new diagnostic tools. To account for this step-change in passive strategy in the model, it was assumed that this would speed up the detection of both stage 1 and 2 cases, and lead to less underreporting. At present, no published study has yet ascertained the quantitative impact of this type of enhanced screening upon the detection rate. The single (unstaged) data point in 2015 is unfortunately insufficient to generate robust estimates here. Therefore model predictions generate approximations for the qualitative change that would be anticipated under the model by doubling the passive detection and reporting rates for 2015 onwards.
The model provides estimates for the decline in new infections, but also the predicted impact of both medical and vector controls on future HAT reporting and transmission. Counterfactual model simulations with medical interventions but no vector control were conducted to establish the expected impact that active and passive screening would have had in the absence of vector interventions. The predictions were used to establish the feasibility of achieving elimination as a public health problem, defined as less than 1 reported case per 10,000 people, by 2020. This mathematical modelling analysis was conducted using Matlab software (see S1 Model Code). Further model details can be found in the S1 Methodology.
A total of 114 human settlements were identified and recorded, comprising 22 encampments (<100 inhabitants), 70 hamlets (100–500 inhabitants) and 27 villages (>500 inhabitants) (Fig 2). Out of the 38 674 inhabitants counted, 1029 were located in encampments, 17 629 in hamlets, and 20 016 in villages. Therefore, in the intervention area (840 km2), the human population density can be estimated to be around 46 inhabitants per square kilometre. The networks of tracks between settlements and towards the swamp were recorded as well as and the main crossing tracks in the swamp (Fig 2).
A total of 145 tsetse were caught by the 108 biconical traps during the T0 survey, all belonging to G. f. fuscipes, with 50 males (34.5%) and 95 females (65.5%). Apparent density in the whole area was 0.65 tsetse/trap/day, ranging from 0 to 26 tsetse/trap/day. Out of the 108 traps, 22 (20.4%) caught tsetse and they were all located on the two main branches of the Mandoul River (Fig 2, S1 Fig).
For the first deployment in 2014, 2600 targets were deployed on both sides of the river and its tributaries. These targets were replaced one year later with new ones along with an extra 108 additional targets, in order to improve the area coverage (Fig 2), giving a total number of 2708 targets. Out of the 840 km2 of the intervention area of the Mandoul focus, the actual area on which targets were deployed represents 45 km2, corresponding to the gallery surrounding the portion of the Mandoul River where tsetse were found. Hence the overall target density can be calculated either as 3.2 targets/km2 if based on the whole intervention area, or 60 targets per linear km if only considering the riverine system.
The total catch of tsetse from 44 sentinel traps was 145 tsetse (1.62 flies/trap/day). At the first evaluation post target deployment (T1, 2 months later), only 2 tsetse were caught (0.02 tsetse/trap/day) were observed (Fig 3). The same density was again observed during the second evaluation (T2) but thereafter no tsetse was caught during the next three successive surveys (T3-T6), even with the additional traps and sticky traps deployed to increase the probability of capture. During the sixth survey (T6, October 2015), trapping duration was extended to six consecutive days and one tsetse was caught. No tsetse were caught during the three surveys conducted in 2016 (T7-T9).
Statistical analysis using GMLE resulted in two selected models differing by only one effect (see Section 1 in S1 Methodology for more detailed results). The best model had only location specific baselines, the pre-target deployment indictor (p<10−18, coeff = 5.3 i.e. a predicted 99.5% reduction due to tiny target intervention) and an overdispersion effect. The other selected model also included a precipitation fixed effect, but the pre-target deployment indicator remained strongly significant (p<10−11, coeff = 5.4).
Model fitting to human case data indicates that active screening has played a major role in the reduction of new HAT infections between 2000 and 2013. Over this time there is an estimated 71.5% (95% CI: 67–75%) reduction in new transmissions (see Fig 4). This trend is not so apparent in the detected HAT cases due to high fluctuation in screening levels during that time period, leading to a variable detection rate. Despite this ambiguity, it is noted that from 2009 onwards, detected cases remained at relatively low levels despite annual screening at moderate levels (see Fig 5). It is seen that passive screening plays a variable role in case detection, and the model can capture this trend. For example with more cases being detected by passive surveillance in 2008 following no active screening in 2007/2008.
To assess the robustness of model predictions, an additional fit was performed using only epidemiological data for 2000–2006, and used 2007–2013 as validation years without vector control and 2014–2014 as validation years with vector control. The model dynamics obtained were very similar to those presented here, despite using only the first half of the data set (see Section 3.2 in S1 Methodology for detailed analysis).
Projecting 2014 and 2015 was performed using four different strategies: (1) basic medical, using known levels of active screening and assuming passive detection continued unchanged in 2015, (2) improved medical, using known active screening and assuming passive detection and reporting rates doubled in 2015, (3) basic medical and vector control, the same as basic medical with vector control from 2014, and (4) improved medical and vector control, the same as improved medical with vector control from 2014. Strategy 4 represents the strategy that occurred in 2014–2015 and is coloured purple in Figs 4 and 5. It is seen that, using the actual strategy, the model captures the trend in reported cases for 2014/2015. Comparing these years to the counterfactual strategies (1–3) predictions demonstrates the additional benefit of having added tsetse control and/or enhanced passive surveillance to the existing medical intervention.
Mathematical modelling estimates that, without vector control, there would have been 129 (95% CI, 105–157) and 123 (95% CI, 97–150) cases per year in 2014 and 2015 respectively with the reduction due to active screening. In reality only 90 and 47 cases were observed (in 2014 and 2015) when a combined vector and enhanced medical strategy took place; in 2013 there were 186 cases. Modelling results indicate that 29.6% (95% CI: 5–49%) of this two-year decline was through active screening, and the remaining 70.4% (95% CI: 51–95%) was due to vector control. In 2015, the new passive strategy was predicted to result in slightly more case reporting than the basic medical strategy and so was not attributed to this particular decline.
Likewise, new infections are calculated to have fallen dramatically. The model surmises that transmission fell from around 311 new infections per year in 2013 to a mean of 7 and 0 new infections per year in 2014 and 2015 respectively. Over these two years, 28.7% (95% CI: 25–33%) of this decline is attributed to active screening, 8.5% (95% CI: 8–9%) to passive and 62.8% (95% CI: 59–66%) to vector control.
Projecting forwards to 2020 using the model gives an optimistic forecast if vector control and screening continue at the 2015 level. Continuing this combined strategy is predicted to result in fewer than 2 annually reported cases and zero transmission in 2020; this strategy appears sufficient to reach elimination as a public health problem by 2018 and full elimination over a decade in advance of the WHO 2030 goal. It is noted that the model predicts this will occur with or without enhanced passive screening. Under the counterfactual strategy without vector control, the reported cases and transmission may have continued to decline steadily, reaching around 47 (95% CI: 31–67) reported cases in 2020 without enhanced passive screening; although this would have still been substantially higher than the 3.9 reported cases threshold that is necessary to achieve elimination as a public health problem in the focus. Results from the counterfactual strategy without vector control but with enhanced passive detection, predicted that there would have been even fewer reported cases compared to the basic medical strategy, with 20 (95% CI: 7–25) reported cases in 2020, however this would still be above the target threshold.
This study shows that adding tsetse control to the ‘screen and treat’ strategy had a marked impact on the transmission of sleeping sickness in the Mandoul focus. This echoes the recent findings from Guinea where tsetse control also resulted in a significant decrease of the incidence of HAT [3].
The total number of cases diagnosed and treated (1143 cases since 2009 out of a population of 38 674 people) suggests that up to 3.0% of the people in this area have had sleeping sickness. A huge medical effort was conducted in this focus with 142 467 people being screened between 2009 and 2015. Given the population number is ~38 674, it means that on average, every individual in the area was screened nearly four times during this period. Despite this effort, a substantial decrease in HAT reporting occurred only when vector control was added to the medical strategy. In 2014, there was a decrease in the stage 1 to stage 2 detection ratio for both active and passive surveillance. For active screening this ratio was 1.5 stage 1 cases for every stage 2 case which decreased from 2.7 in 2012 and 2013 when a similar level of active screening was achieved. For passive surveillance there were 0.16 stage 1 cases per stage 2 case in 2014, which decreased from 0.22 and 0.24 in 2012 and 2013 respectively. This relative decline in the stage 1 to stage 2 case ratio supports the hypothesis that vector control has reduced new transmissions, and is expected to continue to decline over the coming years whilst tsetse control remains in place, although the improved passive surveillance that began in 2015 could skew this ratio. Modelling supports the assertion that tsetse control has greatly impacted past reporting and transmission, and will continue to do so whilst the fly population remains supressed.
It is acknowledged that there are several assumptions that effect the results generated by the mathematical modelling results presented here, many of which have been discussed elsewhere [21, 22]. Given the relatively small size of the Mandoul focus, it is unlikely that spatial effects would alter the presented results, however temporal heterogeneities arising from unknown dates of screenings could explain some of the discrepancies between reported HAT cases and model outputs. Furthermore, the dynamic model does not take importations from other areas into account, which could result in other small differences between the model and data. By additional analysis which used 2007–2015 as validation years (see S1 Methodology), the model seems to perform well at mid-range prediction, providing evidence for the model’s forecasting ability and justification for its use here as a tool to evaluate intervention strategies.
One particular challenge was to estimate the impact of new enhanced medical strategies, which began in the region in 2015, on case detection and reporting. Unfortunately there was little information with which to accurately parameterise the mathematical model to account for this change in strategy, although the presented results demonstrate the expected qualitative impact that this has had, and should continue to have, in Mandoul. The modelling results demonstrate that it very possible that improved screening actually increased the number of passively detected cases in 2015, but would be expected to lead to reduced cases in subsequent years by reducing transmission. As there was uncertainty surrounding this strategy, results without improved screening were also shown for comparison. Future work using other data sets could help to better inform model parameterisation of this type of enhanced medical strategy.
For vector control, the lack of a control (non-intervention) area is a limitation for the study. However, the aim of the intervention was to protect the people of Mandoul from sleeping sickness. This is an isolated focus without a comparable area, which could have served as an appropriate control. In the absence of a control, empirical data, gathered before and after entomological intervention, has been used in conjunction with models to estimate its relative impact.
The vector in the Mandoul area is G. f. fuscipes as reported previously [23]. G. morsitans submorsitans was captured by surveys conducted 20 years ago [11], but none was caught during the current surveys, or previous ones in the area [23]. This confirms its disappearance in places where human encroachment is high and where wildlife disappears, as reported elsewhere [24, 25].
In the Mandoul focus, the medical ‘screen and treat’ strategy saved many lives but the addition of vector control has driven the incidence to even lower levels (0.38%) and this may have interrupted transmission of T. brucei gambiense. Why was this incidence not achieved without vector control? Among several possible causes, the geographical situation and human behaviour in this area may be a key parameter, especially human mobility. The human population frequently crosses the tsetse-infested swamp. The population between the two banks come from the same ethnic group and often belong to the same families. It means that for any social event, people need to cross the swamp. Moreover, there is an important local market and facilities (healthcare, veterinary, school, local government) in the town of Bodo which is visited frequently by local people, many of whom must cross the swamp to reach Bodo.
The rapid and drastic decrease in catch of tsetse suggests that tiny targets can control G. f. fuscipes in this setting. The small size of the area, well defined with limited habitat for tsetse and its relative isolation with low reinvasion pressure may have contributed to the success of the intervention. This contrasts with two other foci where tiny targets have been recently used to control tsetse. In the Boffa focus of Guinea, extensive and difficult-to-access mangrove made vector control (G. palpalis there) much more difficult [5, 26]. Similarly in the Arua focus of northern Uganda, the widely distributed population of G. f. fuscipes was not driven to the low (i.e., close to zero) levels seen in Chad [8].
Present results further demonstrate the efficacy of combining medical interventions with vector control to halt T. b. gambiense transmission, in certain epidemiological settings. Unlike medical interventions, which reduce disease duration and consequently decrease the number of new infections, tsetse control directly prevents people from becoming infected. As neither vaccine nor chemoprophylaxis for HAT exist this provides a valuable protective tool. As in Chad, Uganda and Guinea, this strategy can be applied in other foci where HAT still exists despite medical surveys and treatment of cases. While the use of tiny targets was effective in this operation, other methods may be appropriate elsewhere. For instance, many cattle are present in Mandoul. These animals, owned by sedentary farmers and more mobile pastoralists, are at risk of animal African trypanosomiasis [23] and hence controlling tsetse, using targets and/or treating cattle with insecticide, will also have a positive consequence on both animal and human health. This means that with minimal additional efforts, expanding this one health approach could benefit to people and their animals in this area, as certainly in other areas of the same type [27].
|
10.1371/journal.ppat.1000341 | An Epstein-Barr Virus Anti-Apoptotic Protein Constitutively Expressed in Transformed Cells and Implicated in Burkitt Lymphomagenesis: The Wp/BHRF1 Link | Two factors contribute to Burkitt lymphoma (BL) pathogenesis, a chromosomal translocation leading to c-myc oncogene deregulation and infection with Epstein-Barr virus (EBV). Although the virus has B cell growth–transforming ability, this may not relate to its role in BL since many of the transforming proteins are not expressed in the tumor. Mounting evidence supports an alternative role, whereby EBV counteracts the high apoptotic sensitivity inherent to the c-myc–driven growth program. In that regard, a subset of BLs carry virus mutants in a novel form of latent infection that provides unusually strong resistance to apoptosis. Uniquely, these virus mutants use Wp (a viral promoter normally activated early in B cell transformation) and express a broader-than-usual range of latent antigens. Here, using an inducible system to express the candidate antigens, we show that this marked apoptosis resistance is mediated not by one of the extended range of EBNAs seen in Wp-restricted latency but by Wp-driven expression of the viral bcl2 homologue, BHRF1, a protein usually associated with the virus lytic cycle. Interestingly, this Wp/BHRF1 connection is not confined to Wp-restricted BLs but appears integral to normal B cell transformation by EBV. We find that the BHRF1 gene expression recently reported in newly infected B cells is temporally linked to Wp activation and the presence of W/BHRF1-spliced transcripts. Furthermore, just as Wp activity is never completely eclipsed in in vitro–transformed lines, low-level BHRF1 transcripts remain detectable in these cells long-term. Most importantly, recognition by BHRF1-specific T cells confirms that such lines continue to express the protein independently of any lytic cycle entry. This work therefore provides the first evidence that BHRF1, the EBV bcl2 homologue, is constitutively expressed as a latent protein in growth-transformed cells in vitro and, in the context of Wp-restricted BL, may contribute to virus-associated lymphomagenesis in vivo.
| Cancer almost always develops through the cumulative effects of several independent changes in the target cell. For certain tumors, one step in the chain involves infection of the cell with a particular type of virus. The best example is Burkitt lymphoma (BL), a tumor of B lymphocytes which develops through the combined action of a genetic accident leading to uncontrolled expression of the c-myc oncogene and infection with a common herpesvirus, the Epstein-Barr virus (EBV). Recent evidence suggests that, although latent EBV infection can itself drive B cell growth, the virus plays a different role in the context of BL, namely to counteract the naturally poor survival ability of c-myc–expressing cells while leaving their c-myc–driven growth intact. Here we show that EBV achieves this by unexpectedly switching on a viral protein that was thought never to be seen in latent infection; this viral protein resembles one of the cell's own key survival proteins called bcl2. Furthermore, the work has led us to realise that this virally encoded bcl2-like protein is not only important in the context of BL but, contrary to conventional wisdom, is actually part of EBV's natural strategy for B cell growth transformation.
| Burkitt lymphoma (BL) is a human tumor of B cell origin whose pathogenesis involves complementation between a defined cellular genetic change, translocation of the c-myc oncogene into an active immunoglobulin (Ig) locus, and a B cell-transforming virus, Epstein-Barr virus (EBV) [1],[2]. C-myc deregulation appears to be the crucial lymphomagenic event, since all BLs worldwide carry a c-myc/Ig translocation, and in model systems, expression of c-myc from such a construct converts B cells to the proliferating BL phenotype [3]–[6]. There is, nevertheless, strong selection for EBV as a complementing agent. Thus all cases of BL in its high incidence (endemic) form are EBV genome-positive, as are 15–85% of the low/intermediate incidence (sporadic) BLs seen elsewhere in the world [7]. However, the virus' role in BL pathogenesis has remained obscure, not least because EBV gene expression in tumor cells does not mirror that of a typical growth-transforming infection.
To illustrate the point, Figure 1 (bottom line) shows the pattern of latent gene expression established when EBV transforms normal B cells into permanent lymphoblastoid cell lines (LCLs) in vitro. This entails expression of the non-coding EBER RNAs, the BamHIA rightward transcripts (BARTs) from which most of the EBV micro(mi)-RNAs are derived [8],[9], six nuclear antigens (EBNA1, 2, 3A, 3B, 3C and –LP) and three latent membrane proteins (LMP1, 2A and 2B). The LMPs are each expressed from their own EBNA2-activated promoters. By contrast, the individual EBNA mRNAs are generated by differential splicing of long primary transcripts, initiated immediately post-infection from the BamHIW promoter, Wp, and later from an adjacent pan-EBNA promoter, Cp [10]. Interestingly, the same cDNA cloning studies that first characterised the EBNA mRNAs also identified rare clones that spliced downstream of EBNA2 into BHRF1 [11]–[13], a gene later recognised as a viral homologue of cellular bcl2 [14]. However the BHRF1 protein could never be detected in tightly latent LCLs and it was subsequently identified as an early lytic cycle protein [15] expressed from its own lytic cycle promoter [12]. More recently, transcription of the BHRF1 gene has been detected in freshly-infected B cells but, because this was transient and accompanied by a number of other lytic gene transcripts, it was thought to reflect opportunistic transcription from the incoming un-methylated virus genome [16].
As also illustrated in Figure 1 (upper lines), EBV-positive BL tumors display quite different forms of latency, with many of the transforming proteins being down-regulated. Most BLs express the EBERs, BARTs and just one protein, EBNA1, from an EBNA1-specific promoter, Qp [17]–[19], a form of infection referred to as Latency I. The resident EBV genome in these tumors is wild-type and transformation-competent [17]. However, we recently identified another subset (around 15%) of BLs where, in addition to EBNA1, the EBNA 3A, 3B, 3C proteins and in some cases also a truncated form of EBNA-LP were expressed, always in the absence of EBNA2 and the LMPs [20]. This reflected the use of a different transcriptional programme, called “Wp-restricted latency”, where the EBNA transcripts all derive from the Wp promoter. Interestingly, another defining feature of “Wp-restricted” tumors was the presence of a mutant EBV genome with a deletion removing the EBNA2 gene and some adjacent upstream and downstream sequences. Note that, although EBNA2-deletion events are sometimes detectable by PCR amplification at sites of lytic virus infection in the oropharynx [21],[22], viruses with such a deletion are defective in transformation and have rarely if ever been detected as latent infections of the normal B cell pool. Therefore, detecting such rare mutant genomes in a significant number of BLs strongly suggests that infection with an EBNA2-deletion mutant (or, more likely, the Wp-restricted latency with which such infection is associated) has markedly increased EBV's potential to act as a cofactor in BL development.
As to what that co-factor role might be, studies of c-myc-driven B cell lymphomagenesis in mouse models have shown that complementary changes often act by counteracting the pro-apoptotic effects of high c-myc expression, thereby giving free rein to myc-driven proliferation [23],[24]. Indeed the first evidence suggesting an anti-apoptotic role for the virus in BL came from work with a Latency I BL line, Akata-BL, where EBV-positive sub-clones proved to be slightly less sensitive to apoptotic stimuli than sub-clones that had lost the EBV genome during in vitro passage [25]. This has prompted a large volume of work attempting to identify which Latency I gene products provide a survival advantage to BL cells, with evidence of anti-apoptotic potential being reported for the EBERs [26]–[29], for EBNA1 [30] and most recently for a BART-derived mi-RNA [31] in different experimental contexts. However the levels of apoptosis protection mediated by Latency I infection in vitro are relatively slight and the underlying mechanism remains to be fully resolved. It was therefore notable that Wp-restricted BL cell lines were much more resistant to cell death triggers than either EBV-negative or Latency I BL lines [32]. This observation strongly reinforced the idea that EBV's role in BL pathogenesis was to counteract the pro-apoptotic influence of deregulated c-myc expression. Our objectives in the present work, therefore, were (i) to identify the viral gene expressed in Wp-restricted but not Latency I infection, that was responsible for this large increment in apoptosis protection, and (ii) to determine whether the effect was unique to an EBNA2-deleted virus acting in the context of a BL cell or might be highlighting a Wp-associated function that is a natural feature of wild-type virus infections.
The standard Latency I BL cell lines, Rael-BL, Sav-BL, Kem-BL and Akata-BL, and the Wp-restricted BL cell lines, Sal-BL, Oku-BL and Ava-BL, have been described previously [20], as have the Awia-BL cell line and derived single cell clones (EBV-negative, Latency I and Wp-restricted) and the Awia-LCL [33]. An EBV genome-loss clone of Akata-BL (EBV-loss Akata-BL) was isolated by single cell seeding of the Akata-BL parental cell line. All BL cells were maintained in RPMI 1640 (Invitrogen) containing 10% (vol/vol) selected fetal calf serum and 2 mM glutamine (standard medium), further supplemented with 1 mM pyruvate, 50 µM alpha-thioglycerol and 20 nM bathocupronine disulfonic acid. LCLs, all maintained in standard medium, were generated from the peripheral blood B cells of healthy control donors by infection with wild-type B95.8 strain EBV (WT-LCLs) and with B95.8-derived recombinant viruses lacking an intact BHRF1 gene (BHRF1KO-LCLs) [16] or BZLF1 immediate early lytic gene (BZKO-LCLs) [34]. Note that WT-LCLs typically contain 1–3% of cells in lytic cycle whereas BZKO-LCLs do not contain any lytically-infected cells since the immediate early BZLF1 gene is essential for initiation of the lytic cycle [34]. To provide a reference culture enriched in lytically-infected cells, Akata-BL cells were treated with anti-IgG (Cappell) at a concentration of 0.1% (vol/vol) for 72 hours to induce EBV lytic replication in up to 60% cells [35].
Inducible gene expression was achieved using pRTS-CD2, a derivative of the pRTS-1 expression plasmid [36]. This plasmid carries a truncated rat CD2 gene, the EBV origin of replication (oriP) and the EBNA1 gene (encoding the viral genome maintenance protein), in addition to a bi-directional doxycycline (dox)-regulated promoter controlling expression of GFP and truncated NGF receptor in one direction and the EBV gene of interest in the other direction. Plasmids were constructed that carried the EBV (B95.8 strain) genes encoding either EBNA3A, EBNA3B, EBNA3C or BHRF1. The BHRF1 construct contained a minimal cDNA with no other flanking EBV sequence; as a non-coding control, we also generated a mutated construct (mut-BHRF1) in which the start codon ATG of the above BHRF1 cDNA had been changed to TAG.
The pRTS-CD2 derived expression plasmids (10–15 µg DNA) were electroporated, either alone or in combinations, into 107 Sav-BL, Akata-BL and EBV-loss Akata-BL. Cells were allowed to recover in culture overnight before isolating viable cells by density centrifugation followed by separation of rat CD2-expressing transfected cells by magnetic cell sorting using OX34 anti-rat CD2 antibody and MACS anti-mouse IgG2a/b beads (Militenyl Biotech) according to the manufacturer's guidelines. Cultures were expanded and maintained in standard medium. To induce expression of GFP and the gene of interest, dox was titrated into the medium at concentrations from 1 ng/ml to 1 µg/ml for 24 hrs. Typically this procedure yielded cultures in which 30–80% cells stably carried the plasmid; the remaining 20–70% cells lacked the plasmid and served as internal controls. As an additional control in all experiments, Sav-BL, Akata-BL and EBV-loss Akata-BL cells were also transfected with a control plasmid which lacked any EBV gene insert but carried dox-inducible GFP and the truncated NGFR. Cultures were established using the same protocol as above and, following dox-induction, GFP-positive and GFP-negative cells were compared in the same way.
In the experiment to demonstrate that GFP expression correlated with expression of the inserted EBV gene of interest, Akata-BL cells stably transfected with the pRTS-CD2 EBNA3C expression plasmid were exposed to 1 µg/ml dox for 24 hrs and then sorted using a FACS Vantage into GFP-positive and GFP-negative populations. These cell populations were smeared onto microscope slides and fixed in ice-cold methanol∶acetone (1∶1 vol/vol ratio) at −20°C for 20 minutes prior to immunofluorescence staining for EBNA3C. The slides were incubated for 30 minutes at 37°C in blocking buffer (1×PBS containing 10% heat inactivated normal goat serum) to prevent non-specific antibody staining, before being stained for 1 hr at 37°C with an antibody specific for EBNA3C (E3CA10 [37]) used at a concentration of 5 µg/ml diluted in blocking buffer. Cells were washed three times in 1×PBS and then stained with a goat anti-mouse Alexa Fluor fluorochrome 594 conjugated secondary antibody (Invitrogen) at a dilution of 1 in 1000 in blocking buffer. Cells were washed three times in 1×PBS, mounted in VECTASHIELD medium containing 4′,6 diamidino-2-phenylindole (DAPI) (Vector Labs) before being visualised on a epifluorescence microscope.
Immunoblotting was carried out as described previously [20] using mAbs to: EBNA1 (1H4), EBNA2 (PE2), EBNA3C (E3CA10), LMP1 (CS1-4), BZLF1 (BZ-1) (all used at dilutions of 1 in 50) [20]; BHRF1 (5B11: Millipore, used at a dilution of 1 in 1000), Calregulin (H-170, Santa-Cruz Biotechnology, used at a dilution of 1 in 1000) and polyclonal antibodies specific for EBNA3A and 3B (Exalpha Biologicals, Maynard, MA; the antibodies were used at a dilution of 1 in 1000 to detect EBNA3A and 1 in 500 to detect EBNA3B) and for PARP1 N-terminal region (H-300, Santa-Cruz Biotechnology, used at a dilution of 1 in 1000). All immunoblots were repeated several times on different protein samples.
To quantify mRNA expression, total RNA extraction and cDNA synthesis was carried out as described previously [38]. Quantitative Taqman (Q)RT-PCR assays specific for Wp-initiated, Cp-initiated, Qp-initiated, EBNA2 and LMP1 latent mRNA transcripts and for BZLF1 (immediate early) and gp350 (late) lytic transcripts are described previously, as are the cell lines used as positive controls for each assay [38],[39]. In addition, expression of the early lytic gene BMLF1 was assayed using a new QRT-PCR assay involving a cDNA primer (5′-GAGGATGAAATCTCTCCAT-3′) and the primers (5′- CCCGAACTAGCAGCATTTCCT-3′) and (5′-GACCGCTTCGAGTTCCAGAA-3′) with a FAM labelled probe (5′-AACGAGGATCCCGCAGAGAGCCA-3′). To quantify BHRF1 expression, we designed two assays using a common cDNA primer (5′-TTCTCTTGCTGCTAGCT-3′), reverse primer (5′-TCCCGTATACACAGGGCTAACAGT-3′) and FAM labelled probe (5′-AATAGGCCATCTTGCTCTACAAGATCTGGCA-3′) all within the BHRF1 coding HF exon, but in combination with one of two different forward primers. Latent BHRF1 transcripts were detected using a forward primer either in the Y2 exon (5′-GAGGATGAAGACTAAGTCACAGGCTTA-3′) or in the W2 exon (5′-TGGTAAGCGGTTCACCTTCAG-3′). Note that both of these upstream primers will detect latent BHRF1 transcripts in WT-LCLs, but only the W2 primer will detect latent BHRF1 transcripts in Wp-restricted lines where the deletion has removed the Y2 exon. A standard LCL with 3% of cells in lytic cycle was used as the positive control for the RT-PCR assays detecting lytic BMLF1, BZLF1 and gp350 transcripts and was assigned an arbitrary value of 1. For quantifying the latent W2-BHRF1 and Y2-BHRF1 spliced transcripts an LCL derived from a lytic cycle-deficient BZKO virus was used as a positive control cell line and assigned an arbitrary value of 1. All QRT-PCR assays were carried out in triplicate and all experiments were conducted on at least three occasions.
cDNA was generated as described above using the BHRF1 specific primer. An aliquot of 50 ng cDNA was amplified in a conventional PCR reaction using Expand High Fidelity DNA polymerase (Roche) and the W2 and BHRF1 PCR primers described above. Briefly the cDNA samples were heated to 95°C for 5 minutes before being subjected to 1 minute incubations at 95°C, 59°C, 72°C for 35 cycles. The W2-BHRF1 PCR products were loaded and run on an 8% polyacrylamide gel in order to get good separation of the 110–265 base pair products (the size of the product depends upon the splicing pattern of the transcript in the different cell lines, see Figure S3). The most intense bands were excised from the gel and the DNA extracted and purified. The DNA PCR product was sequenced using the W2 and HF primers described above on an Applied Biosystems ABI 3700 automated sequencer (carried out by the Functional Genomics Laboratory at the University of Birmingham).
For the standard panel of Awia-BL clones, 3×104 cells were seeded into wells of a flat-bottomed 96 well plate and treated with either a final concentration of 0.25–1 µg/ml ionomycin (Sigma) or 2.5–10 µg/ml anti-IgM antibody (ICN Flow) at 37°C. Following 48 hrs ionomycin treatment or 72 hrs anti-IgM treatment, cells were harvested, washed in 1×PBS and resuspended in 0.5 ml saline (pre-warmed to 37°C). Syto 16 (Molecular probes Europe, Leiden, The Netherlands) was added to the cells at a final concentration of 25 nM and incubated at room temperature for 1 hr, at which time 2.5 µg/ml propidium iodide (Sigma) was added and the cells analysed immediately on a flow cytometer. A two-dimensional dot plot was generated of Syto 16 fluorescence (y-axis) versus propidium iodide fluorescence (x-axis). Syto 16 will only stain viable cells whereas propidium iodide will preferentially enter necrotic cells [40],[41]. Viable cells (Syto 16 +ve, propidium iodide −ve), apoptotic cells (Syto 16 −ve, propidium iodide −ve) and necrotic cells (Syto 16 -ve, propidium iodide +ve) can therefore be distinguished. Data for 5,000 cells was collected for each cell line.
For the Akata-BL (parental and EBV-loss) and Sav-BL cultures stably carrying dox-regulatable expression plasmids, cells were plated out at a concentration of 2×104 cells per well in a flat-bottomed microtitre plate in media alone or media supplemented with an appropriate concentration of dox (1–1000 ng/ml dox). Cells were then incubated overnight at 37°C, 5% CO2 for the expression of GFP and the EBV gene of interest to be induced. The cells were then exposed to the apoptosis inducers, anti-IgM (10–20 µg/ml) or ionomycin (5 µg/ml), and apoptosis assayed in GFP-positive and GFP-negative cells within the same population 48–72 hrs later using propidium iodide (PI 2.5 µg/ml) to identify dead cells. In these experiments cells were not dually stained with Syto 16 because this dye is detected by flow cytometry in the same channel as GFP. Cultures were then analysed by flow cytometry immediately for GFP versus PI staining and results expressed as the percentage death induction within the GFP-positive and GFP-negative fractions. All apoptosis assays were carried out on triplicate cultures on each occasion of testing, and each experiment was carried out on at least three independent occasions.
As an additional measure of apoptosis, 1×105 Akata-BL cells were plated out in multiple wells of a 24 well plate. To some wells, dox was added to a final concentration of 500 ng/ml and the cells incubated overnight at 37°C to allow the expression of GFP and the gene of interest to be induced. Ionomycin was then added to all the wells at a final concentration of 5 µg/ml and the cells incubated at 37°C for 18 hrs. The cells were then harvested, washed twice in 1×PBS and the protein extracted. Western blotting was carried out on 20 µg protein and the membranes probed with an anti-PARP 1 antibody specific for the N-terminal region (H-300, Santa-Cruz Biotechnology, used at a dilution of 1 in1000).
To analyse events occurring soon after EBV infection in vitro, B cells isolated from adult peripheral blood mononuclear cells (PBMCs) by positive selection using M-450 CD19 Dynabeads (Dynal) were exposed to recombinant EBV (WT, BZKO or BHRF1KO virus) at a MOI of 100 overnight at 37°C, then resuspended in fresh media and plated out at a concentration of 4×106 cells per well of a 24 well plate. At each time point (0, 8, 12, 24, 48, 72, 120 hours post-infection) cells were harvested for RNA (4×106 cells) and protein (8×106 cells). All infections were carried out on at least three independent occasions.
These experiments involved both freshly-infected B cells studied 4 and 8 days post infection and established LCLs as targets, infections being carried out using recombinant WT virus, the BZKO virus or the BHRF1KO virus. In each case cells were isolated from individuals of known HLA type, positive for DRB1*0401 and DRB1*1501 restricting alleles, or (as a control) from donors mismatched for the alleles. Target cells pre-pulsed for 1 hr with 5 µg/ml relevant epitope peptide served as a positive control. To assay T cell recognition standard numbers (2000 cells per well) of CD4+ T cells specific for the HLA-DRB1*0401-restricted BHRF1 122-133 epitope (designated PYY, [42]), the HLA-DRB1*0401-restricted EBNA2 11-30 epitope (designated GQT, [43]) or the HLA-DRB1*1501-restricted gp350 61-81 epitope (designated LDL, [44]) were incubated in 200 µl medium in 96-well V-bottom plates with 105 target cells per well. The supernatant medium, harvested after 18 hrs, was then assayed for IFNγ by ELISA (Perbio) in accordance with the manufacturer's protocol. All T cell assays were conducted in triplicate and all experiments on freshly infected cells and on established LCLs were conducted on at least three independent occasions.
The numerical data derived from the QRT-PCR and apoptosis assays were statistically analysed using the computer program GraphPad Prism 4 (GraphPad Software, CA, USA). For the QRT-PCR assays, the normalised values from all the replicates of the Wp-restricted and Latency I BL cell samples were compared using an unpaired student t-test (two-tailed, 95% confidence interval). For the apoptosis assays performed on the Awia-BL cell lines (Figure S1B), triplicate values for each cell line were used and an unpaired student t-test (two-tailed, 95% confidence interval) employed for the following comparisons; EBV-negative BLs to Latency I BLs and Wp-restricted BLs to EBV-negative and Latency I BLs. For the apoptosis assays performed on the cell lines carrying the pRTS-CD2 plasmids, the values for the percentage death induction in the GFP-positive and GFP-negative cells within each population from triplicate cultures were analysed. Since the individual readings were derived from the GFP-positive and GFP-negative cells within the same culture, here we carried out a paired student t-test (two-tailed, 95% confidence interval) to compare death induction in the GFP-positive (plasmid-positive) cells to the GFP-negative (plasmid-negative) cells.
The strength of protection from cell death offered to BL cells by a Wp-restricted form of infection is best illustrated in the context of an isogenic system. Awia-BL is an endemic tumor with a characteristic t(8∶14) c-myc/Ig translocation from which we were able to isolate Wp-restricted, Latency I and EBV genome-loss clones in early passage [33]. Figure S1A shows an immunoblot of EBV latent protein expression in these cells, indicating that Wp-restricted clones are distinct from Latency I clones in expressing the EBNA3 proteins in addition to EBNA1. Figure S1B shows representative data from experiments in which these same clones are subjected to graded doses of cell death triggers such as B cell receptor ligation (anti-IgM) or an intracellular calcium ionophore (ionomycin). We have previously shown that the cell death being induced in this system is largely via apoptosis, involving caspase cleavage [33]. Clearly, the Wp-restricted cells are resistant to triggering doses (10 µg/ml anti-IgM, 1 µg/ml ionomycin) that induce widespread death in Latency I and EBV-negative cells. By comparison, the protection being offered by Latency I infection in such assays is much less marked, with differential survival of Latency I compared to EBV-negative clones only being seen as a partial effect at lower anti-IgM and ionomycin doses. Such results strongly suggested that one or more of the viral genes that were exclusive to the Wp-restricted form of infection were responsible for a pronounced increment in cell survival capacity. Referring back to Figure 1, the obvious candidates in that respect were the EBNA3 proteins and/or the predicted product of a truncated EBNA-LP coding sequence (i.e. containing the repeat domains encoded by the W1 and W2 exons, but lacking the unique domains encoded by the Y1 and Y2 exons that are always removed by the deletion). In that regard, while Wp-restricted BL lines and clones are consistently EBNA3-positive, many lack detectable expression of EBNA-LP yet still retain strong apoptosis resistance in the anti-IgM and ionomycin assays [32]. Hence, even though truncated EBNA-LP has been associated with anti-apoptotic effects in some systems through an interaction with protein phosphatase PP2A [45], it could not be responsible for the global apoptosis protection observed in Wp-restricted BLs. At this point, therefore, we focused on EBNA3A, 3B and/or 3C as the potential mediators of protection.
In these experiments, we sought to avoid the problems of inter-clonal variability that can beset gene transfection and drug selection experiments in the BL system. Instead we used a new EBV ori-p-based plasmid, illustrated in Figure S2A, which is designed for stable maintenance as an episome in BL cells [36]. Vectored expression of a surface marker, rat CD2, early after transfection allows transfected cells to be enriched, generating a passageable culture in which typically 30–80% of cells carry the plasmid. Thereafter expression of the gene of choice can be induced in a dose-dependent manner by addition of dox. At the same time, dox-dependent co-induction of GFP allows one to distinguish the plasmid-positive cells by FACS staining (Figure 2A). Note, we have confirmed that GFP and the inserted EBV gene of interest are co-expressed in the same cells following induction of the bi-directional promoter with dox (Figure S2B). Initially, two Latency I BL lines (Sav-BL and Akata-BL) were transfected with vectors expressing either EBNA3A, 3B or 3C. Figure 2B (left) confirms that, in each case, expression of the gene of interest is tightly dox-dependent and can be induced either to physiologic (LCL-like) levels or much higher depending on the dox concentration. Following induction, the culture is subjected to apoptotic triggers and subsequently stained with propidium iodide (PI), thereby allowing comparison of the percentage of dead/dying cells in the GFP-positive (EBNA3-expressing) versus GFP-negative (control) fraction. Figure 2B (right) shows data from such an experiment on the Sav-BL background. None of the EBNA3 proteins, expressed individually, offered any apoptosis protection. We then carried out experiments in which the same Latency I lines were co-transfected with the EBNA3A and 3C vectors, since EBNAs 3A and 3C can act cooperatively to alter other aspects of the BL phenotype [46], or with all three EBNA3 vectors. The appropriate combinations of EBNA3 proteins were detectably induced in each case but again showed no evidence of apoptosis protection, either in Sav-BL (Figure 2C) or Akata-BL cells (data not shown).
In view of these results, we turned to the possibility that viral gene expression in Wp-restricted BL cells was more extensive than first thought and that other anti-apoptotic candidates, perhaps inappropriately expressed as a consequence of the EBNA2 gene deletion, had been missed. In that regard, Figure 3A illustrates the position of the deletions in Wp-restricted BL lines in relation to the EBV genome as a whole. Note that, in the wild-type genome, the whole BamHI W fragment is tandemly reiterated to form a large internal repeat that lies immediately upstream of the BamHI Y fragment containing the EBNA2 gene. While all four Wp-restricted lines analysed (Awia-BL, Sal-BL, Oku-BL, Ava-BL) carry virus genomes with unique deletion boundaries (see Figure 3A, black bars), in each case the deletion extends upstream of the EBNA2 coding sequence into a BamHI W fragment, thereby removing the unique Y1,Y2 exons of EBNA-LP, and downstream into the BamHI H fragment, removing most if not all of the BHLF1 lytic cycle gene. As illustrated, this brings that copy of the Wp promoter which is nearest the 5′ deletion boundary proximal to the previously described lytic cycle gene encoding the viral (v)bcl2 homologue BHRF1 [15].
We therefore designed a QRT-PCR assay for a transcript that splices from the W2 exon (present in all Wp-driven RNAs) into the BHRF1-coding exon. This was then used to look for evidence of such a W2-BHRF1-spliced species in the previously described panel of Awia-BL clones and in other BL lines representative of Wp-restricted and Latency I infections. Figure 3B shows the results of these W2-BHRF1 transcript assays alongside parallel QRT-PCR assays specific for (i) the Qp-driven EBNA1 transcript known to be expressed in Latency I lines, (ii) all Cp-driven transcripts, and (iii) all Wp-driven transcripts. Within the BL cell panel, the Wp-restricted BL lines and clones were, as expected [20], distinguished by high Wp usage in the absence of either Qp or Cp activity; importantly, these same Wp-using cells also expressed correspondingly high levels of the W2-BHRF1-spliced transcript. For all four Wp-restricted BL tumors, the products of RT-PCR amplification with the W2 and BHRF1 primers were then sequenced to determine their splice structure. As fully described in Figure S3, the structures were slightly different in each tumor depending upon the position of the 5′ deletion boundary relative to the W1 and W2 exons and the position of the 3′ boundary relative to the H2 exon that lies immediately upstream of BHRF1. However, all transcripts spliced from the BamHI W fragment into the BHRF1-coding exon.
We therefore screened the same cell line panel for the presence of BHRF1 protein by immunoblotting with the specific mAb 5B11. As illustrated in Figure 3C, the Wp-restricted BL cells did indeed express the protein, although at levels below that seen in a reference track made from a culture enriched in lytically-infected cells; note that, in lytic cycle, BHRF1 is abundantly expressed from its own lytic cycle promoter situated just upstream in the H2 exon [12]. To counter the possibility that BHRF1 expression in Wp-restricted BL lines simply reflected the presence of a few cells spontaneously entering lytic cycle, we screened these same lines by QRT-PCR assay for transcription of the immediate early lytic gene BZLF1 (Figure 3B) and by immunoblotting for BZLF1 protein (Figure 3C), both sensitive indicators of lytic cycle activity. There was no evidence of such activity, strongly suggesting that BHRF1 is indeed being expressed as a latent protein in Wp-restricted BL cells.
We then asked whether expressing appropriate levels of BHRF1 protein in Latency I BL cells would be sufficient to confer the marked resistance to apoptotic triggers characteristic of Wp-restricted BL cells. In this regard, recent work has identified three EBV miRNAs whose expression is associated with BHRF1 transcription and whose coding sequences lie close to, but outside, the BHRF1 protein-coding sequence [8],[9],[47]. Therefore, to avoid any possible contribution from these or other as-yet-undiscovered miRNAs from this region, the following experiments used an expression construct containing only the BHRF1 cDNA sequence and, as a control, the same construct with a mutation in the initial methionine codon (mut-BHRF1). These constructs were cloned into the dox-regulatable vector and introduced into two Latency I BL lines, Akata-BL and Sav-BL. Figure 4A confirms induction of BHRF1 protein expression in Akata-BL cells carrying the wild-type BHRF1 coding sequence; note that with an inducing dose of 1 ng/ml dox, BHRF1 expression was similar to that seen in the Wp-restricted Awia-BL clones whereas, at 1 µg/ml dox, it approached the much higher levels seen in EBV lytic cycle. These BHRF1 transfectants, plus control transfectants carrying either the mut-BHRF1 sequence or an empty vector and also the previously described EBNA3A, 3B, 3C Akata-BL transfectants, were then exposed to different dox concentrations before assaying for sensitivity to a 5 µg/ml ionomycin challenge. As shown in Figure 4B, the wild-type BHRF1 transfectants were completely protected even at the lowest level of BHRF1 expression whereas the other three types of transfectant remained as sensitive as the co-resident non-transfected population. Equally efficient protection from ionomycin- and anti-IgM-induced apoptosis was also mediated by BHRF1 in the Sav-BL cell line (data not shown). To ensure that this effect of BHRF1 was also apparent in an EBV-negative BL cell background, we generated the same panel of transfectants in EBV-loss Akata-BL cells and obtained an exactly similar pattern of results, as shown in Figure S4A. Throughout these experiments we also confirmed that the anti-IgM- and ionmycin-induced cell death was occurring predominantly by apoptosis, with typical PARP cleavage detectable in dying cells and protection from that cleavage in cells induced to express BHRF1 (Figure S4B).
Having observed this connection between Wp activity, BHRF1 expression and apoptosis resistance, we were interested to check its possible relationship to the recent finding that, in vitro, BHRF1 is transiently expressed in newly-infected B cells, thereby promoting their survival immediately post-infection [16]. We therefore asked whether the W2-BHRF1-spliced transcript seen from an EBNA2-deleted genome in Wp-restricted BL cells might also be expressed in normal B cells following infection with a transforming (i.e. non-EBNA2-deleted) virus.
In this regard, it is known that Wp is activated immediately following infection, rapidly rises to a peak and then falls as Cp takes its place as the dominant EBNA promoter; LMP1 transcription is EBNA2-dependent and is not seen until after Cp becomes dominant [48]. Since the Wp (and Cp) promoters specifically give rise to RNAs with a W1W2Y1Y2 splice structure [11],[49],[50], it was anticipated that any latent BHRF1 transcript encoded by such a virus would be detectable both by the previously designed W2-BHRF1 transcript assay and by a newly designed QRT-PCR assay using Y2 and BHRF1 primer pairs. The relevant splice structures and primer/probe locations are illustrated in Figure 5A. Normal B cells from healthy donors were therefore exposed to EBV, cultured and then harvested after intervals up to 120 hrs later. Note that, to avoid possible complication from lytic BHRF1 gene expression in these experiments, we used a recombinant EBV strain (BZKO) that had been rendered incapable of lytic cycle entry by deletion of the BZLF1 immediate early gene. Figure 5B shows the QRT-PCR results obtained when virus gene expression in infected B cells was analysed using the W2-BHRF1 and Y2-BHRF1 assays, as well as the standard assays detecting all Wp-initiated transcripts, all Cp-initiated transcripts and LMP1 transcripts. As shown in Figure 5B, W2-BHRF1-spliced and Y2-BHRF1-spliced transcripts were detected as early as 8 hrs post-infection, peaked within 12 hrs and then fell, exactly matching the kinetics of Wp-activity.
Figure 5C shows the results obtained when cells from the same experiment were assayed for protein expression by immunoblotting. EBNA2 and EBNA-LP, the immediate products of Wp transcription, were easily detectable by 24 hr while low levels of BHRF1 were just detectable by 24–48 hrs, some 2–3 days before LMP1. In view of the low levels of BHRF1 detected by immunoblotting, we sought to confirm protein expression by another method. This took advantage of the fact that cells endogenously expressing BHRF1 are efficiently recognised by CD4+ T cells specific for a derived peptide epitope presented by the HLA-DR4 allele [42]. We therefore raised CD4+ T cell clones specific for this BHRF1 epitope from a DR4-positive EBV-immune donor and tested these on autologous B cells after infection with the BZKO virus. As an internal control, we also tested the same target cells for recognition by CD4+ T cells against another DR4-restricted epitope, this time derived from an antigen known to be expressed early post-infection, EBNA2 [43]. The results of these assays are presented in Figure 5D as histograms of interferon-gamma (IFNγ) release; in each case, recognition of infected targets is shown relative to the maximum seen when the same target cells are pre-pulsed with the relevant synthetic epitope peptide. Both sets of antigen-specific T cells showed clear recognition of target B cells at both 4 and 8 days post-infection; indeed the BHRF1 effectors gave the stronger signals. Note that this recognition required de novo protein synthesis (rather than antigen acquired from the virus preparation) since DR4-positive B cells assayed immediately after overnight exposure to the virus were not recognised (data not shown). To further check that the BHRF1 effectors were specific, we carried out an equivalent experiment this time using a recombinant EBV (BHRF1KO) in which the BHRF1 gene has been inactivated by insertion of a kanamycin resistance cassette [16]. Cells infected with this virus were indeed not recognised by the BHRF1-specific effectors but were recognised by T cells specific for EBNA2 (Figure 5D). Because all of the above experiments had involved recombinant viral strains, we then repeated the work on cells freshly infected with wild-type virus and obtained a similar pattern of results whether assaying for BHRF1 expression by transcription, by immunoblotting or by T cell detection (Figure S5).
Given that Wp has been shown to remain constitutively active at a low level in all LCLs [51],[52], we went on to ask whether latent BHRF1 expression might persist in the longer term. LCLs were therefore established from a range of donors using both wild-type and BZKO virus strains, then assayed after 2–4 months in culture for latent BHRF1 transcripts, as well as for representative early (BMLF1) and late (gp350) lytic cycle RNAs. Latent BHRF1-spliced transcripts were consistently detected in all LCLs, whether transformed with wild-type or BZKO virus; data from the Y2-BHRF1 QRT-PCR assay are shown in Figure 6A; results from the W2-BHRF1 assay were very similar (data not shown). Sequencing of the W2-BHRF1 RT-PCR products confirmed that they did indeed represent RNAs with the predicted W2-Y1-Y2-BHRF1 splice structure (see Figure S3). By contrast BMLF1 and gp350 transcription was only detected in LCLs carrying wild-type virus, reflecting the presence in these lines of a small percentage of cells spontaneously entering lytic cycle. Interestingly, in all the LCLs, residual Wp activity correlated well with latent BHRF1 transcript levels. However these levels were lower than those seen in freshly-infected cells and, accordingly, BHRF1 protein expression in BZKO LCLs was often at or below the borderline of detectability by immunoblotting (data not shown). However, we reproducibly could detect endogenous expression of the BHRF1 protein in both BZKO and wild-type LCLs by CD4+ T cell assay. Figure 6B shows the results of such assays conducted on pairs of LCLs generated from three different donors; note that these donors were chosen because they were positive both for HLA-DR4, the restricting allele for the BHRF1-specific T cells, and for HLA-DR15, the restricting allele for a CD4+ T cell clone specific for gp350, a late lytic cycle protein. As expected, only wild-type LCLs with some cells in lytic cycle could be recognised by the gp350-specific effectors. However, BHRF1-specific CD4+ T cells consistently recognised both the wild-type and the BZKO LCLs at remarkably high levels, up to 33% of that seen on cells pre-pulsed with cognate peptide.
This work was prompted by the recent finding that, in 15% of endemic BLs, tumor pathogenesis appears to have selected for infection of the target cells with an EBNA2 gene-deleted virus, a transformation-defective mutant that we presume represents only a tiny fraction of the total virus load in the B cell system in vivo. Such infection is associated with activation of the Wp latent promoter and a broadening of latent antigen expression beyond that typically seen in Latency I BLs [20]. The fact that this is also accompanied by a marked increase in the tumor cell's resistance to apoptosis [32] immediately helps to explain why such a rare virus mutant had been captured within a disproportionately large number of tumors. This, in turn, greatly strengthens the argument [25] that EBV's role in BL pathogenesis is anti-apoptotic rather than growth-promoting. Here we show that this marked resistance is mediated by Wp-driven expression of the viral bcl2 homologue, BHRF1, a protein hitherto mainly associated with the virus lytic cycle. Furthermore we find that the Wp/BHRF1 connection, though discovered in the context of a mutant virus in BL cells, is also an integral feature of normal B cell infection with wild-type virus. This link with Wp not only explains the burst of BHRF1 expression observed in B cells immediately post-infection [16] but has also led to the finding that BHRF1 remains constitutively expressed as a latent protein in all EBV-transformed LCLs.
We initially focused on the EBNA3 proteins as the most likely mediators of protection from anti-IgM- and ionomycin-induced apoptosis since, at the time, these were the only viral antigens consistently found in Wp-restricted and not in Latency I BL cells. Furthermore in a recent report where EBV-negative BL cells had been stably infected with recombinant viruses (rEBVs) [46], the protection offered by wild-type rEBV infection against cell death induced by nocodazole (which disrupts mitotic spindles), cisplatin (a DNA cross-linking agent) and roscovitine (a cyclin-dependent kinase inhibitor) was lost with virus strains from which either the EBNA3A or EBNA3C gene had been deleted, suggesting that EBNA3A and 3C can act cooperatively to influence the BL phenotype. As a preliminary to the present study, we performed similar infection experiments with EBNA3-knockout viruses on several EBV-negative BL backgrounds but, in our hands, the apoptosis assay results on infected lines proved difficult to interpret because such protocols do not faithfully reproduce Wp-restricted latency. Thus, after the drug selection that is required to establish stable infectants, the selected cells expressed variable levels of viral and cellular anti-apoptotic proteins, namely LMP1 and bcl2, that are never seen in Wp-restricted BL lines (G.L. Kelly, unpublished results). This emphasised the importance of establishing a more controlled experimental system in which to study the effects of specific latent antigens on the apoptosis phenotype. We therefore turned to a novel ori-p-based vector system [36] that allows candidate genes, alone or in combination, to be introduced into Latency I BL lines in silent form and then expression induced to physiologic or supra-physiologic levels by graded doses of dox. This avoids subjecting cells to any drug selection prior to apoptosis assays, and has the added advantage that both vector-positive and control cells exist within the same culture, the former being identified by dox-induction of GFP from the same vector. Using this system, we induced expression of EBNA3A, EBNA3B and/or EBNA3C in two different Latency I cell backgrounds and in an EBV-negative BL backgound, then assayed these cells for sensitivity to triggers of apoptosis (anti-IgM and/or ionomycin) known to distinguish between Latency I and Wp-restricted BL cells. We saw no protection by the EBNA3 proteins, whether expressed individually or together, and whether induced at physiologic levels or much higher.
We therefore began to search for other possible mediators of the anti-apoptotic effect, starting from the observation that all cases of BL presenting with a Wp-restricted pattern of gene expression carried an EBNA2 gene-deleted virus [20],[33]. Close inspection of the different deletions found in individual Wp-restricted tumors showed that each placed a copy of the Wp promoter immediately upstream of the gene encoding BHRF1, a viral bcl2 homologue hitherto thought to be expressed predominantly in lytic cycle [15]. Indeed a W2-BHRF1-spliced RNA could be amplified from all Wp-restricted BL cell lines tested, but never from Latency I lines in which Wp was silent. Moreover a BHRF1 protein could be detected by immunoblotting in all Wp-restricted lines, though at levels much weaker than seen in lytic cycle. Interestingly BHRF1 protein expression and levels of Wp-initiated and W2-BHRF1-spliced transcripts were lower in Awia-BL than in other Wp-restricted lines, perhaps reflecting the single EBV genome copy number in Wp-restricted Awia-BL clones [33]. These results implied that, if ectopic expression of BHRF1 were to explain the apoptosis resistance of these cells, the protein must be active at much lower concentrations than hitherto appreciated. Thus earlier studies had shown that BHRF1 can protect B cells from apoptosis induced by a number of different triggers including growth factor withdrawal [53], TRAIL death receptor signalling [54], γ irradiation and chemotherapeutic drugs [55]. However, such experiments frequently involved vectors giving high level expression. We therefore used the dox-inducible vector system to express BHRF1 in Latency I BL lines at levels ranging from that seen in the Awia-BL clones up to those typical of lytic infection. Remarkably, BHRF1 was fully protective against anti-IgM- and ionomycin-induced apoptosis even at the very lowest level, strongly supporting the view that BHRF1 expressed as a latent protein from the Wp promoter is responsible for the apoptosis resistance of Wp-restricted BLs.
We then went on to ask whether the Wp/BHRF1 connection was unique to the EBNA2-deleted viruses selected for in BL pathogenesis or a hitherto-unrecognised facet of Wp usage in wild-type virus infections. This latter possibility was raised by a recent study showing that BHRF1 was expressed in the first few days following B cell infection with wild-type virus and that this was important for optimal transformation efficiency [16]. There the transient expression of BHRF1 and of a second viral bcl2 family member BALF1, which together appear to protect recently-infected B cells from apoptosis [16], was ascribed to opportunistic transcription from the virus genome following its delivery to the cell nucleus as linear unmethylated DNA. However we found that BHRF1 expression in recently-infected cells was temporally linked to Wp activity and to the detectability of BHRF1 transcripts passing through W2 and Y2 upstream exons. Thus, transcript levels measured by the total Wp, W2-BHRF1 and Y2-BHRF1 QRT-PCR assays all peaked around 12 hrs post-infection and then started to decline as promoter usage switched from Wp to the upstream Cp promoter. These findings suggest that BHRF1 expression is a specific consequence of Wp activation in newly-infected cells and not simply opportunistic transcription from an unmethylated virus genome.
Finally, given recent work showing that Wp is never completely eclipsed by Cp in growth-transforming infections [51],[52], we examined established LCLs for evidence of BHRF1 expression. Both W2-containing and Y2-containing BHRF1 transcripts were consistently detected, whether cells had been transformed with wild-type or lytic cycle-deficient (BZKO) virus. Interestingly some of the earliest analyses of EBV transcription, including work on the tightly latent IB4 LCL, had isolated rare cDNA clones that included W1,W2 and BHRF1 sequences [11]–[13]. However, in the apparent absence of detectable BHRF1 as a latent cycle protein, the significance of the above findings remained obscure. With the advent of more sensitive enhanced chemiluminescent methods for immunoblot detection, we now find that BHRF1 protein is consistently detectable in recently-infected B cells (at times immediately following the peak of Wp activity) and is often just detectable at trace levels in immunoblots of established, tightly-latent, LCLs. The possibility of confirming these observations using T cells, rather than an antibody, as the probe came with the description of CD4+ T cells specific for a defined BHRF1 peptide epitope that recognise target cells endogenously expressing cognate antigen [42]. Here we used the greater sensitivity of these BHRF1-specific CD4+ T cells to show that the protein is indeed constitutively expressed in all established LCLs, even in lines devoid of lytically-infected cells. This puts BHRF1 in a special category of EBV antigens that straddle the lytic/latent divide. Thus it is abundantly expressed from its own promoter in early lytic cycle and also constitutively expressed from a latent cycle promoter in growth-transforming infections.
Setting the present work in its wider context, it is now known that herpesviruses from several different genera have acquired bcl2-homologous genes during their evolution, and express these genes during lytic virus replication, thereby extending survival of the infected cell and maximising virus production [56],[57]. In the case of EBV, and presumably the other gamma-1-herpesviruses of Old World primates [58], the vbcl2 homologue has also been recruited as part of the B cell growth-transforming programme that is unique to these viruses and appears to be important in virus colonisation of the naïve host. This has been achieved by placing BHRF1 under a promoter, Wp, that both initiates transformation and remains constitutively active at some level in transformed cells. While the Wp/BHRF1 connection increases the overall efficiency of B cell transformation [16], it also brings the risk that in other situations inappropriate activation of Wp will lead to unscheduled BHRF1 expression and enhanced survival of the affected cell. We suggest that this is indeed the case in the subset of endemic BLs studied here, where the presence of an EBNA2-deleted virus genome results in high Wp activity and constitutive BHRF1 expression. This draws a direct parallel between the pathogenesis of EBV-positive BL and that seen in mouse models of c-myc-driven lymphomagenesis [23],[24], where the drive towards full malignancy occurs only when a target cell expressing a deregulated c-myc oncogene acquires complementary changes that counteract c-myc-driven apoptosis. The present work suggests that, in a subset of EBV-positive BLs, the complementary factor can be BHRF1. In so doing, it provides the first evidence implicating a herpesvirus bcl2 protein in viral oncogenesis.
|
10.1371/journal.pgen.1007949 | Parental legacy, demography, and admixture influenced the evolution of the two subgenomes of the tetraploid Capsella bursa-pastoris (Brassicaceae) | Allopolyploidy is generally perceived as a major source of evolutionary novelties and as an instantaneous way to create isolation barriers. However, we do not have a clear understanding of how two subgenomes evolve and interact once they have fused in an allopolyploid species nor how isolated they are from their relatives. Here, we address these questions by analyzing genomic and transcriptomic data of allotetraploid Capsella bursa-pastoris in three differentiated populations, Asia, Europe, and the Middle East. We phased the two subgenomes, one descended from the outcrossing and highly diverse Capsella grandiflora (CbpCg) and the other one from the selfing and genetically depauperate Capsella orientalis (CbpCo). For each subgenome, we assessed its relationship with the diploid relatives, temporal changes of effective population size (Ne), signatures of positive and negative selection, and gene expression patterns. In all three regions, Ne of the two subgenomes decreased gradually over time and the CbpCo subgenome accumulated more deleterious changes than CbpCg. There were signs of widespread admixture between C. bursa-pastoris and its diploid relatives. The two subgenomes were impacted differentially depending on geographic region suggesting either strong interploidy gene flow or multiple origins of C. bursa-pastoris. Selective sweeps were more common on the CbpCg subgenome in Europe and the Middle East, and on the CbpCo subgenome in Asia. In contrast, differences in expression were limited with the CbpCg subgenome slightly more expressed than CbpCo in Europe and the Middle-East. In summary, after more than 100,000 generations of co-existence, the two subgenomes of C. bursa-pastoris still retained a strong signature of parental legacy but their evolutionary trajectory strongly varied across geographic regions.
| Allopolyploid species have two or more sets of chromosomes that originate from hybridization of different species. It remains largely unknown how the two genomes evolve in the same organism and how strongly their evolutionary trajectory depends on the initial differences between the two parental species and the specific demographic history of the newly formed allopolyploid species. To address these questions, we analyzed the genomic and gene expression variation of the shepherd’s purse, a recent allopolyploid species, in three regions of its natural range. After ∼100,000 generations of co-existence within the same species, the two subgenomes had still retained part of the initial difference between the two parental species in the number of deleterious mutations reflecting a history of mating system differences. This difference, as well as differences in patterns of positive selection and levels of gene expression, also strongly depended on the specific histories of the three regions considered. Most strikingly, and unexpectedly, the allopolyploid species showed signs of hybridization with different diploid relatives or multiple origins in different parts of its range. Regardless if it was hybridization or multiple origins, this profoundly altered the relationship between the two subgenomes in different regions. Hence, our study illustrates how both the genomic structure and ecological arena interact to determine the evolutionary trajectories of allopolyploid species.
| Allopolyploidy, the origin of polyploids from two different ancestral lineages, poses serious evolutionary challenges since the presence of two divergent sub-genomes may lead to perturbation of meiosis, conflicts in gene expression regulation, protein-protein interactions, and transposable element suppression [1–3]. Whole genome duplication also masks new recessive mutations thereby decreasing selection efficacy [4, 5]. This relaxation of selection, together with the strong speciation bottleneck and shift to self-fertilization that often accompany polyploidy [6], ultimately increases the frequency of deleterious mutations retained in the genome [7, 8]. All of these consequences of allopolyploidy can have a negative impact on fitness and over evolutionary time may contribute to the patterns of duplicate gene loss, a process referred to as diploidization [5, 9, 10]. Yet, allopolyploid lineages often not only establish and persist but may even thrive and become more successful than their diploid progenitors and competitors, with larger ranges and higher competitive ability [11–20]. The success of allopolyploids is usually explained by their greater evolutionary potential. Having inherited two genomes that evolved separately, and sometimes under drastically different conditions, allopolyploids should have an increased genetic toolbox, assuming that the two genomes do not experience severe conflicts [21–23]. This greater evolutionary potential of allopolyploids can be further enhanced by genomic rearrangements, alteration of gene expression and epigenetic changes [4, 5, 24–30].
All of these specific features come into play during the demographic history of allopolyploids. Demographic processes occurring when a species extends its range, such as successive bottlenecks or periods of rapid population growth in the absence of competition, are expected to have a profound impact on evolutionary processes, especially in populations at the front of the expansion range. Species that went through repeated bottlenecks during their range expansion are expected to have reduced genetic variation and higher genetic load than more ancient central populations [31, 32]. Similarly, range expansions can also lead to contact and admixture with related species. Such admixture can in turn shift the evolutionary path of the focal species. Finally, range expansion will expose newly formed allopolyploid populations to divergent selective pressures, providing the possibility of differentially exploiting duplicated genes, and creating asymmetrical patterns of adaptive evolution in different parts of the range.
In this paper, we aim to characterize the evolution of the genome of a recent allopolyploid species during its range expansion. In particular, we explore whether the two subgenomes have similar or different evolutionary trajectories in hybridization, selection and gene expression. The widespread allopolyploid C. bursa-pastoris is a promising system for studying the evolution of polyploidy, with available information on its two progenitor diploid species and their current distribution. C. bursa-pastoris, a selfing species, originated from the hybridization of the Capsella orientalis and Capsella grandiflora / rubella lineages some 100-300 kya [10]. C. orientalis is a genetically depauperate selfer occurring across the steppes of Central Asia and Eastern Europe. In contrast, C. grandiflora is an extremely genetically diverse obligate outcrosser which is primarily confined to a small distribution range in the mountains of Northern Greece and Albania. The fourth relative, C. rubella, a selfer recently derived from C. grandiflora, occurs around the Mediterranean Sea. There is evidence for unidirectional gene flow from C. rubella to C. bursa-pastoris [33]. Among all Capsella species, only C. bursa-pastoris has a worldwide distribution (Fig 1A), which may be partially due to extremely recent colonization and associated with human population movements [34]. A recent study reveals that in Eurasia, C. bursa-pastoris is divided into three genetic clusters—Middle East, Europe, and Asia—with low gene flow among them and strong differentiation both at the nucleotide and gene expression levels [34, 35]. Reconstruction of the colonization history using unphased genomic data suggested that C. bursa-pastoris spread from the Middle East towards Europe and then into Eastern Asia. This colonization history resulted in a typical reduction of nucleotide diversity with the lowest diversity being in the most distant Asian population [34].
Whether adaptation on the two distinct non-recombining [36, 37] subgenomes of C. bursa-pastoris contributed to its rapid population expansion and how they were in return affected by it, remains unclear. Previous studies either ignored the population history of C. bursa-pastoris or failed to consider the two subgenomes separately. In a recent study that does not consider the population demographic history within C. bursa-pastoris, Douglas et al. [10] concluded that there is no strong sign of diploidization in C. bursa-pastoris and most of its variation is the result of the legacy from the parental lineages with some relaxation of purifying selection caused by both the transition to self-fertilization and the greater masking of deleterious mutations. Kryvokhyzha et al. [35] considered population history but did not separate the two subgenomes, and showed that variation in gene expression among Asian, European and Middle Eastern accessions strongly reflects the population history with most of the differences among populations explained by genetic drift. We extend these previous studies by analyzing the genome-wide expression and polymorphism patterns of the two subgenomes of C. bursa-pastoris in 31 accessions sampled across its natural range in Eurasia. We demonstrate that the two subgenomes follow distinct evolutionary trajectories in different populations and that these trajectories are influenced by both range expansion and hybridization with diploid relatives. Our study illustrates the need to account for demographic and ecological differences among populations when studying the evolution of subgenomes of allopolyploid species.
The disomic inheritance of C. bursa-pastoris [36, 37] allowed us to successfully phase most of the heterozygous sites in the 31 samples analyzed in this study (Fig 1A, S1 Table). Out of 7.1 million high confidence SNPs, our phasing procedure produced an alignment of 5.4 million phased polymorphic sites across the 31 accessions of C. bursa-pastoris. Scaling these phased SNPs to the whole genome resulted in the alignment of 80.6 Mb that had the same level of heterozygosity as the unphased data. The alignment of these whole genome sequences of C. bursa-pastoris with 13 sequences of C. grandiflora, 10 sequences of C. orientalis, one sequence of C. rubella (the reference), and one sequence of N. paniculata used here as an outgroup, yielded 12.8 million polymorphic sites that we used in all analyses. The information for each accession is provided in S1 Table.
We analyzed the structure of the phased data with phylogenetic analyses. The separation of the two subgenomes was strongly supported in the reconstructed whole genome tree (Fig 1B). The grouping with a corresponding parental species was also maintained in the phylogenetic analyses of each subgenome separately (S1 Fig) as expected given assumptions of the phasing approach. The tree consisted of two highly supported (100% bootstrap) major clades grouping C. grandiflora and the C. grandiflora / rubella lineage descended subgenome of C. bursa-pastoris (hereafter the CbpCg subgenome), on the one hand, and C. orientalis and the C. orientalis lineage descended subgenome of C. bursa-pastoris (hereafter the CbpCo subgenome), on the other hand. We also analyzed phylogenetic signals at a finer genomic scale using a sliding window approach with 100-kb window size (Fig 1C). Exclusive monophyly of C. orientalis with the CbpCo subgenome, and C. grandiflora and C. rubella with CbpCg subgenome was detected in 95% and 83% of trees, respectively (S2 Fig).
Comparing linkage disequilibrium (LD) decay within and across subgenomes—e.g. comparing r2 between SNP1 and SNP2 in CbpCg and r2 between SNP1 in CbpCg and SNP2 in CbpCo—suggested largely consistent phasing within each subgenome after accounting for population history (S3 Fig). Within homeologues (particularly CbpCg) we observed high LD in the Middle Eastern and European samples with gradual decay as genetic distance increases. LD decayed more rapidly in the Asian samples than in the other populations (possibly due to less population structure and expansion) but also showed the expected decay within both subgenomes. In contrast, r2 with SNPs from the other subgenome or randomly selected from either subgenome was substantially lower and did not show any decay within 10 KB, particularly in the Middle Eastern and European samples. High LD and gradual decay within, and not between, subgenomes suggested phasing was largely effective in separating the two subgenomes and preserving signals of their demographic history.
The genomic data analyzed in this study were phased using the computational phasing of reads mapped to the C. rubella reference (HapCUT method) and was validated with mapping the same reads to the recent assembly of C. bursa-pastoris [37] (HomeoRoq method, see Material and methods). The final alignment between the alternatively phased dataset comprised 800 K SNPs. This alignment was relatively small because mapping to two subgenomes resulted in 2x smaller coverage and subsequent genotype calling, filtering, phasing, and alignment cumulatively further reduced this size (S4 Fig). Phylogenetic trees reconstructed for 10 K sliding windows showed 100% split between CbpCg and CbpCo subgenomes (S5 Fig). Using a smaller window of 1 K resulted in 83% of trees with mutual monophyly of CbpCg and CbpCo subgenomes (S6 Fig). The other 17% still showed a split between CbpCg and CbpCo for the majority of samples, but some single samples were not consistent with the mutual monophyly of the two subgenomes. This could be either due to incomplete lineage sorting or potential local phasing errors. Nevertheless, the major clustering into two subgenomes and three populations was largely maintained in the 17% of trees (S7 Fig). In all trees, the same samples phased with two alternative methods (HapCUT and HomeoRoq) clustered together for most of the European samples. The Middle Eastern cluster showed slightly less consistent clustering, and the Asian samples showed clustering into two groups according to the phasing method. The cause of these discrepancies is difficult to define because errors are possible in both datasets. To localize the problematic regions, we also compared the phased dataset analyzed in this study (HapCUT phased) with the assembly sequences of C. bursa-pastoris [37]. This comparison covered almost all positions of the dataset analyzed in this study (10.9Mb out of 12.7Mb). Most of the discrepancies were located in the vicinity of pericentromeric regions (S8 Fig)), where assembling reads is problematic. In summary, there was a high concordance of the results obtained with the different phasing methods and those that differed are unlikely to have affected the main results of the study.
For both subgenomes, the three C. bursa-pastoris populations, Asia (ASI), Europe (EUR) and Middle East (ME), constituted well-defined phylogenetic clusters (Fig 1B and 1C). However, the relationships of each subgenome with its parental species differed. The CbpCg subgenome formed a monophyletic clade with C. grandiflora at its base. In contrast, the CbpCo subgenome was paraphyletic with C. orientalis that clustered within the ASI group instead of being outside of all C. bursa-pastoris CbpCo subgenomes (this was observed in both phasing methods S9 Fig). This clustering was unexpected and suggested potential gene flow between the ASI group and C. orientalis or multiple origins of the CbpCo subgenome. Nucleotide diversity was higher on the CbpCg subgenome than on the CbpCo subgenome for both EUR and ME (Fig 2, S10 Fig, S2 Table), though the difference was significant only for EUR (p-values: 0.005 and 0.154 for EUR and ME, respectively). The opposite pattern was observed for ASI (Fig 2): there the nucleotide diversity in the CbpCo subgenome was significantly higher than in the CbpCg subgenome (p-value < 0.0001). Interestingly, the diversity of the CbpCo subgenome in all populations was significantly higher than the diversity of its parental species, C. orientalis (p-value < 0.0001).
To reconstruct the changes in effective population size (Ne) over time in the three C. bursa-pastoris populations and the two ancestral species, we used a pairwise sequentially Markovian coalescent model (PSMC). First, we reconstructed the demographic histories of C. orientalis and C. grandiflora (Fig 3). In C. grandiflora, Ne was mostly constant with some slight decrease in the recent past, but the Ne of C. orientalis decreased continuously towards the present. In C. bursa-pastoris, despite a simultaneous rapid range expansion, Ne of EUR and ME populations also gradually decreased starting from around 100-200 kya. The ASI population showed a similar pattern but with population size recovery around 5-10 kya and a subsequent decrease to the same Ne as in EUR and ME. The Ne patterns of the two subgenomes were similar within each population. Overall, the Ne history of C. bursa-pastoris was most similar to that of its selfing ancestor, C. orientalis. We also verified these PSMC results with SMC++, which can consider more than two haploid genomes and incorporates linkage disequilibrium (LD) in coalescent hidden Markov models [39]. The general trend was globally the same but the recent decline of C. orientalis was sharper and fluctuations in Ne were more pronounced (S11 Fig). In summary, the overall pattern of Ne change over time was mostly the same between the two subgenomes and between the three populations of C. bursa-pastoris and it was largely similar to the pattern observed for the diploid selfer C. orientalis.
To quantify the relationships between populations of C. bursa-pastoris and the two parental species, we applied a topology weighting method that calculates the contribution of each individual group topology to a full tree [40]. We looked at the topologies joining each subgenome of C. bursa-pastoris and a corresponding parental lineage. There are 15 possible topologies for three populations of C. bursa-pastoris, a parental species, and the root. We grouped these topologies into five main groups: species trees—topologies that place a parental lineage as a basal branch to C. bursa-pastoris; three groups that join one of the populations of C. bursa-pastoris with a parental lineage and potentially signifies admixture; and all other trees that place a parental lineage within C. bursa-pastoris but do not relate it with a particular population of C. bursa-pastoris (Fig 4A).
These topology weightings varied along the subgenomes and illustrated distinct patterns between the two subgenomes (Fig 4B). In the CbpCo subgenome, the largest average weighting was for the topology grouping the ASI population of C. bursa-pastoris with C. orientalis (Fig 4C), and the species topology had the second largest average weighting. The difference between the average weighting in these two topology groups was statistically significant (S3 Table). In contrast, the species topologies weighting dominated in the CbpCg subgenome, regardless if C. rubella or C. grandiflora were used as a parental lineage (Fig 4C, S12 Fig, S4 and S5 Tables). The topology uniting the CbpCg subgenome of the EUR population with C. rubella had the largest topology weighting among the topologies indicating admixture between these clusters (Fig 4C). Thus, the two subgenomes differed substantially in the pattern of topology weighting and there were signs of a potential admixture of EUR and ASI with C. rubella and C. orientalis, respectively.
The phylogenetic grouping of C. orientalis with the Asian CbpCo subgenome, together with topology weighting results and the relatively elevated nucleotide diversity in this subgenome, suggested the possibility of gene flow between C. orientalis and C. bursa-pastoris in the ASI population. To test this hypothesis, and at the same time to check for possibilities of gene exchange between C. bursa-pastoris and other Capsella species, we conducted two complementary tests of admixture.
We first used the ABBA-BABA test, a coalescent-based method that relies on the assumption that alleles under incomplete lineage sorting are expected to be equally frequent in two descendant populations in the absence of admixture between any of them and a third population that diverged earlier on from the same common ancestor [41, 42]. The deviation from equal frequency is measured with the D-statistic, which ranges between 0 and 1, with 0 indicating no gene flow and 1 meaning complete admixture. The ABBA-BABA test also provides an estimate of the fraction of the genome that is admixed by comparing the observed difference in ABBA-BABA with the difference expected under a scenario of complete admixture (f-statistic). We estimated D and f for triplets including one diploid species and two populations of C. bursa-pastoris represented by the most related subgenome to that species (Table 1). N. paniculata was the outgroup in all tests. The D-statistic were significantly different from 0 in most of the tests, so we considered all three combinations per test group (see Table 1) to determine the pairs that were the most likely to be admixed. The largest fraction of admixture was identified for the pair of the ASI CbpCo subgenome and C. orientalis with an estimate of f indicating that at least 14% of the ASI CbpCo subgenome is admixed. The second largest proportion of admixture was detected between C. rubella and the EUR CbpCg subgenome with f estimate of at least 8%. The estimates for tests with C. grandiflora were the smallest but similar to those obtained for C. rubella. The latter may reflect the strong genetic similarity between these two species rather than real gene flow between C. grandiflora and C. bursa-pastoris which, based on crosses, seems unlikely. Indeed, we crossed individuals from the three populations of C. bursa-pastoris with their three diploid relatives to test for the presence of reproductive barriers. Regardless of the direction of the crosses, all crosses between C. rubella and the three populations of C. bursa-pastoris produced viable F1 seeds. In contrast, crosses involving C. grandiflora mostly failed and the abortion rate approached 100%. Details on these crosses are provided in S1 Appendix. The admixture with C. rubella in EUR and C. orientalis in ASI was also supported by the unphased and complete (i.e. no missing genotypes) genomic data (S6 and S7 Tables). Finally, it should be pointed out that given that evidence for C. bursa-pastoris monophyly is weak, it is also possible that the signals of admixture with the parental species that we are detecting here actually reflect introgression from an independently-arisen C. bursa-pastoris into either CbpCo or CbpCg subgenomes.
We then used HAPMIX, a haplotype-based method, which enables capture of both large and fine-scale admixture, as well as an absolute estimate of the proportion of the genome that was admixed. For the analysis of the CbpCg subgenome of C. bursa-pastoris, the highest levels of admixture were found consistently across regions to be with the diploid C. rubella. In Europe, genomic regions showed an average 18% of admixture with C. rubella, followed by 11% in the Middle East, and just 2% in Asia (S8 Table, S13A Fig). All three populations also showed signs of admixture with C. grandiflora but to a reduced extent compared to C. rubella (7% in Europe, 6% in the Middle East, 0.2% in Asia). C. rubella is highly similar to C. grandiflora, and as noted above, we expect difficulties in discerning between the two, suggesting that much of the signal of admixture with C. grandiflora could in fact be due to admixture with C. rubella. Of the regions putatively admixed with C. grandiflora, 78%-96% of sites called as admixed overlapped with those from C. rubella, none of which occurred in unique regions for C. grandiflora. Because of this, and in combination with the reduced genome-wide probability of admixture with the diploid C. grandiflora compared to C. rubella (e.g. 0.11 compared to 0.24 in Europe), we argue that the signals of admixture with the diploid C. grandiflora were likely an artifact of its similarity with the regions of C. rubella admixture. These findings in accord with the ABBA-BABA results imply that the CbpCg subgenome has experienced significant admixture with C. rubella in Europe, and to a lesser extent in the Middle East.
For the analysis of the CbpCo subgenome, signals of admixture with the diploid C. orientalis were present in all three populations. In the ME population, genomic regions showed an average 18-21% admixture depending on the reference populations used (S8 Table, S13B Fig). Using the Middle East population for the analysis of the CbpCo subgenomes of EUR and ASI, since it was the least admixed in the HAPMIX results, yielded 15% C. orientalis admixture in Asia, and 14% in Europe. These findings suggest admixture of the diploid C. orientalis with the CbpCo subgenome across all three geographic regions. Assuming these levels of admixture accurately reflect reality, we do not have a non-admixed reference population to use for HAPMIX, and are thus violating a key assumption of the method. HAPMIX inferences for the CbpCo subgenome should therefore be taken with caution but we note that the results for ASI and ME are generally congruent with the admixture pattern obtained with ABBA-BABA.
In the present study, we analyzed the genetic changes experienced by the recently formed allopolyploid C. bursa-pastoris since its founding, focusing on the evolutionary trajectories followed by its two subgenomes in demographically and genetically distinct populations from Europe, the Middle East, and Asia. The shift to selfing and polyploidy had a global impact on the species, resulting in a sharp reduction of the effective population size in all populations accompanied by relaxed selection and accumulation of deleterious mutations. However, the two subgenomes were not similarly affected, with the magnitude of the subgenome-specific differences depending on the population considered. The relative patterns of nucleotide diversity, genetic load, selection and gene expression between the two subgenomes in the European and the Middle Eastern populations were distinct from that observed in Asia. The differences between populations were further enhanced by post-speciation hybridization of C. bursa-pastoris with local parental lineages. Below, we discuss these global and local effects in more detail and their consequences for the history of the species.
But before this, a few words need to be said on the reliability of the phasing approach that was chosen here. Our results are based on the computationally phased (HapCUT) genomic data that was generated from the reads mapped to the C. rubella reference genome—the only reference genome available for the Capsella genus at the time of this study. This approach was possible due to the strict disomic inheritance of C. bursa-pastoris [36, 37]. Disomic inheritance and reproduction by selfing resulted in almost no variation within subgenomes and we therefore were able to treat our data as diploid. The major challenge of our approach was to reduce mapping bias in favor of the CbpCg subgenome that is more closely related to C. rubella than the CbpCo subgenome [10]. To minimize this bias, we performed tolerant read mapping, stringent SNP filtering and even more stringent filtering of gene expression data. Our survey of the frequency of non-reference alleles, consistent decay of linkage disequilibrium with a distance, distribution of coverage and heterozygosity across the genome (see Material and methods), and almost an equal ratio between homeologous alleles in the DNA data, indicated that there were no major phasing errors. We also verified the sweeps and admixture results by analyzing only the phased fragments showing 100% consistency with fixed differences between parental species. Furthermore, we phased the same data using an alternative approach (HomeoRoq) and mapping our reads to the assembly of C. bursa-pastoris [37]. The two alternatively phased datasets were largely consistent in the separation between the subgenomes, with a perfect clustering between alternatively phased samples (HapCUT and HomeoRoq) in EUR and a clustering more influenced by phasing in ASI samples. However, even in that case, ASI samples still were more related to each other than to any other cluster and larger-scale grouping remained the same. Given that the assembly of C. bursa-pastoris we used for this validation was done with short reads and that the classification of the scaffolds between homeologues was based only on the coding part and allowed up to 25% of discrepancy between parental species [37], the congruence of the results is encouraging. Overall, we hence believe that our phasing approach did not lead to biases that could invalidate the main conclusions of our study.
The genetic composition of parental species obviously has a strong impact on the evolution of subgenomes in allopolyploids. For example, the allopolyploid Arabidopsis kamchatica originated from the outcrossing diploids Arabidopsis halleri and Arabidopsis lyrata [46] that have rather similar effective population sizes and the two subgenomes of A. kamchatica have thus similar effective population sizes and levels of negative selection [47]. In contrast, the effective population size of the diploid outcrossing ancestor of C. bursa-pastoris, C. grandiflora, is ten times larger than that of its selfing ancestor C. orientalis [10]. Any analysis of the difference in effective population size between the subgenomes of C. bursa-pastoris or of their evolutionary trajectories must therefore account for this initial difference. After the bottleneck associated with the origin of C. bursa-pastoris and the reduction in Ne due to the shift to selfing [48], the effective population sizes of the two subgenomes are expected to progressively converge and decrease along the same trajectory.
While this was indeed the observed overall pattern, the trajectories followed by the two subgenomes in the three populations differed: in Europe the initial level was similar to that in the Middle East but higher than in Asia and the decline of Ne of the CbpCg subgenome was delayed compared to the sudden decline experienced by the CbpCo subgenome. In contrast, in Asia the two subgenomes initially followed similar downwards trajectories but Ne increased again in both subgenomes at around 40,000 ya. In the diploid C. orientalis, there was a period of stasis followed by a steeper decline than in the tetraploid. The difference in demography across the three regions could indicate multiple origins of C. bursa-pastoris as suggested by Douglas et al. [10] and the difference between the diploid and the tetraploid could reflect a mixture of the population expansion experienced by the tetraploid and the buffering effect of tetraploidy against deleterious mutations.
There was a clearly noticeable difference between the two subgenomes in the number of accumulated deleterious mutations. Based on the strong differences in Ne, one would expect the efficacy of selection to be much higher in C. grandiflora than in C. orientalis that has a much smaller Ne [49]. In the analysis of the genetic load, we indeed observed that C. orientalis had a higher proportion of deleterious mutations than C. grandiflora. Hence, the amount of genetic load most likely was different between the CbpCg and CbpCo subgenomes of C. bursa-pastoris at the time of the species emergence. Interestingly, hundreds of thousands of generations of selfing did not totally erase the differences between the two subgenomes and, today, the CbpCo subgenome still accumulates more deleterious mutations than the CbpCg subgenome. Despite being statistically significant, the difference between subgenomes is smaller than between C. orientalis and C. grandiflora suggesting a convergence of the two subgenomes, which is also confirmed by analyses of genetic load dispersion and gene expression regulation [50].
Everything else being equal, genomic redundancy in polyploids is expected to lead to relaxation of purifying selection, and to a higher rate of accumulation of deleterious mutations than in diploids. Douglas and coworkers [10], using forward simulations, concluded that indeed the CbpCg subgenome showed an excess of deleterious mutations compared to what would have been expected from mating system shift alone and ascribed this excess of deleterious mutations to genomic redundancy. Our study also suggests that both effective population size and genomic redundancy contribute to the observed pattern. The proportion of derived deleterious mutations in both subgenomes is, with the exception of the CbpCo subgenome in the Asian population, lower than the same proportion in C. orientalis but much higher than in C. grandiflora where it is extremely low (Fig 5). Furthermore purifying selection is weaker on the CbpCo subgenome than on the CbpCg subgenome. These data can be explained as follows. First, purifying selection on the CbpCo subgenome is stronger than in C. orientalis because the latter has a smaller effective population size than C. bursa-pastoris. Second, purifying selection is weaker on the CbpCo subgenome than on the CbpCg because of genomic redundancy. Or, stated differently, because of its highest initial genetic quality the CbpCg may be more sheltered from deleterious mutations than the CbpCo subgenome.
Nucleotide diversity also demonstrated the effect of parental legacy. The CbpCg subgenome inherited from the more variable outcrosser C. grandiflora was still more diverse in all populations except the Asian one. The maintenance of part of the parental legacy in both cases suggests that, in spite of their initial differences, both subgenomes have experienced similar levels of fixation since the creation of the species. The Asian population is an exception in this regard because it was affected by secondary gene flow as discussed below. Variation in nucleotide diversity in the coding part of the genome also demonstrated similarity in the efficacy of purifying selection between the two subgenomes and their corresponding parental lineages, though the pattern in the ASI population was the reverse of that observed in the parental lineages. The effect of parental legacy in C. bursa-pastoris was also noted in Douglas et al. [10]. Thus, the effect of the genetic background of hybridizing species may not be as overwhelming as the effect of mating system but it still impacts the fate of the two subgenomes long after the species arose.
Based on coalescent simulations and the amount of shared variation between C. bursa-pastoris and its parental species Douglas et al. [10] ruled out a single founder but noted that it would be difficult to estimate the exact number of founding lineages. Douglas et al. [10] did not consider hybridization but an earlier study detected gene flow from C. rubella to the European C. bursa-pastoris using 12 nuclear loci and a coalescent-based isolation-with-migration model [51], a result which is in agreement with the general occurrence of abundant trans-specific polymorphism in the Arabidopsis genus [52]. The present study adds two new twists to the story. First, our results indicate that shared polymorphisms were not symmetrical: namely, the CbpCg subgenome showed signs of admixture with C. rubella in the EUR and ME populations, whereas the CbpCo subgenome was admixed with C. orientalis in ASI. Second, in both the whole genome and density trees, C. orientalis appears as derived from the C. bursa-pastoris CbpCo subgenome rather than the converse as one would have expected. No such anomaly was observed for C. grandiflora that, as expected, grouped at the root of the C. bursa-pastoris CbpCg subgenome. These results could be explained by a mixture of multiple origins and a recent admixture. Multiple origins seem to be common in allotetraploids [27, 53, 54] and interploidy gene flow has already been inferred for Capsella [51] and other plant genera [55–57].
Our crossing results did not reject the possibility of ongoing gene flow between C. bursa-pastoris and parental lineages in both Europe and Asia. The distribution of European and Asian populations of C. bursa-pastoris partially overlap with the distribution ranges of C. rubella and C. orientalis, respectively (Fig 1A). The exact proportion of admixture remains unclear at this stage. Taken at face value, the strongest admixture was between the ASI CbpCo subgenome and C. orientalis. Considering the overlapping estimates of f-statistic and HAPMIX, the proportion of admixture between the ASI CbpCo subgenomes and C. orientalis was around 14%-23%. The admixture between the EUR CbpCg subgenome and C. rubella was also strong, being around 8-20%. There were also signs of minor admixture in the ME population with both C. orientalis and C. rubella. This lack of a non-admixed population posed a problem of correct estimation of the proportion of admixture for both the ABBA-BABA and HAPMIX approaches.
In the ABBA-BABA test, departures from the assumptions can lead to incorrect interpretation of the results. We assumed monophyly of the three populations of C. bursa-pastoris, which may be wrong if these populations were of multiple origins. Thus, the observed shared polymorphism might be due to closer relatedness of our C. orientalis samples with the parent of the ASI population than with the parent of EUR and ME populations, but not admixture. Departures from the assumptions of the ABBA-BABA test can lead to under- or overestimated admixture. In the present case, some proportion of the variation shared between P3 and both P1 and P2 populations could be due to gene flow and not due to incomplete lineage sorting and this would lead to underestimating the amount of admixture. On the other hand, small Ne and recent divergence of the populations used in the test can inflate estimates of D [58]. Further, the behavior of D in tests involving both selfing and outcrossing species has not been assessed yet. The D statistics were significantly different from zero in all our comparisons suggesting that admixture did indeed occur in all populations of C. bursa-pastoris. The f statistic is considered less prone to be affected by these factors [58], and it was more reliable in our tests too. Its values were close to zero in the alternative combinations for the ABBA-BABA tests where we did not expect to find admixture, while D had high estimates (S12 Table). Thus, the f values are the closest to the real proportion of admixture we could obtain.
In HAPMIX, if the reference populations have non-negligible levels of admixture with each other, such that they have few differences, it will be difficult for HAPMIX to distinguish with which reference population the focal population is more likely to share ancestry, driving admixture probabilities to intermediate values. Therefore, we observed a discrepancy between the results of HAPMIX and ABBA-BABA in the estimates of admixture between the EUR CbpCo subgenome and C. orientalis. However, the results for the CbpCg subgenome largely agreed between HAPMIX and ABBA-BABA and, together with the results by Slotte et al. [51] and our crossing experiment bolsters the hypothesis of admixture between C. rubella and C. bursa-pastoris in Europe. On balance, a scenario with a single origin of C. bursa-pastoris with later rampant admixture with C. orientalis in Asia and less extensive admixture with C. rubella in Europe is consistent with our data.
On the other hand, our results could also be obtained under a scenario of multiple origins. Such a scenario seems particularly likely if one looks at Fig 4, where the histories of the CbpCo and CbpCg subgenomes are strikingly different. If we assume that C. orientalis and C. grandiflora are indeed parental lineages and there was no unknown parental lineage that went extinct, this picture can be only explained by a separate and more recent origin of the ASI population (Fig 8). However, the scenario of multiple origins and post-speciation admixture are not mutually exclusive. The signs of gene flow between EUR and C. rubella are still best explained by post-speciation admixture. The weak signs of admixture between C. bursa-pastoris and C. orientalis in EUR and ME are also difficult to fit into a scenario involving only multiple origins. A possibility is that these signs of admixture resulted from gene flow from ASI to EUR and ME within C. bursa-pastoris. The ASI population is more related to C. orientalis and the presence of its alleles in EUR and ME could be spuriously recognized as introgressed from ASI. Regardless of whether a single or a multiple origins scenario is the true one, our results demonstrate that the history of C. bursa-pastoris is far more complex than previously imagined.
Many allopolyploid species show subgenome expression bias, where one subgenome tends to be overexpressed relative to the other one [59–62]. This expression dominance is often observed in synthetic allopolyploids [63–66] and thus the major part of such preferential subgenome dominance is probably established immediately after allopolyploidization. The subgenome expression dominance is also suggested to be largely defined by parental expression differences [67, 68]. Contradictory results on patterns of subgenome specific expression in C. bursa-pastoris have been obtained so far. Douglas et al. [10] concluded that there is no strong homeologue expression bias and those few genes showing HSE could be explained by parental expression differences. However, genes with HSE do show a slight bias towards over-expression of the CbpCg subgenome inherited from C. grandiflora / rubella lineage on the Figure 3B in Douglas et al. [10]. In contrast, Steige et al. [69] reported higher expression of the CbpCo subgenome inherited from C. orientalis in three accessions, and CbpCg over-expression in a fourth one (CbpGR). Steige et al. [69] hypothesized that the over-expression of the CbpCo subgenome might be related to a higher number of transposable elements in this subgenome, but they did not find any evidence of this and could not explain the down-regulation of the CbpCo subgenome in the CbpGR accession and in the artificial hybrid between C. rubella and C. orientalis.
Considering the population histories of C. bursa-pastoris sheds some light on these discrepancies. The results of Douglas et al. [10] and Steige et al. [69] are consistent with the hypothesis that cis-regulatory differences between the C. orientalis and C. grandiflora / rubella genomes result in over-expression of the CbpCg subgenome in a hybrid comprising both genomes. Thus, in the absence of other factors, the slight over-expression of the CbpCg subgenome would be the default HSE pattern in C. bursa-pastoris. In accordance with this, we observed over-expression of the CbpCg subgenome in the ME and EUR populations that are most likely the closest to the region of origin of C. bursa-pastoris [34]. The accessions that show over-expression of the CbpCg subgenome in Douglas et al. [10] (SE14 from Sweden) and in Steige et al. [69] (CbpGR from Greece), as we now know belong to the EUR population [34]. Hence, their results are consistent with ours and expected if the HSE is defined primarily by the differences between the parental lineages. On the other hand, we observed that genes with HSE in the ASI population showed equal expression between the two subgenomes. The accessions showing over-expression of the CbpCo subgenome in Steige et al. [69] also mostly belong to the ASI population (CbpKMB and CbpGY, though not CbpDE that putatively originates from Germany). Thus, the Asian accessions show a HSE that differs from the default pattern. This difference can be caused by the selection preference for the CbpCo subgenome and/or by admixture with C. orientalis that enhanced the cis-regulatory elements of the CbpCo subgenome. The ASI population experienced a strong population bottleneck, so genetic drift played some role as well. These explanations need to be confirmed because HSE can be influenced by many factors (e.g. trans-regulatory elements, gene methylation, transposable elements), but it is clear that there are different directions of HSE in populations of C. bursa-pastoris and they are caused by the different evolutionary histories of those populations.
The reason we observed an equal expression between subgenomes in ASI, whereas Steige et al. [69] detected expression bias of the CbpCo subgenome for Asian samples, could also be due to different approaches in our analyses. First, we extracted RNA from seedlings, whereas Steige et al. [69] obtained RNA from leaves and flower buds. Variation in HSE for different tissues of C. bursa-pastoris is not characterized yet, so the CbpCo expression in seedlings may not be apparent yet. Second, we mapped reads to the C. rubella reference with masked polymorphism, whereas Steige et al. [69] used the reconstructed reference of an F1 hybrid between C. orientalis and C. rubella. The bias in our DNA data was not stronger than in Steige et al. [69], so which method is more appropriate remains to be found out.
Three salient, and sometimes unexpected, features of the evolution of the tetraploid shepherd’s purse that emerged from the present study, are its complex origin and possible admixture with diploid relatives, the long-lasting effects of the difference between its two parental species, and the importance of demography in shaping its current genomic diversity. Hence, the present study suggests that understanding the evolution of tetraploid species without paying due attention to the historical and ecological backgrounds under which it occurred could be misleading.
We obtained the whole genome sequences of 31 accessions of C. bursa-pastoris and the transcriptomes of 24 of these accessions. Transcriptome data used in this study was generated previously from seedling growth in the same growth chamber [35]. Whole genome DNA data consisted of 10 accessions sequenced previously [10] and 21 accessions sequenced in this study. New DNA samples were sequenced using the same technology as the downloaded ones (100-bp paired-end reads, Illumina HiSeq 2000 platform, SciLife, Stockholm, Sweden). The mean genomic coverage of C. bursa-pastoris samples was 47x. We also used the previously generated genomic data of 10 C. orientalis and 13 C. grandiflora samples [10]. For the analysis requiring an out-group, we used the whole genome assembly of Neslia paniculata [70]. Detailed information on the samples is provided in S1 Table.
DNA reads from each individual of C. bursa-pastoris were mapped to the Capsella rubella reference genome [70] and subsequently computationally phased the two subgenomes. We favored this approach, which has been already successfully implemented in [10], over mapping to two alternative genomes or read-sorting genotyping algorithms because the two subgenomes of C. bursa-pastoris are quite similar to each other (∼1-3% divergence), and alternative approaches would have large regions where reads cannot be assigned to the parent of origin [21, 54, 71, 72]. To account for the divergence from the reference and ensure minimal read-mapping bias between the two subgenomes, we performed tolerant read mapping using Stampy v1.0.22 [73] with the substitution rate set to 0.025. Potential PCR duplicates were marked using Picard Tools 1.115 (http://picard.sourceforge.net) and were ignored during genotyping. Genotypes were called using HaplotypeCaller from the Genome Analysis Tool Kit (GATK) v3.5 in the GVCF diploid mode and heterozygosity set to 0.015 [74]. Genotypes were filtered for depth between 6 and 100 reads (the 5th and 99th coverage percentiles, respectively) to remove low confidence genotypes due to low coverage or due to their location in repetitive regions and paralogs, which usually have abnormally high coverage. This approach produced a VCF file containing all called sites. This VCF was used in the analyses requiring both polymorphic and monomorphic sites for correct estimates. To obtain a set of SNPs with the highest confidence possible, we generated another VCF file that contained only polymorphic sites and applied more stringent filtering. We set to no-call all sites that met the following criteria: MQ < 30, SOR > 4, QD < 2, FS > 60, MQRankSum < -20, ReadPosRankSum < -10, ReadPosRankSum > 10. These filtering criteria were defined following GATK Best Practices [75] with some adjustment guided by the obtained distributions of the GATK annotation scores (S16 Fig).
To phase the C. bursa-pastoris subgenomes, we run HapCUT version 0.7 [76] on each sample from the VCF with the stringently filtered SNPs. The phased haplotype fragments were then concatenated into two sequences descended from C. grandiflora and C. orientalis (S17 Fig). The origin of haplotypes in HapCUT fragments was defined using sites with fixed heterozygotes in C. bursa-pastoris and fixed differences between the parental lineages. The fixed differences were defined as fixed between 10 C. orientalis and 13 C. grandiflora with maximum 20% of missing data per position. Fragments that had small (< 2 sites) or no overlap with variation in C. grandiflora and C. orientalis as well as those that looked chimeric (prevailing phasing state was supported by less than 90% of sites) were set to missing data (S18 Fig). Additionally, we also set to missing the sites that were defined as not real variants or not heterozygous by HapCUT (flagged with FV). We checked that the distribution of the length of phased haplotype blocks and the proportion of introduced missing data across samples were not strongly different (S19 Fig). HapCUT phasing produced the alignment that had only heterozygous sites and removed all the sites that were non-variant within but variable between individuals. We restored this inter-individual variation by introducing the same proportion of missing data into non-variant sites as it was introduced to heterozygous sites during the phasing. Similarly, we merged the phased SNPs dataset with non-polymorphic sites from the whole genome data to keep the level of heterozygosity as in the unphased data.
To ensure that there was no sign of a general bias towards the CbpCg subgenome, which is less divergent from the reference genome than CbpCo, we checked that there was no strong bias towards the reference allele in the VCF file. The bias was only ∼4% and it should not affect alleles calling in heterozygous sites even at the lowest coverage. We also surveyed the level of heterozygosity and coverage along the genome and showed that there was no regional dropout of CbpCo haplotypes (S20 Fig). In addition, the accuracy of phasing was assessed with the linkage disequilibrium (LD) variation between and within the subgenomes. From each subgenome, we randomly sampled 1000 SNPs on each chromosome (the principal SNPs) and calculated r2 between them and the next 1000 SNPs. These SNPs could originate from the same genome as the principal SNPs or from the opposite subgenome. We only considered SNP pairs where at least 5 samples had genotypes for both SNPs. As a baseline, we also calculated LD in the same way for a genome containing a random mixture of SNPs from both subgenomes. LD gradually decayed with increasing distance between SNPs within homeologues but not between them or when SNPs from both subgenomes were randomly mixed (S3D Fig), further indicating that there was no major phasing error in our data.
To verify our phasing procedure with an alternative phasing method, we utilized the HomeoRoq pipeline [77] that was successfully used to phase transcriptomic and genomic data in allopolyploid Arabidopsis kamchatica [78, 79]. While HapCUT relies on the read-back phasing of the reads mapped to one reference, HomeoRoq classifies the reads mapped to two references corresponding to the subgenomes. As the two references, we used the recently published assembly of C. bursa-pastoris [37]. We independently mapped our reads to the each assembled subgenome. Then we classified reads with the HomeoRoq to keep only unique and common reads, and to remove unclassified reads from SAM files. These SAM files were processed the same way as described above to obtain SNPs. Next, we aligned the main dataset with the HomeoRoq phased data, according to the alignment between the C. rubella reference and the C. bursa-pastoris assembly. The latter was performed by aligning each subgenome of C. bursa-pastoris to C. rubella with the LASTZ program [80] following the procedure described in [10]. Finally, we analyzed the clustering of the samples phased with the two different methods using a sliding window phylogenetic analysis as described below.
The sequences of C. grandiflora and C. orientalis were created using the GVCF files produced by Douglas et al. [10]. The variants were called as described above with additional filtering for fixed differences between the two species. For some of the analyses, where the software was not able to treat heterozygous genotypes properly, we pseudo-phased the sequences of C. grandiflora and C. orientalis by randomizing alleles in heterozygous genotypes.
The data-sets in all the analyses comprised the alignment of phased C. bursa-pastoris sequences, C. grandiflora, C. orientalis, C. rubella (the reference sequence) and N. paniculata. This alignment was filtered for missing data such that genomic positions with more than 80% of missing genotypes were removed. We also removed the repetitive sequences as annotated in Slotte et al. [70] and pericentromeric regions that we delineated based on the density of repetitive regions and missing data. The final data-set had even proportion of missing data in the three populations of C. bursa-pastoris and diploid species (S21 Fig).
Several analyses presented in this paper required polarized sequence data. The most common approach to polarizing the alleles is to use an outgroup. However, the alignment of Capsella species and N. paniculata, the nearest outgroup with a whole genome sequence available, resulted in substantial reduction of the dataset due to missing data. To overcome this drawback, as well as to track mutations’ origin on the phylogenetic branches, we reconstructed ancestral sequences for major phylogenetic splits. The reconstruction was performed on the tree that was assumed to represent a true history of the Capsella species (S22 Fig) using the empirical Bayes joint reconstruction method implemented in PAML v4.6 [81].
To assess the degree of differentiation among populations for the two subgenomes, we estimated absolute divergence (Dxy) and nucleotide diversity (π) of the phased genomes using a sliding window approach. The estimates were calculated on non-overlapping 100 Kb windows using the EggLib Python module [82]. The p-values for the difference in mean values were estimated using 10,000 bootstrap resamples from 100 Kb windows.
We reconstructed changes of Ne over time with both PSMC [83] and SMC++ [39]. We first masked potential CpG islands and all nonsynonymous sites in the genome to avoid bias caused by variation in mutation rates or selective effects. We randomly paired haplotypes for estimation in C. orientalis and did the same for estimations based on the two subgenomes of C. bursa-pastoris. SMC++ was run on all samples from a population, with default parameter settings. For PSMC runs, we set parameters to “-N25 -t15 -r5 -p 4+25*2+4+6”. Variation in Ne was estimated using 100 bootstrap replicates and three different pairs. We chose a mutation rate equal to the mutation rate of A. thaliana, μ = 7 × 10−9 per site per generation [84] and a generation time of 1 year for all Capsella species.
We reconstructed a whole genome phylogeny to explore the relationship between the phased subgenomes of the three populations of C. bursa-pastoris as well as its parental species. To investigate the local phylogenetic relationships along the genome, we also conducted a sliding window phylogenetic analysis using non-overlapping 100 Kb windows. In both analyses, phylogenetic trees were reconstructed using the neighbor-joining algorithm and absolute genetic distance in R package ape [85]. Additionally, a whole genome phylogenetic tree was also reconstructed using the maximum-likelihood approach with the GTRGAMMA model and 100 bootstrap replicates in RAxML v8.2.4 [86] (S23 Fig). The trees from the sliding window analysis were described by counting the frequency of monophyly of different groups with the Newick Utilities [87]. The variation in topology across the genome was also described using topology weighting implemented in TWISST [40]. The weighting was estimated for 100 SNPs windows where each sample was genotyped for at least 50 SNPs. To test for the difference in mean topology weighting, we fitted the generalized linear model with a binomial distribution and performed multiple comparisons for the contrasts of interest with the glht function from the multcomp library in R [88].
To evaluate the presence of admixture between the parental species and C. bursa-pastoris, we calculated the ABBA-BABA based statistics, D, an estimate of departure from incomplete lineage sorting, and f, an estimate of admixture proportion [41, 42]. These statistics and their significance, which was estimated with a 1Mb block jackknife method, were calculated from population allele frequencies with scripts from Martin et al. [89]. We also used HAPMIX [90] to infer haplotype blocks of admixture with the diploid C. grandiflora, C. rubella, and C. orientalis into the three populations of C. bursa-pastoris for each phased subgenome. We removed sites with more than 20% missing data for each population. The remaining missing data was imputed for the parental populations used in each analysis. As this method determines the probability of ancestry from a diploid progenitor population relative to a non-admixed C. bursa-pastoris subgenome population, we defined regions of the subgenomes as putatively admixed if the probability of ancestry from the progenitor diploid was greater than 50%.
To search for selective sweeps, we used SweepFinder2 [91]. SweepFinder2 was run on the data-set that besides polarized SNPs also included fixed derived alleles. This enables accounting for variation in mutation rate along the genome and increases power to detect sweeps [92]. The critical composite likelihood ratio (CLR) values were determined using a 1% cut-off of the CLR values estimated in 100 simulations under a standard neutral model. The simulations were performed with fastsimcoal2 [93]. We assumed a mutation rate of 7 × 10−9 per site per generation, the population effective sizes for every population and subgenome were inferred from the θ values approximated by genetic diversity (π), and the average recombination rate was estimated using LDhelmet v1.7 [94]. In addition, we estimated the ratio between nucleotide diversity at 0-fold (π0) and 4-fold degenerate sites (π4) in 5-6 samples with the lowest amount of missing data in each group.
To identify differences in genetic load between populations of C. bursa-pastoris (as well as to assess the effect of selfing on accumulation of deleterious mutations), we classified mutations into tolerated and deleterious ones using SIFT4G [44]. We built the SIFT4G Capsella rubella reference partition database and used it to annotate our SNPs dataset. Then we analyzed the frequencies of tolerated and deleterious mutations. We also verified this analysis by using A. thaliana SIFT4G database and annotating C. bursa-pastoris according to the alignment between the two species. This verification was performed to make sure that the observed results were not due to a reference bias because C. rubella is closer to C. grandiflora than to C. orientalis. To get only the annotation of the mutations that occurred after speciation of C. bursa-pastoris, we polarized the mutations with the reconstructed ancestral sequences (see above) and analyzed only derived mutations. We verified this polarization by analyzing only species(subgenome)-specific mutation (e.g. mutations unique to C. bursa-pastoris CbpCo subgenome, C. bursa-pastoris CbpCg subgenome, C. orientalis, C. grandiflora, and C. rubella) (S24 Fig). All the counts were presented relative to the total number of annotated sites to avoid bias caused by variation in missing data between samples. The means of the genetic load were compared using the generalized linear model as we did for the topology weighting except that here we used a quasibinomial distribution due to overdispersion.
Mapping of RNA-Seq reads to the C. rubella reference genome was conducted similarly to the mapping of DNA data using Stampy v1.0.22 [73] with the substitution rate set to 0.025. Although potential PCR duplicates are usually not removed from RNA-Seq data, for the allele-specific expression analysis removing duplicates is recommended [95]. We marked duplicates with Picard Tools 1.115 and did not use them during the genotyping and homeolog-specific expression assessment. Variants were called using HaplotypeCaller (GATK) with heterozygosity set to 0.015, and minimum Phred-scaled call confidence of 20.0, and minimum Phred-scaled emit confidence of 20.0 as recommended for RNA-Seq data in GATK Best Practices [75]. Among the obtained polymorphic sites those that had MQ < 30.00, QD < 2.00, FS > 30.000 were filtered out. Calls with coverage of fewer than 10 reads were also excluded. Alleles counting was carried out using ASEReadCounter from GATK.
Homeolog-specific expression was assessed within the statistical framework developed by Skelly et al. [45]. This framework uses a Markov chain Monte Carlo (MCMC) method for parameter estimation and incorporates information from both RNA and DNA data to exclude highly biased SNPs and calibrate for the noise in read counts due to statistical sampling and technical variability. First, we used DNA data to identify and remove SNPs that strongly deviated from the 0.5 mapping ratio. Second, we estimated the variation in allele counts using unbiased SNPs in the DNA data. Next, we fitted an RNA model using parameter estimated from DNA data in the previous step. Finally, we calculated a Bayesian analog of false discovery rate (FDR) with a posterior probability of homeologue specific expression (HSE) > 0.99 and defined genes with significant HSE given the estimated FDR. All inferences were performed using 200,000 MCMC iterations with burn-in of 20,000 and thin interval of 100. Each model was run three times with different starting parameters to verify convergence.
DNA sequence data generated for 21 accessions of C. bursa-pastoris is submitted to the NCBI database under the sequence read archive number SRP126886. Previously generated DNA sequence data for 10 accessions of C. bursa-pastoris, 10 accessions C. orientalis and 13 accessions C. grandiflora is available in the NCBI (SRP050328, SRP041585, SRP044121). RNA-Seq data is also available in the NCBI (SRA320558). Both phased and unphased SNPs, phylogenetic trees, reconstructed ancestral sequences, estimates of π and Dxy with sliding window approach, results of PSMC and SMC++, SIFT4G annotations, CLR estimates of sweepFinder2, TWISST and HAPMIX outputs, homeologue-specific gene expression values are deposited to the Open Science Framework Repository (DOI: 10.17605/OSF.IO/5VC34) [96].
|
10.1371/journal.pgen.1002923 | Genome-Wide Association Studies Identify Heavy Metal ATPase3 as the Primary Determinant of Natural Variation in Leaf Cadmium in Arabidopsis thaliana | Understanding the mechanism of cadmium (Cd) accumulation in plants is important to help reduce its potential toxicity to both plants and humans through dietary and environmental exposure. Here, we report on a study to uncover the genetic basis underlying natural variation in Cd accumulation in a world-wide collection of 349 wild collected Arabidopsis thaliana accessions. We identified a 4-fold variation (0.5–2 µg Cd g−1 dry weight) in leaf Cd accumulation when these accessions were grown in a controlled common garden. By combining genome-wide association mapping, linkage mapping in an experimental F2 population, and transgenic complementation, we reveal that HMA3 is the sole major locus responsible for the variation in leaf Cd accumulation we observe in this diverse population of A. thaliana accessions. Analysis of the predicted amino acid sequence of HMA3 from 149 A. thaliana accessions reveals the existence of 10 major natural protein haplotypes. Association of these haplotypes with leaf Cd accumulation and genetics complementation experiments indicate that 5 of these haplotypes are active and 5 are inactive, and that elevated leaf Cd accumulation is associated with the reduced function of HMA3 caused by a nonsense mutation and polymorphisms that change two specific amino acids.
| Cadmium (Cd) is a potentially toxic metal pollutant that threatens food quality and human health in many regions of the world. Plants have evolved mechanisms for the acquisition of essential metals such as zinc and iron from the soil. Though often quite specific, such mechanisms can also lead to the accumulation of Cd by plants. Understanding natural variation in the processes that contribute to Cd accumulation in food crops could help minimize the human health risk posed. We have discovered that DNA sequence changes at a single gene, which encodes the Heavy Metal ATPase 3 (HMA3), drives the variation in Cd accumulation we observe in a world-wide sample of Arabidopsis thaliana. We identified 10 major HMA3 protein variants, of which five contribute to reduce Cd accumulation in leaves of A. thaliana.
| Cadmium (Cd) is a significant pollutant and naturally occurring trace element that is potentially toxic to both plants and animals, including humans. The human body receives Cd from many sources, but mainly from food, drinking water and smoking [1]–[3]. An important step for Cd to enter the human food chain is its accumulation in plant tissues, especially the aerial parts that form the majority of the food sources consumed either directly by humans or through eating meat produced from animals raised on a plant-based diet [4]. High level accumulation of Cd in the harvestable, above-ground tissues of plants is also essential for successful phytoremediation of environments contaminated with potentially toxic concentrations of Cd [5]. Understanding the mechanism of Cd accumulation in plants is therefore an important step towards being able to control the health risk environmental Cd poses.
Accumulation of Cd in the aerial tissues of plants is determined by several factors, including the bioavailability of Cd in the soil, uptake from the soil solution by roots and radial transport within the root to the vascular system, translocation from the root, and storage in the above ground tissues. Plants take up Cd from the soil through the symplastic pathway, though apoplastic transport may also be important [6]. Translocation of Cd from the roots to the shoots requires loading of Cd into the xylem from the symplast in the stele. Xylem loading of Cd in plants requires the Heavy Metal ATPases AtHMA4 and/or AtHMA2 [7]–[13].
Unlike most animals, plant cells have large vacuoles that can be used for Cd detoxification through compartmentalization and storage. To date three types of transporters have been identified that are responsible for sequestering Cd into plant vacuoles. They are CAX-type antiporters, such as CAX2 and CAX4 [14], [15], the Heavy Metal ATPase 3 (HMA3) 16–19 and phytochelatin transporters ABCC1 and ABCC2 [20], [21]. Among these, the H+/Cd2+-antiporters and HMA3 transport the ionic form of Cd (Cd2+), whereas the phytochelatin transporters transport Cd chelated with phytochelatins [14]–[21]. Further, the various transport systems involved in the accumulation of leaf Cd play different roles. Heterologous expression of AtCAX2 and AtCAX4 in all tissues enhanced Cd accumulation in tobacco leaves [15], whereas selective expression only in roots decreased leaf Cd accumulation [14]. Similarly, A. thaliana plants over expressing AtABCC1 and AtABCC2 in all tissues accumulate higher leaf Cd than controls [20]. Such data suggest that enhancement of a root sink for Cd reduces foliar Cd accumulation where as an enhanced leaf sink can increase foliar Cd accumulation.
HMA3 shows high amino acid sequence similarity to both HMA2 and HMA4, but its function is distinct from either [10]. In contrast to the plasma membrane localization of both HMA2 and HMA4 [10], HMA3 is localized to the tonoplast [16]–[19]. Studies have established that HMA3 orthologs in many plant species function in sequestering heavy metals into the vacuole, but the metal specificity and their role in leaf Cd accumulation appear to vary. In rice, HMA3 was identified as the responsible locus underlying a shoot Cd accumulation QTL [17]–[18]. Functional HMA3 was found to specifically restrict Cd accumulation in rice seeds and leaves [17]. HMA3 is highly expressed in the Zn/Cd hyperaccumulators Noccaea caerulescens (previously named Thlaspi caerulescens) and Arabidopsis halleri [16], [22], suggesting it may play a positive role in Zn/Cd hyperaccumulation. Heterologous expression of HMA3 from rice and A. thaliana in Saccharomyces cerevisiae (yeast) suggests that HMA3 can function to sequesters Cd into vacuoles [18], [23], whereas HMA3 from A. halleri appears to function in Zn but not Cd detoxication [22]. Further, overexpression in A. thaliana of HMA3 from A. thaliana enhanced Cd, Zn and Co tolerance and accumulation [19]. It is not clear if these differences in substrate specificity of HMA3 in the different species are a result of evolutionary divergence or the use of different experimental systems. The role of HMA3 in regulating foliar Cd accumulation in A. thaliana also remains inconclusive. However, the overall evidence supports the conclusion that HMA3 functions at the tonoplast in vacuolar compartmentalization of multiple heavy metals including Cd, Zn, cobalt (Co) and lead (Pb) [16]–[19], [23].
Natural variation is a powerful resource for studying the molecular function of genes as well as understanding their ecological function [24]–[29]. Natural variation has been observed at HMA3 in a limited number of species including rice, N. caerulescens and A. thaliana accessions [8], [16]–[19], and this variation has been established to impacts foliar Cd accumulation in rice and N. caerulescens. However, to date population-wide variation in foliar Cd and the potential link with variation at the HMA3 locus have not been investigated in any species. Arabidopsis thaliana is broadly distributed throughout the northern hemisphere growing in a diversity of climatic, edaphic and altitudinal habitats where it is likely to be exposed to a range of selective pressures [30]. The A. thaliana genome contains extensive diversity throughout its global range and at least part of this genetic diversity is associated with broad phenotypic variability [31], and also local adaptation [27]–[29]. This extensive natural variation in A. thaliana has also been utilized to uncover specific genes and QTLs involved in controlling natural variation in a variety of traits [24].
Traditionally, QTLs have been identified using experimental populations such as recombinant inbred lines (RILs) in which homozygous alternative alleles are segregating. These mapping populations have high power to detect QTLs because each allele is present in 50% of the recombinant lines. However, these populations are time consuming to develop and also suffer from low resolving power due to the limited number of recombination events that occur during their development. This leads to the identification of QTLs that span relatively large genomic regions, making identification of causal genes more difficult. Further, each mapping population is generated from a cross between two parental accessions potentially captures only two alternative alleles of any locus. This leads to very limited sampling of natural allelic diversity in a population and the low probability of detecting important minor alleles. An alternative approach to using experimental recombinant populations for QTL analysis is genome-wide association (GWA) mapping. This approach takes advantage of the large number of historic recombination events that have occurred within a population, and couples these events with linked DNA polymorphisms in order to associate phenotypic diversity with a relatively small region of the genome. However, unlike RIL populations where each allele is at a frequency of 0.5, in samples of natural populations rare alleles will occur at lower frequency making it difficult to detect their phenotypic effect. GWA mapping has been successfully used in A. thaliana [26], [32]–[37], rice [38]–[40]) and maize [41], [42] for the identification of QTLs and candidate genes for various ecological and agricultural traits. However, few if any of these studies have verified the candidate genes and polymorphisms identified using GWA mapping. Here, we report the use of GWA mapping for the identification of a major QTL for foliar Cd accumulation in A. thaliana. Further, we extend the GWA mapping with fine mapping in an experimental F2 population, genetic and transgenic complementation and with analysis of whole genome re-sequencing data for the identification of HMA3 as the causal gene, and the identification of the specific protein coding haplotypes of HMA3 that underlie natural variation in leaf Cd accumulation in the global A. thaliana population.
In a previous GWA study using a population of 93 A. thaliana accessions we were unable to identify a major peak of linked SNPs associated with leaf Cd accumulation, though we did identify several SNPs with −log(p-value)>5 [31]. The absence of strong associations might be a result of the small population size used in this previous study combined with an underpowered experimental design (fewer control genoypes in each experimental block for inter block normalization). This is supported by the observation that in the Atwell et al. [31] study, which used a population of 93 accessions, only two SNPs linked to HKT1 were observed to be significantly associated with leaf Na, whereas in an expanded population of 349 accessions Baxter et al. [26] observed 12 SNPs significantly associated with leaf Na and linked to HKT1. We therefore employed this enlarged mapping population of 349 accessions [26] for our current GWA study to identify reliable QTLs contributing to leaf Cd accumulation in the globally sampled A. thaliana population.
Each accession was grown in a controlled common garden in potting mix soil with Cd supplied in the soil at a sub-toxic concentration of 90 µg kg−1. After 5-weeks of vegetative growth leaves were harvested individually from each plant and analyzed for Cd using inductively couple plasma mass spectrometry (ICP-MS) as described previously [43]. After normalization across experimental blocks using common genotypes and normalization of the ICP-MS data to an estimated leaf dry weight [26], we observed that leaf Cd concentrations varied across the 349 accessions from 0.5 to 2.0 µg g−1 dry weight (Figure 1A). From the 349 accessions 337 had previously been genotyped using the 256K SNP-tilling array Atsnptile 1, which contains a probe sets for 248,584 SNPs [26]. Using the genotype and leaf Cd concentrations for this subset of 337 accessions we performed a GWA analysis in which a population structure correction method implemented in EMMA was applied [31], [44]. In this genome-wide scan we observed a single region on chromosome 4 that contained multiple SNPs highly associated with leaf Cd concentrations (Figure 1B and 1C). In a 100 kb interval within this region we observed 54 SNPs significantly associated with leaf Cd (p-value<10−5), 39 of which were highly significantly associated with leaf Cd concentration (p-value<10−10). The most highly associated SNP was found at Chr4:14736658 (−log (p-value) = 21.32), which explains 30% of the total variance in leaf Cd accumulation we observed. In contrast, no SNP contributing to more than 8% of the variance in leaf Cd was observed in any other region of the genome, suggesting the causal gene in linkage with SNP Chr4:14736658 is the major genetic locus responsible for natural variation in leaf Cd accumulation in A. thaliana. At this peak SNP accessions with the cytosine (C) allele have leaf Cd on average 34.4% higher than accessions with thymine (T) allele. The minor allele (T) is represented in 42.4% of the population of 337 accessions. Within 40 kb either side of SNP Chr4:14736658 (LD decay distance in this region) there are a total of 13 genes (Table 1), including HMA2 and HMA3. Given that HMA2 and HMA3 have been shown to function as Cd and/or Zn transporters [10], [19], [23,], these two genes made good candidates for the causal gene underlying the observed Cd QTL centered on SNP Chr4: 14736658.
To some extent, the geographic distribution of a genetic locus may reflect if there is selection for a particular allele in a certain environment. Using a genotyped worldwide collection of 1178 A. thaliana accessions in which the genotype at SNP Chr4:14736658 is known, we plotted the geographical distribution of the two alleles at SNP Chr4:14736658. From this map we observe both alleles are widely distributed within Europe and central Asia and the USA (Figure S1). However, the enrichment of the two alleles varies by geographical region. For example, accessions with the T allele are enriched in the United Kingdom and western France, while accessions with the C allele predominantly occur in eastern Spain, eastern France, Germany, the Czech Republic and Sweden (Figure S1). The east-west structure in the geographical distribution of the alternate alleles at SNP Chr4:14736658 in Europe may well be related to the known large-scale A. thaliana metapopulations that also have an east-west structure, related to range expansion from various southern glacial refugia [45].
To further genetically characterize the Cd QTL on chromosome 4 identified using GWA analysis we generated an experimental F2 population in which the alternate alleles of the diallelic SNP Chr4: 14736658 were segregating. To achieve this we outcrossed the low leaf Cd A. thaliana accession CS28181, with a T at SNP Chr4: 14736658, to Col-0 which contains average leaf Cd and has a C at SNP Chr4: 14736658. The F1 generation of this cross had the same leaf Cd concentration as the CS28181 parent (Figure 2A), indicating that the CS28181 allele for leaf Cd accumulation is dominant over the Col-0 allele. eXtreme Array Mapping (XAM) was performed in which we combined bulk segregant analysis (BSA) with microarray genotyping [46], [47] using the CS28181×Col-0 F2 mapping population. A total of 314 F2 individuals in 4 experimental blocks were grown vegetatively in potting mix soil, leaves harvested after 5 weeks and analyzed by ICP-MS for Cd. Data was normalized across experimental blocks using the parental genotypes common within each block and normalization to estimated dry weight [26]. Consistent with the dominance of the CS28181 allele observed in the F1 generation the center of the distribution is shifted towards CS28181 leaf Cd accumulation (Figure 2B). 58 plants from the extreme high side of the Cd distribution (leaf Cd>0.85 µg g−1 dry weight) and 79 plants from the extreme low side of the Cd distribution (leaf Cd<0.55 µg g−1 dry weight) were pooled separately. Genomic DNA from each pool was isolated, labeled and hybridized to the Affymetrix SNP-tilling array Atsnptile 1. The allele frequency differences for all polymorphic SNPs were assessed according to hybridization signals as previously described [47]. Based on the allele frequency differences between the two pools, the causal locus of leaf Cd accumulation was mapped to a 3 Mb interval on chromosome 4 (from 13 Mb to 16 Mb) (Figure 3A), with the peak centered on the mapping interval identified in our GWA analysis (Figure 1B and 1C). The observation of a single strong XAM peak (Figure 3A) provides good supporting evidence for there being a single major QTL responsible for natural variation on leaf Cd accumulation.
PCR-based genotyping was used to further narrow down the mapping interval obtained using XAM. 314 F2 recombinants from the CS28181×Col-0 cross were individually genotyped at five CAPS markers spanning the 13–16 Mb interval on chromosome 4 and 20 recombinants between marker Fo13M and Fo16M were identified. According to the genotypes of these 20 recombinants and their leaf Cd accumulation in the F2 and/or F3 generations, we mapped the casual locus to a 500 kb region between marker Fo14.5M and marker Fo15M (Figure 3B), in which HMA2 and HMA3 are located (Figure 3C). Our linkage mapping in the CS28181×Col-0 mapping population confirmed the results we obtained from our GWA analysis, and further supported HMA2 and/or HMA3 as candidate genes driving the natural variation in leaf Cd accumulation we observed in our global A. thaliana population sample.
As both association mapping and linkage mapping in A. thaliana indicate HMA2 and HMA3 are the best candidates for being responsible for natural variation of leaf Cd accumulation, we sequenced the genomic region covering the two genes in the accession CS28181, including the promoters, intergenic regions and 3′ termini. According to the assembled sequence, there are a total of 23 polymorphic sites between CS28181 and Col-0, of which 21 are SNPs and two are 1-bp deletion/insertions (Table 2). Of those polymorphic sites three are located in HMA3 exons, three in HMA2 exons, six in the HMA3 promoter and two in the HMA2 promoter. The polymorphisms in exons lead to differences of two amino-acid residues in HMA2 (Thr131Ala [CS28181 to Col-0 applied throughout] and Thr759Ala) and three amino-acid residues in HMA3 (Asn426Tyr, Ile448Arg and Leu543Stop). The premature stop codon in Col-0 HMA3 is likely to eliminate the activity of the translated protein as it will be truncated after amino acid 542. The truncated product would lack the conserved ATP binding site and it is therefore likely to be non-functional [8]. It is also possible that the observed SNPs may contribute to differences in gene function between the CS28181 and Col-0 alleles.
Given that the sequence polymorphisms cannot exclude HMA2 as a possible candidate gene, we used transgenic complementation to determine which gene underlies the observed leaf Cd QTL on chromosome 4. We constructed DNA vectors to introduce the CS28181 genomic DNA fragments of HMA3 (HMA3CS28181) and HMA2 (HMA2CS28181) separately into Col-0. These separate genomic DNA fragments included the 2 kb promoter region, the whole gene body and the 3′ terminal for both HMA2CS28181 and HMA3CS28181. In the T2 generation, transgenic lines were grown vegetatively in potting mix soil, leaves harvested after 5-weeks and analyzed for Cd using ICP-MS. Because transgenic plants were segregating in the T2 generation, the reporter gene GUS was used as a marker for the transformation construct using histochemical staining in order to assess if an individual had the transgenic fragment. Individuals without the transgenic fragment were removed from further analysis. All seven independent Col-0 lines transformed with HMA3CS28181 showed significantly reduced leaf Cd compared to Col-0 wild-type, with leaf Cd similar to, or even lower than CS28181 (Figure 4A). However, none of the Col-0 lines transformed with HMA2CS28181 had reduced leaf Cd concentrations compared with Col-0 wild-type (Figure 4B). These results clearly indicate that it is HMA3 and not HMA2 that is the causal gene underlying the leaf Cd QTL we observe on chromosome 4.
To determine which tissue (root or shoot) is responsible for controlling the HMA3 dependent variation in leaf Cd between Col-0 and CS28181 we performed a reciprocal grafting experiment (Figure 5). Both self-grafted and non-grafted Col-0 had similar leaf Cd concentrations, as did the self-grafted and non-grafted CS28181. Further, both self-grafted and non-grafted CS28181 showed significantly lower leaf Cd than Col-0 (self-grafted or non-grafted) as expected. Grafted plants with a CS28181 root and a Col-0 shoot contained leaf Cd concentrations the same as CS28181 (self-grafted or non-grafted). Whereas, plants with a Col-0 root and a CS28181 shoot had leaf Cd concentrations the same as Col-0 (self-grafted or non-grafted). From this experiment we conclude that the variation in leaf Cd accumulation between Col-0 and CS28181, determined by HMA3, is driven by physiological processes in the root.
Given that in many cases natural phenotypic variation is caused by cis-element polymorphisms driving changes in the level of gene expression [24], we used quantitative Reverse Transcription PCR (qRT-PCR) to quantify steady state levels of HMA3 mRNA in Col-0 and CS28181. We observe that HMA3 is primarily expressed in roots of Col-0, though we do observe expression in leaves to a lesser degree (Figure 6A). This is consistent with previous observations also using qRT-PCR [22]. Primary expression of HMA3 in the root is also consistent with our observation that the root controls the HMA3-dependent variation in leaf Cd (Figure 5). However, we observe no significant difference between the steady state levels of HMA3 mRNA in roots of Col-0 and CS28181 (Figure 6A). These results suggest that differences in the level of expression of HMA3 between Col-0 and CS28181 cannot explain the differences in HMA3-dependent leaf Cd accumulation.
To further extend this analysis we used qRT-PCR to examine the steady state levels of HMA3 mRNA in roots of 14 A. thaliana accessions grown on media solidified with agar (Figure 6B), representative of eight of the HMA3 protein coding haplotypes we have identified (Figure 7) from a set of 149 re-sequenced accessions. We compared the root expression of HMA3 to the leaf Cd accumulation in the same plants across all 14 accessions and observed that expression of HMA3 varies among these 14 accessions but there is no correlation (R2 = 0.005) between HMA3 mRNA levels and leaf Cd accumulation (Figure 6B). Further, we found a strong correlation between leaf Cd of the same accessions grown in potting mix soil and on agar solidified media (Figure S2). These results support our conclusion that root-driven HMA3-dependent variation in leaf Cd accumulation in A. thaliana is not due to variation in HMA3 expression level.
Given that HMA3 expression level polymorphisms do not appear to drive HMA3-dependent variation in leaf Cd in A. thaliana we investigated the possibility that this variation is due to differences in the function of the HMA3 protein. To test this hypothesis we examined the predicted protein coding haplotypes of HMA3 from a set of 149 genome re-sequenced A. thaliana accessions that we had previously phenotyped in potting mix soil grown plants for leaf Cd accumulation (Table S1). A total of 31 amino acid substitutions were found in the HMA3 predicted amino acid sequence within this set of 149 genome re-sequenced accessions. Fourteen of those substitutions are only present in one accession (), which could represent sequencing errors. Seven of them are only found in 2–4 accessions, which are also considered as rare alleles. Removal of these 21 polymorphisms left 10 amino acid substitutions which were used to conservatively estimate the existence of 10 HMA3 protein coding haplotypes. (Figure 7A; Table S1). Given that the premature stop codon likely produces an inactive truncated HMA3 protein [8], [19] we put the two haplotypes with the 1-bp deletion causing the premature stop codon together and classify them as Type X (Figure 7A). We identified nine accessions in this class (Figure 7A). For each haplotype group we calculated the average leaf Cd concentration from leaf Cd data collected on all 149 accessions grown and analyzed individually (Figure 7B). A clear association between haplotype groups and leaf Cd concentration is observed, with accessions with haplotype I–V and haplotypes VI–X forming two distinctly separate low and high leaf Cd groups (Figure 7B).
Given that the group X haplotype is defined by a loss of function allele of HMA3 [8], [19], we propose that elevated leaf Cd in this group is caused by loss of HMA3 activity. This is supported by the fact that the Col-0 allele of HMA3 (which falls into haplotype group X) is a recessive allele compared to CS28181 (haplotype group I) (Figure 2). If elevated leaf Cd is associated with hypofunctional alleles of HMA3, such as the loss of function alleles in haplotype group X, then the haplotypes in groups VI, VII, VIII and IX are also likely to represent hypofunctional alleles of HMA3. Conversely, the CS28181 allele of HMA3 is functional since it is dominant over the loss of function Col-0 allele (Figure 2), and therefore low leaf Cd is associated with hyperfunctional alleles of HMA3 in protein coding haplotype groups I–V. Consistent with this, the Ws-2 allele of HMA3, which was previously established to be functional [19], falls into haplotype group III. To test our predicted functional classifications of the different HMA3 protein coding haplotypes we examined leaf Cd accumulation in F1 plants from crosses between Col-0 and accessions with HMA3 protein coding haplotypes VII, VIII and IX. None of these three haplotypes were able to complement the loss of function HMA3 allele in Col-0 (Figure 7D), establishing that HMA3 alleles in protein coding haplotype groups VII, VIII, IX are hypofunctional like the Col-0 allele.
Interestingly, the classification of the functional protein coding haplotype groups is consistent with our GWA study with the T allele at SNP Chr4:14736658 being highly enriched in most of the functional haplotype groups, while the C allele is highly enriched in the hypofunctional groups (Figure 7A). However, this association is not perfect as might be expect for a linked yet non-causal polymorphism. We do though observe an absolute linkage between the HMA3 protein coding haplotypes and function. The substitution of a glutamine at residue 564 (Q564) with a methionine (M564), or a tyrosine at residue 480 (Y480) with an aspartic acid residue (D480), are absolutely associated with loss of function of HMA3, reflecting the tight linkage between phenotype and genotype that would be expected for these putative casual polymorphisms.
A comparison of the functional HMA3 orthologs in Arabidopsis halleri and rice [17], [22] with the functional HMA3 protein coding haplotypes (I–V) in A. thaliana reveals that in A. halleri and rice the Q564 and M480 are also conserved, further supporting the conclusion that changes at these two residues in A. thaliana generate a non-functional HMA3 protein. The location of these two amino acid residues in the important ATP binding domain (Figure 7C) is also consistent with this inference.
In rice, HMA3 was found to specifically control leaf accumulation of Cd, but not Zn or other elements [17]. In A. thaliana HMA3 was found to be involved in the transport of Cd and also possibly Zn, Co and Pb [19], [23]. To investigate a possible function for HMA3 in controlling variation in accumulation of these trace metals in A. thaliana we compared the foliar concentrations of Zn and Co in Col-0 (hypofunctional allele of HMA3) with CS28181 (hyperfunctional allele HMA3). A significant difference in leaf Zn was observed between Col-0 and CS28181 (Figure 8), with Col-0 having increased leaf Zn concentrations compared to CS28181. This elevated Zn was partially reduced by transformation of Col-0 with a genomic DNA fragment containing the CS28181 HMA3 promoter, gene body and 3′ terminus (Figure 8). No significant differences in leaf Co were observed between CS28181, Col-0 or Col-0 transformed with the CS28181 HMA3 genome clone (Figure 8). From these results we conclude that the hypofunctional allele of HMA3 in Col-0 also affects the concentration of leaf Zn but has no effect on leaf Co.
Using GWA mapping on 349 A. thaliana accessions selected from a worldwide collections of 5810 accessions and genotyped at approximately 250,00 SNPs [26], [48] we successfully identified a single strong peak of SNPs associated with leaf Cd accumulation near to HMA2 and HMA3 (Figure 1B). The most highly associated SNP in this peak accounting for 30% of the total variance in leaf Cd after accounting for population structure. To confirm the GWA mapping result and identify the causal gene, we performed linkage mapping and transgenic complementation experiments and established that polymorphisms at HMA3 are the major genetic determinant for the variation we observe in leaf Cd in this global A. thaliana population sample.
Expression level polymorphisms in HMA3 do not appear to be responsible for the HMA3-dependent variation in leaf Cd we observe. This contrasts what we have previously found for natural variation in A. thaliana leaf Na and Mo levels which are driven by expression level polymorphisms in HKT1 and MOT1, respectively [26], [49], [50]. In the reference accession Col-0 it had previously been observed that a 1-bp deletion in HMA3 results in a premature stop codon, which was believed to cause a loss of function HMA3 variant [8], [19]. However, the effect of this loss of function allele of HMA3 on leaf Cd accumulation was not investigated, though a loss of function T-DNA insertion allele in Ws-2 was known to increase sensitivity to Cd [8], [19]. We compared the protein coding haplotypes of HMA3 across 149 accessions and identified 10 major HMA3 protein coding haplotypes. From the association of these haplotypes with leaf Cd concentrations in these accessions, and a comparison with the predicted amino acid sequence of HMA3 from A. halleri and rice, we infer that five of the protein coding haplotypes of HMA3 in A. thaliana are functional and the other five are non-functional. To confirm this hypothesis, we performed genetic complementation tests for 4 accessions representing 4 haplotype groups (I, VII, VIII and IX). Our results show that the known Col-0 loss of function protein coding haplotype cannot be complemented by HMA3 alleles from haplotype group VII, VIII and IX (represented by Duk, Van-0 and Ler-0), which establishes that these three HMA3 protein coding haplotypes also represent loss of function alleles. In contrast, the group I haplotype (represented by CS28181) is able to complement the loss of function allele in Col-0 confirming that this protein coding haplotype represents a functional allele of HMA3. From this we conclude that a major portion of the genetically determined natural variation in leaf Cd observed in our world-wide A. thaliana population sample is driven by variation in the level of function of the HMA3 protein. The sequence differences among active and inactive A. thaliana protein coding haplotypes of HMA3 allowed us to further conclude that amino acid changes Gln564Met and Tyr480Asp are responsible for the impaired activity of the HMA3 alleles in the protein coding haplotype group VI–IX. Furthermore, we could also confirm the previous observation that the premature stop codon is responsible for loss of function of protein coding haplotype group X. Interestingly, these amino acid changes occur in the ATP binding domain (Figure 7B) with Gln564Met and Tyr480Asp potentially affect ATP binding or ATP hydrolysis. Although further evidence is necessary to confirm the biochemical effects of these amino acids changes, our discoveries contribute to our understanding of the functional mechanism of HMA3 and other heavy metal ATPases.
It is interesting to note that the hypofunctional HMA3 alleles we identify are more common than the hyperfunctional alleles in the genome re-sequenced population of 149 A. thaliana accessions we investigated. Ninety seven accessions contained hypofunctional HMA3 protein coding haplotype, suggesting that the hypofunctional HMA3 alleles are widely distributed in the A. thaliana population. This is supported by the high frequency and wide geographical distribution of the C genotype at SNP Chr4: 14736658 linked to the hypofunctional HMA3 allele. This raises the question of is the effect of the hypofunctional alleles of HMA3 neutral or do they provide an adaptive benefit to the plant under certain environmental conditions? Recent genome-wide estimations of selection in A. thaliana did not reveal any evidence for selection at the HMA3 locus [29]. However, it is possible that alleles could be adaptive in one environment but neutral in another [27]. Signals of selection of such locally adaptive alleles would be more difficult to identify in the world-wide A. thaliana sample used [29]. The adaptive function of these natural alleles of HMA3 in Cd or Zn homeostasis, if there is any, remains unknown. We could speculate that the hypofunctional HMA3 in A. thaliana might be neutral in soils with normal concentrations of Zn and beneficial in soils with low Zn where translocation of Zn to shoots needs to be maximized. Alternatively, the hyperfunational allele may be neutral in regions of low Cd and beneficial in areas of elevated Cd where enhanced vacuolar sequestration of Cd would potentially reduce the plants sensitivity to Cd. Further studies are required to eliminate the need for such speculation.
Genetics research indicates that loss of function of rice HMA3 only affects accumulation of Cd, and not other heavy metals, and based on this observation it was concluded that HMA3 is a highly specific Cd transporter [17]. Our results presented here for A. thaliana are similar to rice in the sense that genetic alteration of HMA3 function primarily effects Cd accumulation in leaves. However, unlike rice we observe in A. thaliana that HMA3 also contributes to a lesser degree to leaf Zn accumulation. We would however caution against using such evidence to conclude that HMA3 in A. thaliana has higher specificity for Cd transport over other essential metals such as Zn. It is possible that the primary function of HMA3 in A. thaliana is in Zn transport, as has been proposed for HMA3 in A. halleri [22]. We observe that variation in HMA3 function is not reflected in a large variation in leaf Zn accumulation in A. thaliana and propose this could be due to other genes involved in Zn homeostasis (e.g. ZIPs, MTP1/3, HMA2/4) responding to maintain normal tissue Zn concentrations. Because Cd accumulation in A. thaliana is unlikely to be tightly regulated in the same way that Zn is, variation in HMA3 function is more clearly manifest in variation in leaf Cd accumulation. In a sense, variation in Cd accumulation is revealing hidden variation in Zn homeostasis mechanisms. However, further experiments are required to validate this model.
In previous studies in A. thaliana, HMA3 has been shown to function in the detoxification of Cd [19], [23], but its role in limiting Cd translocation to the shoot was not investigated. We determine genetically that HMA3 drives natural variation in leaf Cd concentration in A. thaliana, and grafting determined that HMA3 functions in the root to determine leaf Cd concentration. Further, the known root expression pattern of HMA3 is consistent with this observation. The expression pattern of HMA3 in different plant species may be very important in determining its roles in regulating leaf Cd accumulation. Similar to HMA3 in A. thaliana, rice HMA3 is also predominantly expressed in root. Since HMA3 functions in sequestering Cd into the vacuolar this expression pattern is consistent with HMA3 acting to reduce leaf Cd accumulation in both A. thaliana and rice. In contrast, the Cd/Zn hyperaccumulators N. caerulescens and A. halleri express HMA3 to extremely high levels in leaves where HMA3 is thought to enhance Cd sequestration into the vacuole, increasing its uptake [16], [22]. Consistent with this, constitutive over expression of a functional HMA3 in A. thaliana increases leaf Cd accumulation two-fold [19].
In conclusion, our data supports a model of HMA3 functioning in roots of A. thaliana to limit long-distance transport of Cd from root to shoot. We establish that the genetically determined natural variation in leaf Cd accumulation we observe in the A. thaliana global population is primarily controlled by variation of the function of HMA3 driven by DNA polymorphisms in the protein coding region of the gene. Further, we propose there are two polymorphic amino acid residues and a nonsense mutation distributed among 10 protein coding haplotypes that drive this population-wide variation in HMA3 function. These discoveries in A. thaliana improve our understanding of the mechanism of natural variation in Cd accumulation in plants. Further, they extend our knowledge of the function of HMA3 which could contribute to the engineering or breeding of low Cd accumulating crop plants.
The 349 A. thaliana accessions for the GWA study were selected from 5810 worldwide accessions as described previously [26], [48]. 82 of the genome re-sequenced accessions used in this paper were obtained from the Arabidopsis Biological Resource Center. Most plants used for elemental analysis by ICP-MS were grown in a controlled environment [26], [43]. Briefly, seeds were sown on moist soil (Promix; Premier Horticulture) with non essential elements (As, Cd, Co, Li, Ni and Se) added at subtoxic concentrations in a 20-row tray. After stratification at 4°C for 3 days the tray was moved into a climate-controlled room for growth with a photoperiod of 10 h light (90 µmol·m−2·s−1)/14 h dark, humidity of 60% and temperature ranging from 19 to 22°C. Plants were bottom-watered twice a week with modified 0.25× Hoagland solution in which Fe was replaced by 10 µM Fe-HBED (N,N′-di(2-hydroxybenzyl)ethylenediamine- N,N′-diacetic acid monohydrochloride hydrate; Strem Chemicals, Inc.). Plants used for studying the relationship between expression of HMA3 and leaf Cd concentration were grown in axenic conditions. Briefly, seeds were surface sterilized using 50% bleach and 0.05% SDS for 15 min, washed 8 times with sterilized deionized water and sown on ½ strength Murashige and Skoog (Sigma-Aldrich, St. Louis, USA) media solidified with agar containing 1% sucrose in Petri dishes. Plates were placed at 4°C for 3 days for seed stratification and then maintained at 16 h light (90–120 µmol·m−2·s−1)/8 h dark and 22°C. After 3-weeks growth, roots were harvested and used for RNA extraction and shoots were harvested for elemental analysis.
Seedlings were grafted as previously described [49]. Graft unions were examined before transfer to potting mix soil under the stereoscope to identify any adventitious root formation from graft unions or above. Healthy grafted plants were transferred to potting mix soil in a 20-row tray and grown in a controlled environment and after 4-weeks leaf samples were harvested as described above. After harvesting graft unions were examined again, and grafted plants with adventitious roots or without a clear graft union were removed from subsequent analysis.
The determination of leaf elemental concentrations was performed as described previously [43]. One to two leaves (∼2–4 mg dry weight) were harvested from A. thaliana plants grown vegetatively for 5 weeks, leaves were rinsed with 18 MΩ water and placed into Pyrex digestion tubes. Samples were placed into an oven at 92°C to dry for 20 hours. After cooling, 7 reference samples from each planted block were weighed. Subsequently, all samples were digested with 0.7 ml concentrated nitric acid (OmniTrace; VWR Scientific Products) and diluted to 6.0 ml with 18 MΩ water. Gallium (Ga) was added in the acid prior to digestion to serve as an internal standard for assessing errors in dilution, variations in sample introduction and plasma stability in the ICP-MS instrument. Analytical blanks and standard reference material (NIST SRM 1547) were digested together with plant samples in the same manner. After samples and controls were prepared, elemental analysis was performed with an ICP-MS (Elan DRCe; PerkinElmer) for Li, B, Na, Mg, P, K, Ca, Mn, Fe, Co, Ni, Cu, Zn, As, Se, Mo and Cd. All samples were normalized to calculated weights, as determined with a heuristic algorithm using the best-measured elements, the weights of the seven weighed samples and the solution concentrations, detailed at www.ionomicshub.org. For GWA analysis data was normalized using common genotypes across experimental blocks as previously described [26], and this normalized data has been deposited on the iHUB (previously known as PiiMS [51]) for viewing and download through www.ionomicshub.org.
The selection and genotyping of accessions for GWA analysis was described previously [26], [48]. Briefly, 5810 A. thaliana accessions were collected worldwide and genotyped at 149 genome-wide SNPs [26], [48]. These accessions were classified into 360 groups based on their genotypes at the 149 SNPs. One accession from each group was chosen to make a core set with 360 accessions. Among the core set of 360 accessions, 349 were phenotyped by ICP-MS for ionomic traits. Of this phenotyped subset 337 accessions were genotyped for at least 213,497 SNPs using the custom-designed SNP-tilling array Atsnptile 1 [26], [31], [48]. The GWA analysis was done using a linear mixed model to correct confounding by population structure [44] implemented in the program EMMA (Efficient Mixed-Model Association), which was described previously [31].
The SNP-Tilling array-based eXtreme Array Mapping (XAM) was done following the description of Becker et al. [47]. First, F2 progeny from an outcross of CS28181 and Col-0 were sorted by leaf Cd concentration. Approximately 25% of the total progeny at each end of the leaf Cd concentration distribution were pooled separately. From these pools approximately 300 ng genomic DNA was labeled separately using the BioPrime DNA labeling system (Invitrogen) and hybridized to the Affymetrix SNP-tilling array Atsnptile 1. The CEL files containing raw data of signal intensity for all probes were read and spatially corrected using R scripts from Borevitz et al. [52] with the R program and the Bioconductor Affymetrix package. The original CEL files used in this study have been submitted to the Gene Expression Omnibus (GEO) under accession GSE39679. Polymorphic SNPs between the two parents identified previously [52] were used for further analysis. There are 4 probes for each SNP, antisense and sense probes for two alleles. The allele frequency difference between the two pools for each SNP was then assessed based on the signal intensity difference of the 4 probes. The whole process can be carried out using R scripts that are available at http://ars.usda.gov/mwa/bsasnp [47].
PCR-based genotyping was used to further narrow down the mapping interval for the leaf Cd accumulation QTL. All 312 F2 plants that were phenotyped by ICP-MS were genotyped individually at 5 cleaved-amplified polymorphic sequence (CAPS) markers. The primers and restriction enzymes for the CAPS markers are listed in Table S2. Recombinants between marker Fo13M and Fo16M were selected for further analysis. The F2 recombinants with a clear low leaf Cd phenotype similar to CS28181 were directly used for determination of the candidate region. The F2 recombinants without a clear phenotype, or with a low Cd phenotype were selfed and 24 F3 progeny of each F2 individual further phenotyped for leaf Cd contetnt. According to the leaf Cd concentration of the F3's the genotype in the mapping interval was inferred and the region further narrowed.
The candidate genomic region of CS28181 was sequenced through overlapping PCR. Firstly, 20 overlapping fragments were amplified using KOD hot start DNA polymerase (Novagen, EMD Chemicals, San Diego, CA USA) from the genomic region of CS28181 covering HMA2 and HMA3 and their promoters. The primers for the PCR reactions were designed using Overlapping Primersets (http://pcrsuite.cse.ucsc.edu/Overlapping_Primers.html) and are listed in Table S2. After purification, each fragment was sequenced using its amplification primers in two directions. The sequenced reads were assembled using SeqMan Lasergene software (DNASTAR; http://www.dnastar.com), with Col-0 sequence used as the reference. The HMA3 haplotypes were analyzed using 149 genome re-sequenced A. thaliana accessions. Genomic sequence data of the 149 accessions was downloaded from the 1001 Genomes Data Center (http://1001genomes.org/data/MPI/MPICao2010/releases/2011_08_23/full_set/TAIR10, http://signal.salk.edu/atg1001/index.php,). The genomic sequences of the HMA3 region were extracted using Text File Splitter 2.0.4 (http://www.softpedia.com/get/System/File-Management/Text-File-Splitter.shtml). The sequence data was introduced into Microsoft Excel and polymorphic nucleotides identified. The coding sequence (CDS) of each HMA3 allele was predicted according to the reference cDNA of Col-0. Variations in protein amino acid sequence were identified according to the polymorphic nucleotides in the DNA sequence.
For construction of the expression vector of A. thaliana HMA3 and HMA2 two genomic DNA fragments for the two genes were PCR amplified from CS28181 using KOD hot start DNA polymerase and primers as listed in Table S2. The fragment for HMA3 is ∼4.9 kb including 1.6 kb promoter region and 0.8 kb 3′ downstream sequence. The fragment for AtHMA2 is ∼6.7 kb including 2.0 kb promoter region and 0.4 kb 3′ downstream sequence. The fragments were cloned into pCR-XL-TOPO vector (Invitrogen Life Technologies, http://www.invitrogen.com) for sequencing and subsequently recombined into binary vector pCAMBIA1301 by restriction enzymes of Sal I and BamH I. The expression vectors with the two genes were transformed into Agrobacterium tumeraciens strain GV3101 and were introduced into Col-0 using the floral dip method [53]. Transgenic lines were screened on ½ strength Murashige and Skoog (Sigma-Aldrich, St. Louis, USA) medium solidified with agar containing 50 µg/ml Hygromycin and 1% sucrose.
Total RNA was extracted from 3-week old plants grown on ½ strength Murashige and Skoog (Sigma-Aldrich, St. Louis, USA) medium solidified with agar containing 1% sucrose using TRIzol Plus RNA Purification kit (Invitrogen Life Technologies, http://www.invitrogen.com). Two microgram of total RNA was used to synthesize first strand cDNA with SuperScript VILO cDNA Synthesis Kit (Invitrogen Life Technologies, http://www.invitrogen.com). Quantitative real-time PCR was performed using SYBR Green PCR Master Mix (Applied Biosystems, USA) with the fist strand cDNA as a template on a Real-Time PCR System (ABI StepOnePlus, Applied Biosystems lco., USA). Primers for qRT-PCR were designed using Primer Express Software Version 3.0 (Applied Biosystems, USA). One primer of a pair was designed to cover an exon-exon junction. The primer sequences are shown in Table S2. Expression data analysis was performed as described previously [54].
|
10.1371/journal.pgen.1005750 | The Nuclear Matrix Protein Megator Regulates Stem Cell Asymmetric Division through the Mitotic Checkpoint Complex in Drosophila Testes | In adult Drosophila testis, asymmetric division of germline stem cells (GSCs) is specified by an oriented spindle and cortically localized adenomatous coli tumor suppressor homolog 2 (Apc2). However, the molecular mechanism underlying these events remains unclear. Here we identified Megator (Mtor), a nuclear matrix protein, which regulates GSC maintenance and asymmetric division through the spindle assembly checkpoint (SAC) complex. Loss of Mtor function results in Apc2 mis-localization, incorrect centrosome orientation, defective mitotic spindle formation, and abnormal chromosome segregation that lead to the eventual GSC loss. Expression of mitotic arrest-deficient-2 (Mad2) and monopolar spindle 1 (Mps1) of the SAC complex effectively rescued the GSC loss phenotype associated with loss of Mtor function. Collectively our results define a new role of the nuclear matrix-SAC axis in regulating stem cell maintenance and asymmetric division.
| Like many stem cells, the adult Drosophila male GSC often divides asymmetrically to produce one new stem cell and one gonialblast. The asymmetric division of GSC is specified by perpendicular orientation of the mitotic spindle to the hub-GSC interface and localization of Apc2. Here we show that Tpr/Mtor regulates GSC self-renewal and asymmetric division through the SAC complex. We found that Mtor cell-autonomously required in both GSCs and CySCs to regulate their self-renewal. Loss of Mtor function affects expression and localization of Apc2 and E-cadherin. We further found that Mtor is required for the correct centrosome orientation, mitotic spindle formation, and chromosome segregation. These defects are rescued by SAC complex components, Mps1 and Mad2. These data together suggest that Mtor regulates GSC asymmetric division and maintenance through the mitotic spindle checkpoint complex. We suggest that disruption of the Tpr-SAC pathway might lead to chromosome instability, chromosome lagging, and aneuploidy, stem cell division defects, and thereby tumor development.
| Germline stem cells (GSCs) from the Drosophila testis provide one of the best genetic systems to study stem cell regulation. At the tip of the Drosophila testis (apex) is a germinal proliferation center, which contains the germline and somatic stem cells that maintain spermatogenesis (Fig 1A) [1–5]. Each GSC is encysted by two somatic cyst stem cells (CySCs). Both GSCs and CySCs anchor to a group of 12 nondividing somatic cells, called the “hub”, through cell-adhesion molecules [6–9]. The hub defines the stem-cell niche by expressing the growth factor Unpaired (Upd), the ligand that activates the Janus kinase–signal transducer and activator of transcription (JAK-STAT) pathway in adjacent GSCs and CySCs to regulate their self-renewal [10,11]. In addition, hedgehog (Hh) [12–14], and bone morphogenetic protein (BMP) [9,15] signaling also play important role in GSC and CySC maintenance. During GSC division, the mother (old) centrosome remains anchored near the niche, while the daughter centrosome migrates to the opposite side of the cell, thereby assembling a mitotic spindle perpendicular to the hub [16–18]. In addition, the mother centriole nucleates more microtubules than the daughter centriole, which may help in asymmetric delivery of Apc2 to the cortex where GCSs contact the hub [18]. At the cortex, Apc2 and E-cadherin together anchor the spindles of mitotic GSCs perpendicular to the hub [19]. Therefore, only one daughter cell will contact the hub and receive JAK-STAT signaling to maintain stem cell identity, while the daughter cell at the other end of the mitotic spindle will experience a weaker signal and initiate differentiation. However, it is not known what molecular mechanism regulates asymmetric Apc2 localization at the niche-GSC interface.
To identify new regulators of stem cell asymmetric division in the Drosophila testis, we carried out a screen in which a collection of transgenic RNAi lines [20–22] were crossed with act-Gal4; tub-Gal80ts flies (referred to as Actts). The adult flies were shifted to the restrictive temperature (29°C-to inhibit Gal80 activity and induce Gal4-activity) from 18°C (Gal80 is active) and cultured at different time intervals. The flies were then dissected, stained, and examined for GSCs phenotype under confocal microscopy. One of the first few genes identified in this screen was Mtor. Mtor knockdown by transgenic RNAi resulted in a significant decrease of GSCs in the testes compared to the wild-type flies. Mtor belongs to a conserved family of coiled-coil proteins [translocated promoter region (Tpr) in vertebrates and myosin-like proteins 1 and 2 (Mlp1/2) in yeast] [23–25] that make up nuclear basket of the nuclear pore complex in vertebrates [26–29], and nuclear matrix in flies [27]. Tpr/Mtor plays an important role in regulating the SAC. The SAC delays anaphase until chromosomes are bioriented on the mitotic spindle. Recent studies demonstrated that Tpr/Mtor regulates the SAC by either controlling the kinetochore localization of Mad1 and Mad2 [26–27, 29–30] or by targeting Mad1 to the nuclear pore to direct mitotic checkpoint complex (MCC) assembly during interphase [28].
In this study, we found that loss of Mtor function affects GSC maintenance and asymmetric division. Knockdown Mtor results in Apc2 mis-localization, defects of mitotic spindle formation and chromosome segregation, and eventual GSC loss. Expression of Mad2 and Mps1 of the SAC complex effectively rescued the GSC loss phenotype associated with loss of Mtor function. Our results suggest that a nuclear matrix-SAC axis regulates GSC maintenance and asymmetric division through the Mtor-Mps1/Mad2 pathway.
As described above, we identified Mtor in a genetic screen for new regulators of stem cell fates in the Drosophila testis. Mtor knockdown by transgenic RNAi resulted in a significant decrease of GSCs in the testes compared to the wild-type testes. To further understand the function of Mtor in the germline or in the soma, we knocked down Mtor by using cell-type–specific Gal4s. We used three independently generated UAS-MtorRNAi transgenic fly lines. We found that depleting Mtor in the germ cell lineage using Nanos (Nos)-Gal4 resulted in a significant decrease in the number of GSCs associated with the hub compared to the wild-type control (Figs 1B–1F and S3A and S3B).
We also knocked down Mtor in the CySC lineage using the c587-Gal4 driver in combination with a temperature-sensitive Gal80 inhibitor [31]. Depleting Mtor in the CySC lineage (c587ts>MtorRNAi) caused defects in GSC differentiation (S1B and S1F and S1H Fig) similar to those seen in cyst cells defective in the epidermal growth factor receptor (EGFR) signaling pathway [32–36; S1C and S1D Fig]. We further examined CySC changes in c587ts>MtorRNAi flies using Traffic Jam (Tj) (a transcription factor expressed in CySCs, early cyst cells, and the hub cells) and Eyes absent (Eya) (a transcription factor expressed in early cyst cells) markers (S1E–S1H Fig). In comparison with those in the wild-type testes, the Tj-positive (S1 Fig, compare S1E and S1F Fig) and Eya-positive cells (S1 Fig, compare S1G and S1H Fig) were pushed away from the niche by the expanding undifferentiated germ cells in the testes of c587ts>MtorRNAi flies, suggesting that the Mtor-deficient CySCs have disadvantage in niche occupancy.
We further knocked down Mtor in the hub cells using unpaired (upd)-Gal4 (S2B Fig), in adult fly posterior midgut intestinal stem cells (ISCs) and progenitors using escargot (esg)-Gal4 in combination with a temperature-sensitive Gal80 inhibitor [31] (esgts>MtorRNAi) (S2D Fig), and in nine other types of cells using the cell-type–specific Gal4s (S2E Fig). Knockdowns of Mtor in the hub cells, ISCs and progenitors, and eight other cell types [lymph gland (plasmatocytes and crystal cells, collagen (Cg)-Gal4) [37], Drosophila insulin like peptide 2 (Dilp2)-Gal4 [38], leading edge (LE-Gal4) [39], fat body (pumpless (ppl)-Gal4) [40], hindgut (brachyenteron (byn)-Gal4) [41], hemocytes (serpenthemo (srphemo)-Gal4) [42], corpus allatum (Aug21-Gal4) [43], and hemocytes (hemolectin (hml)-Gal4) [44]], have no obvious phenotypes compared to those cells in the wild-type flies. Only Knockdowns of Mtor using wing-specific (MS1096)-Gal4 [45] (ms1096>MtorRNAi-2) resulted in wing phenotype (S2E Fig). Further, we found that knock down of Mtor in adult ISCs has no effect on spindle orientation and chromosome segregation (S6E and S6F Fig). These data suggest that Mtor functions specifically in GSC of Drosophila.
Using antibodies to Mtor [46], we detected Mtor in both the germline and soma in the wild-type (S3C Fig), but not in the Mtor-depleted testis (Actts>MtorRNAi) (S3D Fig), suggesting that the RNAi almost completely depleted Mtor protein expression. Further, we found that Mtor expresses at the nuclear pores, which is confirmed by co-staining the Mtor with nucleoporin 98 (NUP98) [32] (S3E and S3F Fig).
To further examine the function of Mtor in GSCs, we generated negatively marked GSC clones of wild-type or Mtork03905 [46] flies using the FLP/FRT mosaic analysis technique [47]. Testes with LacZ (arm-lacZ)-negative clones were counted 1, 2, and 7 days after clone induction (ACI). As expected, in FRT42D control testes, we were able to find many LacZ-negative GSCs and their differentiated progenies (Fig 2A–2B’ and 2E) at 1, 2, and 7 days ACI. At 1 day ACI, we were also able to find LacZ-negative GSCs and their differentiated progenies (Fig 2C, 2C’ and 2E) in FRT42D-Mtork03905 and FRTG13-Mtork03905 testes. At 2 days ACI, LacZ-negative Mtor homozygous mutant GSCs were recovered at negligible levels compared to control clones (Fig 2D, 2D’ and 2E). At 7 days ACI, we were unable to find a single LacZ-negative GSC or differentiated germ cell in Mtor-mutant testes (Fig 2E).
We examined cell death using anti-caspase 3 (Cas3) staining and found a significant increase in dead cells in the testes of FRT42D-Mtork03905 mosaic clones at 1 day ACI (S4B Fig) compared to wild-type control flies (Nos>LacZRNAi) (S4A Fig). However, the dead cells were outside the GSC zone. Consistent with this observation, coexpression of the pan-caspase inhibitor p35 did not suppress the GSC loss phenotypes in the Mtor-deficient testes (S4C and S4D Fig), indicating that the GSC loss in the Mtor-deficient testes maybe not through apoptosis.
We also generated GFP positively marked CySC clones of wild-type or Mtork03905 flies using the mosaic analysis with a repressible cell marker (MARCM) technique [48] (Fig 3). As expected, in FRT42D control testes, we were able to find many GFP-positive CySCs and their differentiated progenies (Fig 3A, 3A’, 3F, 3F’ and 3I) at 1, 2, and 4 days ACI. In FRT42D-Mtork03905 testes, GFP-positive Mtor homozygous mutant CySCs were recovered at negligible levels compared to the control clones (Fig 3B–3E’ and 3I) at 1 and 2 days ACI. At 4 days ACI, we were unable to find a single GFP-positive CySC (Fig 3G–3H’ and 3I). However, we could find many GFP-positive differentiated cyst cells (Fig 3D’, 3E’, 3G’ and 3H’) in Mtor-mutant testes at 2 and 4 days ACI. These results together suggest that Mtor is required for CySC self-renewal or attachment to the niche as suggested previously [32] in the c587ts>MtorRNAi flies (S1H Fig). Further, we did not detect any cell death in CySCs lineages in loss of Mtor testes.
During wild-type GSC division, the mother (old) centrosome remains anchored near the niche and Apc2 is localized to the interface of the niche and GSCs to anchor the spindles of mitotic GSCs perpendicular to the hub [18,19]. We examined Apc2 localization in wild-type or Mtor-depleted GSCs. After shifting the flies from 18ºC to 29ºC for 3 days, most of the Apc2 was enriched at the cortical region of GSCs adjacent to the hub-GSC interface in wild-type GSCs (Fig 4A, 4A’ and 4C), while a significant amount of Apc2 was distributed over GSC cortex from the hub-GSC interface in the GSCs remaining at the hub in Mtor-depleted (Nos>MtorRNAi) testes (Figs 4B, 4B’ and 4C and S4E and S4F). Similarly, we found that the expression and localization of E-cadherin at the hub-GSC interface were significantly reduced in Mtor-depleted testes in comparison with those in the wild-type testes (Fig 4D and 4E).
We further examined centrosome orientations in wild-type or Mtor-depleted GSCs. After shifting the Nos>MtorRNAi flies from 18°C to 29°C for 3 days, we found that in the Mtor-depleted testes, remaining GSCs at the hub had an increased frequency of misoriented or multiple centrosomes (Figs 5B, 5C and 5E–5G and S5B–S5D) as compared to GSCs in the wild-type testes (Figs 5A, 5D and 5G and S5A and S6A).
Besides the Apc2 localization and centrosome orientation defects, we also observed severe microtubule spindle and chromosome segregation defects in the Mtor-depleted GSCs. After shifting the Nos>MtorRNAi flies from 18°C to 29°C for 3 days, we found that many phospho-Histone H3 (pH3)-positive GSCs exhibited lagging and scattered chromosome phenotypes (S5B–S5D Fig). The phenotypes became worse after shifting the Nos>MtorRNAi flies from 18°C to 29°C for 7 days (Figs 5C and 5E and S5F–S5H and S6B–S6E). Most of the pH3-positive GSCs were detached from the hub (Figs 5C and S5F–S5H and S6B–S6E). The microtubule spindles formed were incomplete, unfocused, only half, and/or without clear spindle poles (Figs 5B, 5C and 5E–5H and S5F–S5H and S6B–S6E). At anaphase, some spindles remained, bridging the separated chromosomes (S6B Fig), and the chromosomes were lagging and scattered (Figs 5E and 5H and S5F–S5H and S6E).
These observations suggest that the function of Mtor is cell-autonomously required for the correct centrosome orientation, mitotic spindle formation, chromosome segregation, and localization of Apc2 to the hub-GSC interface.
In the Drosophila S2 and several human cell lines, Tpr/Mtor plays an important role in regulating the spindle assembly checkpoint (SAC) [26–29]. During interphase, the Tpr/Mtor-Mad1-Mad2 complex regulates generation of a premitotic anaphase inhibitor to protect genome integrity [27–29]. In Mtor-depleted Drosophila S2 cells, accumulation of Mad2 and Mps1 at kinetochores is significantly reduced and the cells enter anaphase prematurely [27]. Therefore, Mtor may regulate GSC maintenance and asymmetric division through SAC. To test this possibility, we expressed UAS-Mtor, UAS-mad2, UAS-Apc2, and UAS-E-cad with the UAS-MtorRNAi lines using the Nos-Gal4 driver. Expression of either UAS-Mtor or UAS-mad2 significantly rescued the stem cell loss phenotypes of the UAS-MtorRNAi lines (Fig 6, compared panel A to B and C), while expression of UAS-Apc2 or UAS-E-cad had no function (Fig 6D, 6E and 6G). Further, we also found that expression of UAS-mad2 significantly rescued the reduction of Apc2 localization phenotypes of the UAS-MtorRNAi line (Fig 6F). Overexpression of Mtor (Nos>Mtor), mad2 (Nos>mad2), Apc2 (Nos>Apc2) and E-cad (Nos>E-cad) alone resulted in no abnormal phenotype (S6H–S6K Fig).
We further expressed Mtor, mad2, mps1, and Apc2 in Mtor-mutant GSC mosaic clones (Fig 7A). Expression of Mtor, mad2, and mps1 significantly rescued the GSC loss phenotypes of the Mtor-mutant GSC clones, while expression of Apc2 had no significant function. These data suggest that the correct localization of Apc2 and E-cad regulated by the Mtor-SAC axis through mitotic spindle rather than the relative amounts of these proteins are important for GSC maintenance.
These above data together suggest that Mtor/Tpr regulates GSC asymmetric division and maintenance through the mitotic spindle checkpoint complex (Mps1 and Mad2).
The asymmetric division of CySCs occurs through a cellular mechanism strikingly distinct from the one used by GSCs. The mitotic spindle of CySCs first forms in a random location and then repositions during or near the onset of anaphase so that one pole is close to the hub cells [49]. Our above data demonstrated that depletion of Mtor in GSCs resulted in GSC loss, while depletion of Mtor in CySCs resulted in differentiation of CySCs, indicating that Mtor may function differently in GSCs and CySCs. To find out whether Mtor regulates CySCs through the SAC complex, we expressed UAS-mad2 and UAS-mps1 in the UAS-MtorRNAi lines using the c587-Gal4 driver. Expression of UAS-mad2 and UAS-mps1 did not significantly rescue the phenotypes associated with expressing UAS-MtorRNAi in somatic cyst cells (they show GSC tumor phenotype; S6L and S6M Fig), suggesting that Mtor functions in GSCs and CySCs through distinct molecular pathways.
In male Drosophila GSCs, the asymmetric outcome of stem cell division is specified by an oriented spindle and cortically localized Apc2. However, the molecular mechanism that regulates asymmetric Apc2 localization and formation of the oriented spindle is unclear. In this study, we identified a nucleoporin and spindle matrix protein Tpr/Mtor that regulates GSC asymmetric division and maintenance. Loss of Mtor function results in abnormal Apc2 localization, incorrect centrosome orientation, defective mitotic spindle formation, and abnormal chromosome segregation. We further demonstrated that Mtor regulates GSC asymmetric division and maintenance through the SAC, which regulates asymmetric localization of Apc2 and E-cad. At the cortex, Apc2 and E-cad together anchor the spindles of mitotic GSCs perpendicular to the hub for asymmetric GSC division [19]. Defects in the Mtor-regulated processes may first block cytokinesis, result in polyploidy, and cause eventual loss of GSCs. We do not know how SAC/Mad2 affects APC2 localization. More experiments are needed to find the detailed molecular mechanism. However, in yeast SAC/Mad2 regulates Kar9 (the APC2 homologue) localization through the mitotic exit network (MEN) and Kip2 (kinesin-like protein) [50]. SAC/Mad2 may regulate APC2 localization through a similar mechanism in Drosophila male GSCs.
In Drosophila S2 cell, Mtor forms a nuclear complex with Mad1/2 in interphase; after nuclear envelope breakdown (NEB), Mad2 is recruited to unattached kinetochores and functions in the SAC complex while Mtor reorganizes into a fusiform structure coalescent with spindle microtubules and plays a role in spindle elongation [27]. In Mtor-knockdown GSCs, we found that many pH3-positive GSCs exhibited lagging and scattered chromosome phenotypes, were detached from the hub; the microtubule spindles formed were incomplete, unfocused, only half, and/or without clear spindle poles. These phenotypes cannot be entirely explained by SAC defects alone. Consistent with this, expression of Mad2 in Mtor GSC mosaic clones could only partially rescue the GSC loss phenotype (Fig 7A). These results suggest that there could be additional factors that together with Mad2 and Mps1 mediate Mtor’s function in GSCs. Further, expression of Mad2 and Mps1 did not rescue Mtor-mutant phenotypes in CySCs, suggesting that Mtor functions differentially in GSCs and CySCs. The yeast homologue of Tpr, Mlp2p, binds to the yeast spindle pole body (SPB) and promotes its efficient assembly [51]. Most recently, it has been shown that Mtor in Drosophila directly binds a tau-tubulin kinase, Asator, and colocalizes to the spindle region with Asator during mitosis [52]. Asator may represent a link between Mtor and the microtubule-based spindle apparatus that facilitates Mtor’s function in regulating microtubule dynamics and microtubule spindle function. Therefore, besides Mps1 and Mad2, Mtor may also regulate mitotic spindle and Apc2 localization through Asator-CLIP190-EB1 and/or the centrosome (Fig 7B).
In the Drosophila testis, the two types of stem cells, GSCs and CySCs, use distinct molecular mechanisms for their asymmetric division. The mitotic spindle of CySCs first forms in a random location within an irregularly shaped CySC, then repositions through functional centrosomes, dynein, and the actin-membrane linker moesin during or near the onset of anaphase so that one pole is close to the hub cells [49]. Therefore, CySCs require moesin, but not Apc2, and GSCs require Apc2, but not moesin, for their orientation.
We demonstrated that expression of Mad2 and Mps1 could rescue Mtor-mutant phenotypes in GSCs but not in CySCs, suggesting that the Tpr/Mtor-SAC pathway regulates asymmetric Apc2 localization for asymmetric GSC division, but not asymmetric moesin localization for asymmetric CySC division. In c587ts>MtorRNAi flies, the Tj-positive and Eya-positive cells were pushed away from the niche by the expanding undifferentiated germ cells, suggesting that the Mtor-deficient CySCs have disadvantage in niche occupancy. Consistent with the MtorRNAi result, the Mtor mutant CySCs generated by the MARCM technique were quickly moved out of the niche and became differentiated cyst cells (Fig 3). These results together suggest that the main function of Mtor in CySCs is to regulate their attachment to the niche.
Dividing eukaryotic cells have to establish correct bipolar attachment of a pair of sister kinetochores residing on each mitotic chromosome to the mitotic spindle for delivering duplicated chromosomes to separate daughter cells. This process is regulated by components of the SAC. Defects in the SAC will result in improper segregation, aneuploidy, and chromosome lagging, which are linked to birth defects and cancer in animals and humans. Aneuploidy has been recognized as a major driver of cancer [53]. Our study results have demonstrated that the Tpr/Mtor through SAC regulates Apc2 localization and asymmetric GSC division. Disruption of Mtor function resulted in defective mitotic spindle and abnormal chromosome segregation. Tpr was fused together with oncogenes met, raf, and kit in several kinds of tumors [54–56]. It is reasonable to propose that disruption of the Tpr-SAC pathway in these tumors might lead to chromosome instability, chromosome lagging, and aneuploidy, stem cell division defects, and thereby tumor development.
Oregon R or UAS-lacZRNAi was used as the wild type. Mtork03905 was previously described [46] and obtained from the Bloomington stock center. The P-element insertion in the Mtork03905 resulted in a 9-base pair duplication, including 8 base pairs of upstream genomic sequence and a duplicated +1 residue and may represent a null mutation.
To generate transgenic strains of UAS-mad2, UAS-mps1, and UAS-Mtor, the corresponding full-length cDNAs of the Drosophila genes were amplified by PCR and inserted into the pUAST transformation vector [57]. Second-chromosome UAS-Apc2 transgenic flies were from David Roberts. UAS-DEFL (full-length shg) #6–3 was obtained from Kyoto stock center. UAS-EgfrDN was obtained from the Bloomington stock center (BL5364).
All constructs were confirmed by DNA sequencing. The UAS constructs were injected into w1118 embryos using standard procedures.
RNAi stocks used in this study: MtorRNAi-1 (Vienna Drosophila RNAi Center-VDRC) Transformant ID 110218 [v110218]), MtorRNAi-2 (BL32941), MtorRNAi-3 (v24265), and rlRNAi (v35641). The sequences used for VDRC knockdown strains are available for each line at https://stockcenter.vdrc.at and sequences for Bloomington knock-down strains are available for each line at http://flystocks.bio.indiana.edu.
The following Gal4 alleles were used to drive UAS lines: Nos-Gal4 (nanos-Gal4VP16) [58], obtained from the Bloomington stock center (BL4937), and upd-Gal4 and c587-Gal4, provided by Ting Xie. esg-Gal4 was obtained from Shigeo Hayashi. Collagen (Cg)-Gal4 [37], Dilp2-Gal4 [38], LE-Gal4 [39], pumpless (ppl)-Gal4 [40], brachyenteron (byn)-Gal4 [41], serpenthemo (srphemo)-Gal4 [42], Aug21-Gal4 [43], hemolectin (hml)-Gal4 [44], and MS1096-Gal4 [45].
Flies were raised on standard fly food at 25°C and at 65% humidity, unless otherwise indicated.
Clones of mutant GSCs were generated as previously described [6]. To generate Mtor-mutant GSC clones, FRT42D+ and FRT42DMtork03905/Cyo virgin females were mated with males of genotype FRT42Darm-lacZ/Cyo; MKRS, hs-flp/+, or FRTG13Mtork03905/Cyo virgin females were mated with males of genotype FRTG13arm-lacZ/Cyo; MKRS, hs-flp/+. One- or 2-day-old adult males carrying an arm-lacZ transgene in trans to the mutant-bearing chromosome were heat shocked four times at 37°C for 1 hr, at intervals of 8–12 hr. The males were transferred to fresh food every day at 25°C. The testes were removed 1, 2, or 7 days after the first heat-shock treatment and processed for antibody staining.
To induce MARCM clones of FRT42D-piM (as a wild-type control) and FRT42D-Mtork03905, we generated the following flies: FRT42D tub-Gal80/FRT42D Mtork03905 (or piM); MKRS, hs-flp/ tub-Gal4,UAS-mCD8.GFP. Three- or 4-day-old adult male flies were heat-shocked twice at 37°C for 45 min, with an interval of 8–12 hr. The flies were transferred to fresh food daily after the final heat shock. The testes were removed at 1, 2, or 7 days after the first heat-shock treatment and processed for antibody staining.
Male UAS-RNAi transgene flies were crossed with female virgins of genotype Nos-Gal4, upd-Gal4, c587-Gal4; tub-Gal80ts (c587ts), or esg-Gal4, UAS-GFP; tub-Gal80ts (esgts). The flies were cultured at 18°C (Gal4 is inactive and Gal80 is active). Three- to 5-day-old adult flies with the appropriate genotype were transferred to new vials at 29°C (Gal4 is active and Gal80 is inactive) for 3 or 7 days before dissection.
Normal immunofluorescence staining was performed as described previously with some modifications [6]. Briefly, testes were dissected in phosphate-buffered saline (PBS), transferred to 4% formaldehyde in PBS, and fixed for 30 minutes. The testes were then washed in PBST (PBS containing 0.1% Triton X-100) for 3 times, 10 Minutes each time, then blocked with 5% goat serum in PBST for 1 hour. Samples were the incubated with primary antibody in PBST at 4°C overnight. Samples were washed for 30 minutes (three 10-minute washes) in PBST, incubated with secondary antibody in PBST at room temperature for 2 hours, washed as above, and mounted in VECTASHIELD with DAPI (Vector Labs).
For the γ-tubulin staining, testes were dissected in PBS, transferred to 4% formaldehyde in PBS, and fixed for 20 minutes, followed by incubation with methanol for 10 minutes. Then washed for 10 minutes with PBST, and two 10-minutes washes with 5% goat serum PBST. Then incubated in primary antibody in 5% goat serum in PBST overnight at 4°C. Then washed three times, 15 min each in PBST, followed with 2 hrs incubation with secondary antibody in 5% goat serum in PBST. Then washed for at least an hour with PBST, and mounted as above.
Caspase-3 activity was assessed using Live Green Caspase Detection Kits (I35106, Molecular Probes) according to standard protocol.
Confocal images were obtained by using a Zeiss LSM510 system, and were processed with Adobe Photoshop 7.0. GSCs were scored as Vasa-positive cells adjacent to the hub (detected using Fas3) and containing dot spectrosome (detected using 1B1). Only image with a clear view of the complete hub were used.
The following antisera were used: rabbit polyclonal anti-Vasa antibody (1:5000; gift from R. Lehmann), rabbit polyclonal anti-β-Gal antibody (1:1000; Cappel), mouse monoclonal anti-β-Gal antibody (1:100; Invitrogen), mouse monoclonal anti-Hts antibody 1B1 (1:4; Developmental Studies Hybridoma Bank [DSHB]), mouse monoclonal anti-Fas 3 antibody (1:10; DSHB), rat polyclonal anti-Tj (1:400; Li et al. [59], mouse monoclonal anti-Eya (1:20, DSHB), mouse monoclonal anti-Dl (1:20; DSHB), mouse monoclonal anti-Pros (1:50; DSHB), rabbit polyclonal anti-GFP antibody (1:200; Molecular Probes), mouse monoclonal anti-GFP antibody (1:100; Invitrogen), rabbit polyclonal anti-Caspase 3 antibody (1:1000; gift from B. Hay); rabbit polyclonal anti-Thr3-phosphorylated histone H3 antibody (1:200; upstate), mouse monoclonal anti-α-tubulin antibody (1:100; Sigma), mouse monoclonal anti-γ-tubulin antibody (1:100; Sigma), guinea pig polyclonal anti-Zfh1 (1:2000; gift from J. Skeath), mouse monoclonal anti-Mtor antibody (1:100; gift from K. Johansen), rabbit polyclonal anti-Apc2 antibody (1:5000; gift from M. Bienz). Secondary antibodies were goat anti-mouse, goat anti-guinea pig, and goat anti-rabbit IgG conjugated to Alexa 488 or Alexa 568 (1:400; Molecular Probes). DAPI (Molecular Probes) was used to stain DNA.
We scored the centrosome misorientation and spindle misorientation following the protocol described by Yamashita et al. [19, 60]. Specifically, centrosome misorientation was noted when neither of two centrosomes were closely associated with hub-GSC interface during interphase and at mitosis. Spindle misorientation was scored when neither of the two spindle poles was closely associated with hub-GSC interface during mitosis [19, 60].
Statistical analyses were performed using Microsoft Excel 2010 or GraphPad Prism 6 software. Data are shown as means ± SD or standard error of the mean (SEM). P-values were obtained between two groups using the Student’s t-test or between more than two groups by analysis of variance (ANOVA).
|
10.1371/journal.pcbi.1000710 | Interplay between Pleiotropy and Secondary Selection Determines Rise and Fall of Mutators in Stress Response | Mutators are clones whose mutation rate is about two to three orders of magnitude higher than the rate of wild-type clones and their roles in adaptive evolution of asexual populations have been controversial. Here we address this problem by using an ab initio microscopic model of living cells, which combines population genetics with a physically realistic presentation of protein stability and protein-protein interactions. The genome of model organisms encodes replication controlling genes (RCGs) and genes modeling the mismatch repair (MMR) complexes. The genotype-phenotype relationship posits that the replication rate of an organism is proportional to protein copy numbers of RCGs in their functional form and there is a production cost penalty for protein overexpression. The mutation rate depends linearly on the concentration of homodimers of MMR proteins. By simulating multiple runs of evolution of populations under various environmental stresses—stationary phase, starvation or temperature-jump—we find that adaptation most often occurs through transient fixation of a mutator phenotype, regardless of the nature of stress. By contrast, the fixation mechanism does depend on the nature of stress. In temperature jump stress, mutators take over the population due to loss of stability of MMR complexes. In contrast, in starvation and stationary phase stresses, a small number of mutators are supplied to the population via epigenetic stochastic noise in production of MMR proteins (a pleiotropic effect), and their net supply is higher due to reduced genetic drift in slowly growing populations under stressful environments. Subsequently, mutators in stationary phase or starvation hitchhike to fixation with a beneficial mutation in the RCGs, (second order selection) and finally a mutation stabilizing the MMR complex arrives, returning the population to a non-mutator phenotype. Our results provide microscopic insights into the rise and fall of mutators in adapting finite asexual populations.
| The dramatic rise of mutators has been found to accompany adaptation of bacteria in response to many kinds of stress. Two views on the evolutionary origin of this phenomenon emerged: the pleiotropic hypothesis positing that it is a byproduct of environmental stress or other specific stress response mechanisms and the second order selection which states that mutators hitchhike to fixation with unrelated beneficial alleles. Conventional population genetics models could not fully resolve this controversy because they are based on certain assumptions about fitness landscape. Here we address this problem using a microscopic multiscale model, which couples physically realistic molecular descriptions of proteins and their interactions with population genetics of carrier organisms without assuming any a priori mutational effect on fitness landscape. We found that both pleiotropy and second order selection play a crucial role at different stages of adaptation: the supply of mutators is provided through destabilization of error correction complexes or, alternatively, fluctuations of production levels of prototypic mismatch repair proteins (pleiotropic effects), while the rise and fixation of mutators occurs when there is a sufficient supply of beneficial mutations in replication-controlling genes. This general mechanism assures a robust and reliable adaptation of organisms to unforeseen challenges. This study highlights physical principles underlying biological mechanisms of stress response and adaptation.
| Bacterial populations often respond to various stresses by inducing mutagenesis whereby mutator clones rise to fixation, at least transiently, during adaptation to stressful environments [1]–[5]. The rise of mutator clones has been observed as a universal response regardless of the nature of stress, despite the diversity of detailed molecular mechanisms associated with such responses (reviewed in [1],[5]). (See, however, [6] where this interpretation is questioned for a particular experimental system.) The evolutionary significance of this observation has been controversial, and two distinct views emerged in the literature [3],[7]. The pleiotropic hypothesis posits that high mutation rate is a by-product of genetic mechanisms invoked in response to stress or other physical mechanisms unrelated to adaptation [8]. The key aspect of the pleiotropic hypothesis is that high levels of error correction and maintenance may be energetically costly so that bacteria would not fully activate them in stable environments [3]. Consistent with that view is the observation that natural populations exhibit a broad range of mutator allele frequencies, which are relatively higher than expected [2],[9]. Higher mutation rates during adaptation may be then due to the trade-off between the requirement to repair diverse lesions in genomes and the energetic cost maintaining the fidelity of DNA polymerases involved in this process.
An alternative view is a second order selection hypothesis [10]–[12]. Mutators, which can rapidly produce beneficial mutations, could get fixed in the population by hitchhiking [12]. However, they mostly burden the population with deleterious mutations, which eventually outnumber the beneficial ones, and thus mutation rate tends to decrease to a minimum in well-adapted populations [10], [13]–[15]. Computer simulations employing population genetics models provided some evidence that mutators can hitchhike to fixation when population size is large enough and stress is sufficiently profound and durable [2],[15],[16]. However, these theoretical studies were based on a number of phenomenological assumptions. In particular, alleles were classified into a few discrete forms such as “deleterious”, “normal” and “beneficial” and fitness effects were assumed additive between alleles. Furthermore, most population genetics models are based on certain a priori assumptions about the appearance and reversion of mutations. Implicit in these models is a peculiar effect of saturation whereby all or most alleles get fixed in their beneficial forms, essentially eliminating further supply of beneficial mutations, which causes populations to reverse to a non-mutator phenotype. However, while many postulates of mathematical population genetics are rooted in experimental observations, the reality is certainly much more complex. In particular, in complex crowded cellular environments, mutations in coding regions are more likely to have a broad impact on many properties of cellular proteins such as their stability, interactions with their functional and non-functional partners and of course their catalytic activity, which results in a continuous effect of mutations on fitness. Furthermore, the effect of fitness on supply of beneficial (as well as deleterious) mutations is hard to evaluate a priori due to enormous plasticity and size of sequence space of functional proteins [17],[18]. To this end, it is very important to go beyond the phenomenological postulates of traditional population genetics models and develop a new model where population genetics is coupled to a realistic yet tractable biophysical model of proteins and their interactions in the cytoplasm. A first step in that direction has been made in [13] where we studied evolution of mutation rates in a population of simple organisms each carrying 3 genes. The key distinctive feature of the approach proposed in [13] is that properties of cellular proteins – their stability and interactions – were derived directly from sequences of their genomes and a simple biologically realistic relationship connected these biophysical properties with fitness (growth rate) of the model cell population.
Here, we further develop this microscopic multiscale approach to study evolutionary dynamics of stress-induced adaptation in a finite asexual population. In particular we focus on emergence (or lack thereof) of mutators during the adaptation process. In the present model, each organism carries four genes expressing corresponding protein products. The first three genes are housekeeping genes responsible for cell growth and division, (replication controlling genes or RCGs), and protein products of gene 4 homodimerize to form a mismatch repair (MMR) complex – mimicking the mutS system in bacteria whose proteins are active in vivo as tetramers (dimers of dimers) [19],[20]. While diverse molecular mechanisms can lead to stress-induced mutagenesis in bacteria (e.g. rpoS dependent SOS responses [21]), here we focus on a prototypical MMR system, for simplicity. The deficiencies and down regulation of MMR genes are known in many instances to be the main cause of constitutive mutators, which are constantly supplied to the population regardless of environmental requirements, [2],[9],[22] as well as a major molecular event in stress-induced mutagenesis [2],[23],[24]. Three RCGs form the simplest functional protein-protein interaction (PPI) network where protein 1 functions in isolation and proteins 2 and 3 must form a functional heterodimeric complex. The model with three RCGs was used in our recent study [13] where it was shown that this minimal model, which takes into account protein function (in the form of PPI), is capable of reproducing the rich biology of mutation rate evolution. Fitness, i.e. the growth rate, b of an organism is proportional to the monomer concentration of the protein product of the first RCG and the concentration of functional dimers of protein products of the second and the third RCGs:(1)where is a base growth rate, is the concentration of monomeric protein i and is the concentration of heterodimer complex between protein i and j in all possible binding configurations. is the Boltzmann probability of binding between protein 2 and 3 in the native, functional binding configuration whose binding energy has the lowest value of all possible mutual configurations. is thermal stability (Boltzmann probability to be in the native state) for the protein product of gene . is intracellular concentration for protein i, and is an optimal total concentration for all proteins in a cell. Deviation from this optimal level causes a drop in fitness, reflecting a metabolic cost of protein production and degradation, and is a control coefficient, which sets fitness penalties for deviations from the optimum production level. The importance of fitness cost for protein overproduction has been established by Dekel and Alon [25]. Phenomenologically, the overexpression cost function in the denominator of Eq.(1) prevents an artificial scenario when the increase of fitness is achieved by merely overexpressing proteins rather than by evolving their sequences.
The protein product of the fourth gene determines the mutation rate of its genome by acting as a prototype of mutS, which forms dimers of dimers. The fidelity of an organism's DNA replication is proportional to the concentration of functional MMR homodimers formed by products of gene 4 (see Model for details). Protein concentrations are epigenetically heritable but can fluctuate; reflecting long-time correlated noise in protein production in living cells [26]. and are exactly calculated for a given set of by solving equations of the Law of Mass Action (LMA) (see Model for details). Thus, mutation rates can increase upon a drop in concentration of functional MMR homodimers, or upon mutations of the MMR gene that disfavor its functional homodimerization, or both. (See Figure 1 and Model below for illustration and details.)
Using this ab initio model we study adaptation to various stresses such as higher temperature, stationary phase and starvation. In particular we focus on the importance, universality and causes of transient fixation of the mutator phenotype in adapting finite asexual populations.
Each evolutionary simulation started from a population of 500 organisms each having the same seed genome. The population size was limited at 5000 organisms so that excess organisms were randomly culled when this size limit was reached. Seed protein sequences were optimized to have sufficiently high initial stability () to avoid an immediate lethal phenotype (see Model for details). However neither protein sequences nor their concentrations C's were optimized to achieve beneficial protein-protein interactions. Correspondingly initial fitness of the seed populations was quite low and the initial adaptation increased fitness through optimization of expression levels and protein-protein interactions (see below). Then at a later time (at t = 20000) we subjected adapted populations to stress. We modeled three types of stress. The first type was “heat shock” whereby we instantly raised temperature from T = 0.85 to T = 1.0 and kept it fixed afterwards. The second type of stress mimicked entrance into stationary phase whereby we instantly dropped growth rate of all organisms threefold (i.e. decreased in Eq.(1) threefold). The third type of stress simulated “starvation” accompanied by a sharp drop in protein production. To this end we instantly dropped the optimal protein production level (see Eq.(1)) tenfold at t = 20,000. For each type of stress we ran 100 simulations to obtain statistically significant results.
Figure 2 shows evolution of the populations. The first key event is an initial adaptation of seed sequences, which resulted in dramatic improvement of fitness. At this stage initial seed sequences evolve into adapted organisms where functional and non-functional PPI are optimized (see below). Three broad classes of populations (strains) distinguished by their fitness () emerged after initial adaptation (see Table 1 for detailed distributions), suggestive of a highly non-trivial fitness landscape in the model, containing at least three local fitness peaks. In all cases the initial adaptation occurred via transient fixation of the mutator phenotype as can be seen in the bottom panel of Figure 2. Second, transient fixation of mutators took place in most cases except entrance into stationary phase in highly fit strains (Figure 2B bottom panel), which eventually did not increase fitness upon adaptation after starvation stress. For heat-shock stress the highly fit population went briefly to transient fixation of the mutator phenotype but quickly eliminated it (Figure 2A bottom panel). It is also interesting to note that populations of higher fitness carried a greater fraction of constitutive mutators (before stress but after initial adaptation) but after stress this relation was reversed. This is similar to experimental observation of Matic and coworkers that mutation in aging colonies is anticorrelated with the fraction of constitutive mutators [2]. The starvation stress resulted in a dramatic drop of fitness for all three strains. Correspondingly all three strains transiently fixed mutator phenotypes upon adaptation to new conditions.
Our model provides a unique opportunity to get detailed insight into possible mechanisms, which lead to the rise, fixation and fall of mutators. The dynamics of microscopic variables such as protein concentrations , the stability of MMR proteins , and Boltzmann probability to form functional MMR complexes are shown in Figure 3 for the same three fitness classes (strains) of evolved populations (same color code as in Figure 2). These data provide insights into molecular mechanisms underlying the emergence, fixation and disappearance of mutator clones. Two factors are potentially responsible for the emergence of mutators: epigenetic stochastic switching through fluctuation of concentrations and mutations changing the stability of the MMR protein or interactions between them in a functional homodimeric complex . The initial set of quickly converged to a more optimal distribution by reallocating resources for better fitness: The total concentration of replication-controlling proteins (, , and ) increased, while concentrations of the MMR proteins, () decreased. Similar parallel changes in gene expression pattern were also reported in long-term evolutionary experiments [12],[23],[27].
Change in concentration of the MMR protein, due to stochastic fluctuations, was the primary factor causing the rise of mutators in initial adaptation. As for adaptation to stress which took place at a later time t = 20,000, fluctuation in MMR protein production level was primarily responsible for the rise of the mutator phenotype at stationary phase and starvation stresses (see Figure 3B and Table 1), except for the strain in stationary phase stress which reached high fitness b∼1 in initial adaptation. For this strain no further adaptation took place after stress, and a mutator phenotype did not fix. For heat-shock stress, destabilization of the MMR protein and its homodimeric complex at higher temperature was the primary cause of the rise of a mutator phenotype (Figure 3A). The recovery of a normal, non-mutator phenotype was mostly due to mutations in the MMR gene, which increased stability of the functional MMR complex. In order to determine precisely the microscopic causes of phenotypic switches between mutators and non-mutators, we traced all transitions between them for all adaptation events on all trajectories. The summary picture is presented in Figure 4. Green lines in all panels of Figure 4 highlight the instances when the mutator phenotype was switched on or off by variation of concentration of MMR protein . Most mutators in the bottom panel of Figure 4, except in the temperature jump case, initially emerged from stochastic variation of protein concentrations, i.e. they represented switches due to epigenetic events. The transitions between fixation of mutators and non-mutators mostly occurred in a specific microscopic order, depending on the nature of stress (see Figure 4 and Table 1). The heat-shock stress resulted in thermal destabilization of the MMR complex, which gave rise to higher mutation rates. On the other hand, the stationary phase and starvation stresses decreased the growth rate, which prevented the constitutive mutators from being purged away from their finite populations by genetic drift. Sequentially, highly mutating strains in all cases discovered mutations, which stabilized functional interactions in RCGs providing strains of higher fitness, so that mutator strains hitchhiked to fixation in stationary phase and starvation stresses. Finally a mutation in the MMR protein stabilized the complex bringing mutation rates in the population back to the original low level. On a microscopic level, the behavior of generating mutator strains in response to temperature stress is somewhat different from the behavior to stationary phase and starvation stresses. In the former case, stress induces the mutator strain directly by disrupting the MMR complex. Meanwhile, in the latter case it does not induce mutator strains per se but sets in motion a chain of microscopic and populational events, such as hitchhiking, which result in a similar phenotypic phenomenology as adaptation to a temperature jump.
Why did mutators preferentially emerge through epigenetic stochastic switching rather than a genotypic change (mutation)? To address this question we studied adaptation in response to the stress of stationary phase at various rates r of stochastic fluctuation of protein concentrations, from r = 10−2 to 10−3, 10−4, and r = 0 – the case where no fluctuations of protein concentration were allowed (Figure 5; see Model and Table 2 for definition of fluctuation rates r). To correctly compare simulations in four different conditions with one another, we assigned unequal concentrations to the RCG proteins and MMR protein, setting them similar to those reached after the first adaptation event. Otherwise the inability to relax an imbalance among equally fixed protein concentrations at the control of might constraint the evolution of fitness. Deceleration of fluctuation rate delayed fixation of mutators, and furthermore, no mutators (and, strikingly, adaptation) were observed when r = 0.
The upper panel of Figure 5 points to a peculiar feature: while populations with highest rate r of protein copy number fluctuation evolved to highest fitness b, their initial population growth was not the highest. In order to resolve this apparent contradiction we carried out a simulation where species with a high rate of protein concentration fluctuation r = 0.01 competed with the ones with no protein concentration fluctuation (Figure 6). For the first 500 time steps, the fractional population of highly fluctuating r = 0.01 species decreased. The species with a high fluctuation rate provided more mutators due to epigenetic stochastic switching and their high mutation rate effectively reduced the growth rate of the population due to the heavy genetic load of deleterious mutations. Fitness curves shown in Figure 6 (red and blue curves) indicate that the initial drop in fractional population of the r = 0.01 species (black curve) was not caused by the difference in growth rates between two competing species. The initial decrease of fractional population of the r = 0.01 species is reversed after it found a beneficial mutation in RCGs which provided higher fitness at t∼500, and the population with r = 0.01 started to dominate in the competition. We conclude that the genetic load of high mutation rate initially burdens the population with r = 0.01, which is enriched in mutators and its growth curve is effectively limited until it finds a beneficial mutation.
Visser et al. showed that the initial level of fitness of the founder population dramatically affects the rate of adaptive evolution: the rate of adaptation was much slower for populations founded by an adapted strain than for the populations founded by an initially unadapted strain [28]. This finding is in direct agreement with our results. Figure 2 shows that populations that achieved high fitness (b = 1, blue lines) did not further evolve after stationary phase stress and only briefly fixed mutators upon heat shock with no significant adaptation afterwards, while in starvation stress where fitness dropped more significantly all three strains showed some degree of adaptation (see Figure 2C). In contrast, less evolved populations adapted significantly after stress by reaching the characteristic level of fitness of more adapted populations at longer times. Most importantly, such difference in post-stress adaptation patterns is directly matched by the difference of the frequencies of mutators caused by stress: it is markedly narrower for initially well-adapted populations than for less adapted populations (see Figure 2). Visser et al. hypothesized that more adapted populations have lower supply of strongly beneficial mutations, making the wait time for them to arrive longer [28]. In order to evaluate the importance of changes in fitness landscape upon adaptation we determined local “fitness landscapes” of populations (i.e. distributions of relative fitness change upon point mutations), immediately prior to heat shock, after heat shock and after post-stress adaptation (at t = 25,000) (see Figure 7). We found that while differences in fitness landscapes may be noticeable, there were no pronounced patterns of differences except for extremely rare mutations that change fitness significantly which were not found in populations which adapted to high temperature. Furthermore, our simulations also confirmed that the mutators were able to get fixed in the populations in response to the stresses of stationary phase and starvation, because the overall patterns of fitness landscapes were conserved against those stresses.
What is then an explanation for the bias to emerge and fix mutators in lesser adapted strains? In order to address this question we performed a control simulation where fitness is constant, independent of sequences (i.e. not determined by Eq.(1)), so that supply and fixation of mutators are decoupled. Since the main reason for fixation of mutators in case of “stationary phase” and “starvation” stresses appears to be hitchhiking with beneficial mutations in RCG, by eliminating hitchhiking, the sequence-independent fitness model focuses entirely on the supply of mutators rather than their fixation. We compared the average fraction of mutators in populations having different values of fixed fitness b. However, we still left a protein structural constraint which removes organisms due to protein malfunction if any of its proteins lost stability i.e. its . This constraint provided a weak selection against deleterious mutations, which arose more frequently in mutator clones. The results shown in Figure 8 suggest that populations of higher fitness contain less mutators. In order to understand this finding, we note that our chemostat regime simulates a limited-resource environment in which excess organisms are removed at random. High fitness in such an environment causes greater production of new organisms and hence a larger excess of organisms over the carrying capacity. Thus, more organisms must be culled at high fitness per unit time, which means a faster random drift. The trend shown in Figure 8 indicates that the level of fitness determines the frequency of mutator clones through random drift, because mutators are supplied at a constant rate by epigenetic stochastic switching and fitness determines the rate at which they are purged from the population due to genetic drift.
In this work we presented a model that combines biophysical principles of protein folding and protein-protein interactions in crowded cellular environments with population genetics and applied it to study universal principles of adaptation in asexual populations. The model is still mesoscopic as it includes simplified representation of proteins and their functional and non-functional interactions. However it is much more detailed and microscopic than more traditional population genetics models of evolution of mutation rates [14]–[16],[29],[30] because it derives fitness directly from an organismal genotype and protein concentrations in the cell and therefore can directly and explicitly assess the evolutionary consequences of genomic mutations. Unlike the conventional population genetics models, our model does not make a priori assumptions about the supply of beneficial and deleterious mutations and their effects on fitness and it does not assume a fixed fitness landscape or for that matter any a priori phenomenological fitness landscape. The main assumption of this model is the microscopic genotype-phenotype relationship Eq.1, which is is based on a number of intuitive biological assertions. First, in order to function, proteins have to be in their native (folded) state and participate in functional protein-protein interactions, when needed. Proteins in this model (and of course in reality) may participate in non-functional interactions (red boxes in Figure 1), however that would result in lower copy numbers of proteins available for functional interactions and biological activity and consequently would lead to lower fitness of an organism. Second, our model considers two types of genes: housekeeping genes (or RCGs) and genes that carry out control/regulation function, in this case gene 4, whose product is responsible for control of mutation rates. The fitness of an organism is proportional to concentrations of replication controlling proteins in their functional form as stipulated by Eq.(1). Third, production of proteins incurs a cost invoking a fitness penalty for overproduction. This constraint makes it detrimental for the total concentration of all proteins to go beyond some optimal level, and therefore it causes in some cases redistribution of resources between productions of different proteins rather than the increase of overall protein production. Expression levels of different genes determining, along with other factors, copy numbers/concentrations of their protein products can fluctuate on time scales, which are much faster than time scales of mutations in upstream regions which also cause changes in protein productions and be epigenetically inherited, reflecting epigenetic phenomena. This factor reflects extrinsic noise in gene expression, which were observed in many cell types [31],[32]. The epigenetic inheritance of protein concentrations is not a genetic phenomenon - rather it is due to long-time correlations in extrinsic noise in protein production which were found by Elowitz and coauthors [26]. Fourth, model cells replicate at the rates corresponding to their fitness levels, so that their population can grow until it reaches a threshold size, after which excess organisms are removed randomly to maintain fixed population size. This process sets an effective total death rate, which is equal to replication rate when population size is kept fixed. The prototypic system to model mutational control here is a mismatch repair (MMR) system, which involves several proteins that are functional in their dimeric (mutL) or tetrameric (dimer of dimers) form (mutS) [20],[33] Accordingly our model MMR proteins are functional in a homodimeric form. While diverse molecular mechanisms exist which determine mutation control under known stressful environments (e.g. rpoS dependent responses to DNA damage and other known stress responses [1],[5],[21]), most of the mutator bacteria isolated in the laboratory and in nature have been shown to downregulate or be defective in the MMR system [2],[10],[22]. Most importantly, changes in expression level or mutations in MMR system proteins represent the most universal response to stress, regardless of the bacterial species or the character of challenge. Our aim here is to elucidate the role of mutation rates in stress response in an ab initio model, and therefore using the MMR system as a prototype appears to be a logical choice.
In this study we investigated three types of stresses, which affect different properties of our model cells. Upon temperature stress, a mutator phenotype emerges simultaneously in most organisms due to destabilization of MMR complexes at higher temperature. There is no follow-up mutator fixation stage in this case (Figure 2). The rise of mutators after temperature jump is a clear example of a pleiotropic phenomenon where physical factors rather than adaptive mechanisms are responsible for the rise of mutators in the population. Second order selection does not play a significant role in the rise of mutators in temperature jump but plays a role in their fall by providing stabilizing mutations in the MMR complex, which bring mutation rates in the population back to the original low level. The sequence of events upon adaptation in two other types of stresses is quite different from the temperature jump stress. Here, both pleiotropy and second order selection play an important role in rise of mutators. Both decrease of the baseline value and drop in the optimal protein production level lead to an instant drop of fitness for all organisms (see Figure 2). Why would then such a uniform change as drop in result in a response? It may seem, at a first glance, that drop in should be equivalent to change in time scales without any material consequences. However, we found that the immediate consequence of fitness drop is the increased supply of mutators in the population of fixed size due to diminished genetic drift (Figure 8). The reason for that is the interplay of two time scales: a faster time scale at which fluctuations in protein production level supply organisms with mutators whereby MMR complexes fail to dimerize, and the time scale at which excess organisms are randomly killed in the environment which maintains a finite population size. As a result, at lower fitness levels, the net supply of mutators is greater providing a necessary diversity of mutation rates in the population which will give rise to subsequent fixation of mutators via hitchhiking. The initial supply of mutators is certainly a pleiotropic phenomenon in the sense that it is caused by physical processes, which are unrelated to adaptation. The increased supply of organisms which have higher mutation rates provides ample opportunity to acquire mutations in RCGs, which increase the fitness of an organism. This is clear from Figure S1, which provides clear evidence that mutations in RCGs are responsible for all increases in fitness. Fixation of the mutator phenotype in this case is a classical example of hitchhiking, i.e. second order selection.
Our model points out that noise in protein production levels is a major source of mutators, which are supplied in epigenetic and pleiotropic manners in stationary phase and starvation adaptation. The key feature of this mechanism is that it epigenetically produces greater diversity of mutation rates in populations than would have been possible due to genotypic diversity only at a very low natural mutation rate of approximately 0.003 mutations per genome per generation [34]. This factor supplies mutators, which improve fitness through beneficial mutations in RCGs (see Figure S1). Other hypothetical possibilities such as an elevated mutation rate in the upstream regions of the MMR genes might generate similar diversity, however we do not have evidence that such a mechanism does indeed exist.
In real biological systems, the noise-induced mechanism, which supplies mutators, can be supplemented and strengthened by directed regulation of copy numbers of MMR proteins. Experiments show that expression of mutS or mutL genes are often downregulated upon entering into stationary phase [23],[24],[35]. A decrease in copy numbers of MMR proteins is predicted by our model as a universal initial step in adaptation in stationary phase and starvation, leading to a quick transient fixation of mutator clones. A key component of the Escherichia coli MMR system, mutS, is efficient in its tetrameric form as dimer of dimers. It is noteworthy that at the conditions of exponential growth, the concentration of mutS dimers is close to the threshold of the dimer-tetramer equilibrium transition [23],[36]. The proximity of the concentration of the MMR components to the critical threshold makes the number of functional mutS tetramers most susceptible to noise and it explains the persistent presence of a small proportion (1–10%) of mutators in the adapted populations observed in our simulations (Figure 2) and in experiment [10],[27]. The importance of noise in gene expression for adaptation is indirectly supported by the observation that expression of stress-related proteins in Saccharomyces cerevisiae is controlled by TATA-containing promoters which are known to give rise to noisy gene expression while housekeeping genes are mainly under TATA-less promoters [37],[38]. Furthermore, Blake and coauthors showed that increasing noise in expression of stress related genes (by mutating the TATA region) resulted in greater benefit in adaptation to acute environmental stress [39]. More immediately, we predict that modulating noise in production of mutS proteins in E. coli without affecting the mean (e.g. by introducing mutations which decrease binding affinity of dimers concurrently increasing the expression level of the gene) would result in dramatically altered response to an unknown stress. The work to test these predictions is underway.
Our study highlights an important interplay of pleiotropic and genetic factors in generating mutator clones and suppressing them when the population adapts. In particular, Table 1 shows that the dominant mechanism by which populations return to normal mutation rates after adaptation is genetic - acquiring a mutation in the MMR gene which makes the complex more viable. The important role of recurrent losses and reacquisition of MMR gene functions was highlighted in the study by Denamur et al. who found that phylogeny of the MMR genes in E. coli is very different from that of the housekeeping genes [40]. These authors found the evidence that horizontal gene transfer of MMR genes may play an important role by increasing the rates of reacquisition of MMR function over those expected from compensating mutations only as implemented in our model. While our model does not allow for gene transfer it also points out an importance of changes in MMR genes in adapting populations. Gene transfer mechanisms may make these processes faster eliminating the need to wait for specific point mutations in the MMR genes.
Our microscopic evolutionary model of mutations and adaptation in populations of asexual organisms is still simple and minimalistic. It represents proteins at a coarse-grained level. Another important simplification of the model is mean-field treatment of PPI using the LMA approach. Such approach is good at time and length scales at which a protein participates, permanently or transiently, in multiple PPI. While certainly applicable to highly expressed proteins, the LMA treatment may be an oversimplification for proteins whose copy number in a cell is small. In this case either direct simulation of PPI in crowded cellular environment as in [41] or corrections to mean-field LMA as in [42] would be required for a more complete analysis. Furthermore, our simple 4-gene model, despite its explicit character, is certainly a major simplification of reality with its thousands of genes operating in a crowded cytoplasm. Nevertheless, the unique feature of this approach, in contrast to traditional population genetics studies of mutation rates, is that it couples first principles consideration of protein folding and protein-protein interactions with population genetics. We find that important aspects of our findings are due to the fact that the fitness landscape is not a priori pre-determined but is evolving as populations evolve. As such, this model provides a description of physical principles of adaptation on all scales, from individual proteins to their assemblies in cytoplasm to populations of asexual organisms. On the population level, we found that adaptation always proceeds through transient fixation of a mutator phenotype (except in cases of high fitness pre-stress populations). This is realized, on a microscopic level of proteins and their interactions, through a sequence of events involves a peculiar interplay of intrinsic noise and genomic variation. Utilization of noise in protein copy numbers to trigger a set of adaptation events provides a clear evolutionary advantage in meeting unforeseen challenges for which no detailed molecular response mechanism may be available. The fact that a minimalistic “first principles” model was able to describe realistically many principal aspects of molecular and cellular mechanisms of adaptation in real bacteria suggests that evolution uses general physics as its “design scaffold”, around which it builds a beautiful structure of living cells.
In our model, organisms carry 4 genes whose sequences and structures are explicitly represented. Each gene contains 81 nucleic acids, encoding 27-mer model proteins. Once it is expressed into a protein, it folds into a 3×3×3 compact lattice structure [43],[44]. Lattice models have been instrumental in gaining key insights into mechanisms of protein folding [44]–[47], protein design [17],[48] and evolution [49]–[51].
We reduce the range of all possible 3×3×3 lattice structures, which totals 103,346 [43], to randomly chosen and evenly distributed representative set of 10,000 structures for faster calculation. is the Boltzmann probability that a protein stays in its native structure whose energy is the lowest out of all 10,000 structures. There exist 144 rigid docking modes between two 3×3×3 lattice proteins, considering 6 surfaces for each protein and 4 rotations for each surface pair of two proteins (6×6×4). is the probability that two proteins i and j form a stable dimeric complex in the correct docking mode. and are proportional to the Boltzmann weight factors of the native structure energy, E0, and the lowest binding energy, as follows:(2)Where is energy of the native, i.e. the lowest energy, conformation (out of all 10,000 conformations) of a protein, which is a product of gene number . is energy of native, i.e. the lowest energy, binding mode (out of 144 possible ones) between proteins, which are products of genes .
The binding constant between proteins i and j is calculated as follows:(3)and these values are substituted into the Law of Mass Action (LMA) equations in Eqs.(5) and (6) to determine free concentrations of proteins and concentrations of their complexes . We use Miyazawa-Jernigan pairwise contact potential for both protein structural and interaction energies [52]. We report environmental temperature T in Miyazawa-Jernigan potential dimensionless energy units.
Simulations start from a population of 500 identical organisms (cells) each carrying 4 genes with initial sequences designed to be stable in their (randomly chosen) native conformations with . At each time step, a cell can divide with probability b given by Eq.(1). A division produces two daughter cells, whose genomes are identical to that of mother cells apart from mutations that occur upon replication at the rate of mutations per gene per replication. Mutation rate m depends on concentration of functional (homodimeric) MMR proteins (products of gene 4) as specified below. The stability loss of any protein by a mutation () incurs lethal phenotype [53], and the cell carrying such gene is discarded. Constant death rate d of cells is fixed to 0.005/time unit, and the parameter is adjusted to set the initial birth rate equal to the fixed death rate (b = d). The control coefficient in Eq.(1) is set to 100. All parameters are listed in Table 2.
We simulated a chemostat regime: when the population size exceeded 5000 organisms, the excess organisms were randomly culled to bring the total population size back to 5000. Initially total protein concentrations are set equally for each protein at . Protein concentrations (determined in vivo by expression levels of corresponding genes and translation/degradation) values can fluctuate with rate determined by parameter , (see below) unrelated to the mutation rate reflecting primarily the epigenetic factors such as long-time correlated extrinsic noise in protein production [26],[54]. Due to the long time correlation in protein production levels, values of appear epigenetically inherited. Fluctuations in protein production levels are modeled in the following way. At each time step the value of may stay unchanged with probability or, with probability , change. The magnitude of the change is random:(4)where and are the new and old concentrations of protein product of i-th gene, is drawn from Gaussian distribution whose mean and standard deviation are 0 and 0.1, respectively.
Parameter characterizes the rate of fluctuations of protein copy numbers; we take unless otherwise is noted.
The concentration of free (uncomplexed) proteins is determined from the LMA equations that assume that monomers and binary complexes can form:(5)where is the binding constant of interactions between protein i and protein j [55] and concentrations of binary complexes between all proteins (including homodimers) are given by the LMA relations:(6)
We determined, after each change (a mutation or a fluctuation in Ci), all necessary quantities by solving the LMA Eqs.(5) and (6) to find , , and evaluate the new for mutated protein(s) and and for the complex of protein pairs to be in their specific functional states as explained above. We solve coupled nonlinear LMA equations by iterations. Once changes or a mutation occurs, the old set of is substituted into the right hand side of Eq.(5) and a new set of is calculated. This procedure iterates until the difference between old and new values of drops below 0.1% of the new value.
To simulate a variable mutation rate, we consider protein 4 to be a component of DNA mismatch repair (MMR) machinery using an important part of it – mutS - as a prototype. Mutation rate at any time step t depends linearly on the concentration of functional MMR dimers, which is(7)where the first term in the right hand side of Eq.(7) is a concentration of binary dimeric complex of the MMR protein, the second term is the thermal probability that homodimer forms a fixed functional conformation out of 144 possible docking modes and the last term stems from the requirement that both members of a functional MMR complex have to be in their native folded conformations.
The mutation rate depends linearly on concentration of functional MMR complexes:(8)where is maximal mutation rate. We define as a mutator a clone whose mutation rate is greater than 0.01, i.e. 100 times or more higher than lowest (i.e. wild-type) value of 0.0001. The latter value is typical of non-mutator E. coli strains [34]. While the dependence presented by Eq.(8) is the most natural one, the results do not depend significantly on this assumption: a threshold-like dependence where mutation rates can take two values depending on whether is below or above a certain threshold gives qualitatively the same adaptation behavior as presented here for the model Eq.(8) (data not shown).
In order to seed simulations with organisms whose initial state is non-mutator, we designed the initial sequences for protein 4 to be stable and to form a strong homodimer using design algorithms described in [41],[48]. We did not initially design interactions between products of RCGs, so that populations start from non-adapted growth rate conditions.
|
10.1371/journal.pntd.0000879 | Control of Tungiasis through Intermittent Application of a Plant-Based Repellent: An Intervention Study in a Resource-Poor Community in Brazil | Tungiasis, an ectoparasitosis caused by the female sand flea Tunga penetrans, is an important health problem in many impoverished communities in the tropics. Sand flea disease is associated with a broad spectrum of clinical pathology and severe sequels are frequent. Treatment options are limited.
We assessed the effectiveness of the intermittent application of the plant-based repellent Zanzarin to reduce infestation intensity and tungiasis-associated morbidity in a resource-poor community in Brazil, characterized by a very high attack rate. The study population was randomized into three cohorts. Initially, during a period of four weeks, the repellent was applied twice daily to the feet of all cohort members. This reduced the number of embedded sandfleas to 0 in 98% of the participants. Thereafter members of cohort A applied the repellent every second week twice daily for one week, members of cohort B every fourth week for one week, and members of cohort C served as controls. Infestation intensity and tungiasis-associated morbidity were monitored during five months. The intermittent application of Zanzarin for one week every second week significantly reduced infestation intensity from a median 4 lesions (IQR 1–9) during the whole transmission season. In contrast, in cohort B (application of the repellent every fourth week) the infestation intensity remained twice as high (median 8 lesions, IQR 9–16; p = 0.0035), and in the control cohort C 3.5 times as high (median 14 lesions; IQR 7–26; p = 0.004 during the transmission season). Tungiasis-related acute pathology remained very low in cohort A (median severity score 2; IQR 1–4) as compared to cohort B (median severity score 5; IQR 3–7; p<0.001), and control cohort C (median severity score 6.5; IQR 4–8; p<0.001).
Our study shows that in a setting with intense transmission, tungiasis-associated morbidity can be minimized through the intermittent application of a plant-based repellent.
| Tungiasis is a parasitic skin disease caused by the female sand flea Tunga penetrans. The disease is frequent in resource-poor communities in South America and sub-Saharan Africa and affects the poorest of the poor. Sand flea disease is associated with a considerable morbidity and may lead to tetanus in non-vaccinated individuals. The degree of morbidity depends on the intensity of infestation, i.e., the number of embedded sand fleas a person has. Since tungiasis is a zoonosis involving a host of animal reservoirs, and because an effective treatment is not at hand, in resource-poor settings elimination is not feasible. Preventing morbidity to develop is therefore the only means to protect exposed individuals from sand flea disease. Similar to other arthropods, sand fleas can be repelled before they penetrate into the skin. In this study we show that the intermittent application of a plant-based repellent, of which the major component is coconut oil, reduces the intensity of infestation dramatically during the whole transmission season and prevents tungiasis-associated morbidity from developing. The prevention can be performed at the household level by the affected individuals themselves with minimal input from the health sector.
| Tungiasis is a common, but neglected health problem in economically disadvantaged communities in tropical and subtropical countries [1]–[5]. The female sand flea Tunga penetrans penetrates into the epidermis of its host, undergoes a peculiar hypertrophy, expels several hundred eggs for a period of three weeks, and eventually dies in situ [6]. In endemic areas constant reinfestation is the rule and affected individuals frequently harbor dozens, sometimes hundreds of embedded parasites [7]. Tungiasis is acquired peri-domiciliary, but also inside the house [8]. Different species of animals act as reservoirs [4].
The ectoparasitosis mainly affects marginalized populations in urban squatter settlements, in villages in the hinterland, and in traditional fishing communities along the littoral [1], [3], [8]–[13]. In these settings tungiasis is commonly associated with an important morbidity, such as intense inflammation of toes and heels, painful fissures, ulcers and abscesses, as well as deformation, and loss of nails [4]–[6], [13]–[16]. Difficulty in walking is common; gangrene and tetanus are life-threatening sequels [4], [14], [17]–[20].
Since it is virtually impossible to eliminate tungiasis as long as the precarious living conditions characteristic for impoverished communities exist, morbidity control remains the only option. There is no effective chemotherapy available to kill embedded sand fleas. Parasites need to be extracted surgically with a sterile instrument. However, this requires a skilled hand and good eyesight. In resource-poor communities surgical removal is inconsistently performed and causes more harm than good if not done correctly [6], [21], [22]. We have previously shown that a regular twice daily application of Zanzarin, a repellent based on coconut oil, for a period of three weeks, reduced the rate of newly embedded fleas by 92%, and reversed tungiasis-associated clinical pathology almost completely [20]. In this study we investigated, whether an intermittent application of the repellent protects inhabitants of an area with intense transmission against penetrating sand fleas during the transmission season, in Northeast Brazil a period of six month.
The study was conducted in five neighborhoods (Luxou, Morro das Sandra's, Placas, Morra da Vitória and Novo Rumo) of the shantytown Vincente Prinzón, a typical conglomeration of urban squatter settlements (favela) in Fortaleza, Northeast Brazil. The area is characterized by intense transmission of T. penetrans [7]. The five neighborhoods are located on dunes near to the Atlantic Ocean. The area had been occupied by landless poor (“sem terra”) in the beginning of the 1950’s [23]. The poor living conditions have been described previously [24]. In brief, most houses are constructed with recycled litter and do not posses a concrete floor. Illiteracy, unemployment, crime, alcoholism, drug traffic, abuse, as well as adolescent prostitution, and domestic violence are common. Sixty percent of the population have a monthly family income of less than two minimum wages (1 monthly minimum wage 300 Real ≈ 100 Euro) [23]. In 2005, about 20.000 people inhabited the shantytown and were served by a single primary health care center.
A randomized controlled trial was carried out in the five neighborhoods between June and December 2005. This period (dry season) coincides with the high transmission season of T. penetrans resulting in an attack rate of up to 10 newly embedded sand fleas per persons per day in a similar setting nearby [25].
Individuals with tungiasis were identified with the assistance of community health workers. They were included in the study provided they had at least 5 embedded sand fleas in stage 1 to 4 of the Fortaleza Classification, or a similar number of sand flea lesions manipulated with a perforating instrument [26]. Individuals who intended to change their place of residence during the next six months were not eligible. Individuals with ulcerated lesions necessitating antibiotic treatment, and children less than one year were excluded. In total 142 participants were recruited.
The intensity of infestation and the degree of tungiasis-associated morbidity was assessed as described previously [16]. After the admission examination the participants were randomized into three cohorts (A, B and C). In the first phase of the study individuals received a twice-daily application of Zanzarin for a period of four weeks. It was anticipated that this reduced the number of embedded sand fleas to almost zero [20]. Thereafter participants were examined again and the degree of the tungiasis-associated morbidity was assessed. These data provided the baseline for the subsequent study phase.
During a period of five months members of Cohort A applied the repellent every second week twice daily for one week, and members of Cohort B every fourth week twice daily for one week. Cohort C served as control group and did not receive any protection.
During the intervention periods Zanzarin was applied by trained community health workers on the skin of the feet (up to the ankle) including the interdigital areas. The average volume applied was 3 ml per person and day (Calculation based on the number of 100 ml bottles used per day, divided by the number of treated individuals). Prophylaxis was performed in the morning between 6 and 8 a.m., and in the evening between 6 and 8 p.m. The exact time of application was recorded for each study participant. The application of the repellent was regularly checked by random visits to the households of the study participants by one of the investigators (J.B.). In addition, at each examination the participants were asked whether the repellent had been applied regularly. This ensured that the repellent was applied exactly as defined in the study protocol, not spilled, given away for money, or stolen. The participants were asked not to wash their feet for at least two hours after application of the repellent. However, they were allowed to take a shower whenever they wanted.
Unexpectedly, staff members were assaulted during the second part of the study. For safety reasons we decided to interrupt the regular follow-ups at calendar week 45. Monitoring was resumed at calendar week 47. Nonetheless, the application of the repellent was continued during this period.
The repellent used was Zanzarin, a lotion based on coconut oil (Cocos nucifera), jojoba oil (Simmondsia chinesis) and Aloe vera (Engelhard Arzneimittel GmbH & Co. KG, Niederdorfelden, Germany). The lotion is sold as a biocide with repellent activities against ticks and biting insects. The exact composition is described in an annexed file (Text S1). As Zanzarin has a peculiar odor, neither the investigator nor the patient were blinded to group assignment. Adverse reactions were interrogated and documented at each visit. A previous study has shown that Zanzarin, is highly effective in preventing the infestation with sand fleas [27].
Since tungiasis can occur at any area of the body [28], [29], the whole body surface was examined for the presence of immature, egg-producing or dead fleas, and manipulated lesions. Tungiasis lesions were classified according to the Fortaleza Classification [26]. The following findings were considered diagnostic for tungiasis:
– Flea in statu penetrandi (stage I)
– A dark and itching spot in the epidermis with a diameter of 1 to 2 mm, with or without local pain and itching (early lesion, stage II)
– Lesions presenting as a white halo with a diameter of 3 to 10 mm with a central black dot (mature egg producing flea, stage III)
– A brownish–black circular crust with or without surrounding necrosis of the epidermis (dead parasite, stage IV).
During monitoring the number of viable (stage I to III), and dead (stage IV) sand fleas, and the total number of sand flea lesions were determined. Clinical pathology was documented every four weeks. Lesions manipulated by the patient (such as partially or totally eliminated fleas leaving a characteristic crater-like sore in the skin), and suppurative lesions caused by the use of non-sterile perforating instruments, such as needles, and thorns, were documented as well. The exact topographic localization of each lesion, its stage, and appearance were documented on a visual record sheet.
Clinical pathology was assessed in a semi-quantitative manner using a previously elaborated severity score for acute tungiasis (SSAT), and a severity score for chronic tungiasis (SSCT) [16]. The SSAT score comprises the following signs and symptoms: erythema, edema, pain upon pressure or spontaneously, itching, sleep disturbance due to itching, difficulty walking as indicated by an altered gait; abscess, and suppuration as indicators of superinfection; fissures, and ulcers as characteristic chronic skin defects [12], [26]. The score can take a value from 0–24 points.
The SSCT ranges from 0 to 33 points and comprises the presence of nail deformation, nail loss, brilliant skin (an indicator of chronic edema), deformation of toes; hypertrophic nail rim, and perilesional desquamation; the latter two characteristics are indicators of repeated tungiasis experienced in the past [12], [16], [26].
Households were randomized using a permuted block design (block size six, allocation ratio 1∶1). An investigator not involved in the follow-up visits created the randomization code. A computer generated random list was used.
All data were entered into an Epi-Info database (CDC, Atlanta, Ver. 6.04d) and checked for errors, which might have occurred during data entry. The database was exported into SigmaStat and SigmaPlot (Systat Software GmbH, San José, Version 2007). The main outcome measure was the intensity of infestation, i.e. the number of sand flea lesions present at the time of examination. Secondary outcome measures were the severity score of acute, and the severity score of chronic pathology. As the variables assessed were not normally distributed and variances varied considerably, the median and the interquartile ranges were used to indicate the average and dispersion of data. To compare results between the cohorts, the Wilcoxon Signed Rank Test was used; for correlation analysis, Spearman’s Rho was calculated. In order to detect a difference of 50% in infestation intensity between cohorts A and C, a sample size of 42 individuals for each cohort was calculated (level of significance 95%, power of the test 80%). In order to compensate for drop-outs it was decided to recruit 140 individuals to the study.
The study was approved by the Ethical Committee of the Federal University of Ceará, Brazil (43/05, SINESP) and was registered at Controlled-trials.com (ISRCTN16910507). Informed written consent was obtained from all participants and in the case of minors from the parents or legal guardians. At the end of the study, all participants as well as their household members were carefully examined for the presence of embedded sand fleas. Individuals with tungiasis were treated with Zanzarin twice daily for a period of three weeks, a measure effectively reducing the number of embedded sand fleas and the degree of clinical pathology to almost zero [20].
The flow diagram of the study is depicted in Figure 1. The median age of all participants was 8 (range 1–66) with no difference between the cohorts. 43.9% of the participants were males, and 56.1% were females. The parasitological characteristics of the study population are depicted in Table 1. There was no difference between the three cohorts.
After the initial intervention (application of Zanzarin twice daily in all cohort members during a period of four weeks) the infestation intensity decreased from a median of 17 (IQR 11–30), to a median of 0 (IQR: 0–1; p<0.001). Only two participants showed more than one embedded sand flea. This was paralleled by a drastic reduction of the SSAT: median before intervention 8.5 (IQR: 6–11), versus 0 (IQR: 0–1; p<0.001). The SSCT score was also reduced: median before intervention 12 (IQR: 9–15), versus 7 (IQR: 5–10; p<0.001). No adverse reactions to the repellent were reported.
In this study we investigated, whether the intermittent application of Zanzarin keeps the infestation rate at an acceptable low level, and prevents severe clinical pathology to develop during the transmission season, which coincides with the dry season of the year in most endemic areas [15], [26]. In Northeast Brazil transmission occurs mainly between July and December, i.e. a period of 6 months.
Since infestation rate and infestation intensity are closely related [8], [16], and assessment of the infestation rate is very laborious, we determined the infestation intensity in three cohorts at regular intervals, together with the degree of tungiasis-associated pathology. In cohort A (application of Zanzarin every second week for one week) the number of embedded sand fleas only slightly increased during the transmission season. In cohort B (application of the repellent every fourth week for one week) the intensity of infestation was considerably higher during this period. However, even the long intervention-free intervals resulted, at the end of the transmission season, in an infestation intensity of approximately half the number of embedded sand fleas found in the control cohort (Figure 2).
If sand fleas are effectively repelled, only a few new parasites will penetrate per unit of time. This is reflected by a low number of viable lesions (stage I to III of the Fortaleza classification). In fact, the median number of viable sand flea lesions per individual showed different patterns in each of the cohorts. Whereas in cohort A the median number of viable sand fleas was 1, in cohort B – and even more so in cohort C - the number of viable sand flea lesions increased in a step-wise manner during the intermittent application of the repellent (Figure 3). These findings indicate that after the application of Zanzarin for one week a residual effect of the repellent seems to occur for a couple of days, but that an interval of four weeks without the application of the repellent is too long to prevent sand fleas invading the skin. Previous observations had raised the expectation that a residual effect may persist for more than one week [20].
The irritation and pain caused by embedded sand fleas is the reason why affected individuals try to get rid of the parasites with sharp instruments. Supposedly, the more sand fleas penetrate and embed per unit of time, the higher is the proportion of lesions manipulated with instruments. Hence, an effective repellent will be reflected by a low percentage of manipulated lesions. Indeed, the effectiveness of Zanzarin to prevent a high intensity of infestation to build up, is mirrored by the pattern of manipulated lesions in the three cohorts. In cohort A, manipulated lesions were completely absent until almost the end of the transmission season, whereas in cohort C the number of manipulated lesions started to increase almost constantly after the end of the initial intervention, and with a certain delay also in cohort B (Figure 4).
As expected, the almost complete interruption of transmission in cohort A prevented severe pathology to develop. This is mirrored by a decrease between 64% and 88% of the SSAT score during the transmission season, as compared to the degree of acute pathology at admission (Figure 5). A median SSAT score of 2 points at the end of the study in this cohort indicates an insignificant degree of acute pathology [20]. When the intervals between the application of Zanzarin were extended to three weeks, the degree of clinical pathology developing during the transmission season almost doubled, as compared to cohort A, and was only slightly less than in the control group (Figure 5). Hence, regular application of Zanzarin every second week is necessary to reduce clinical pathology to an insignificant level.
The interval application of Zanzarin was less effective in reducing the presence of chronic pathology (Figure 6). This is plausible, since chronic pathology, such as nail deformation and fibrosis of the skin around the nail rim, needs long time without new infestations, to resolve at least partially [20]. However, it is important to note that the application of the repellent every second week significantly reduced the SSCT score during the follow-up period, as compared to the value at admission.
Based on an average of 3 ml of Zanzarin applied per individual per day, a member of Cohort A needed a total of 210 ml of the repellent for the whole transmission season. If Zanzarin is bought in bulk quantity, 10–12 US$ would be sufficient to protect one person against the debilitating sequels of tungiasis during the whole transmission season. Since children and the elderly are particularly affected by severe manifestations of sand flea disease [12], prophylaxis could be targeted to these population groups, reducing the costs of the repellent. The application of a repellent is a simple and sustainable means to relief the poorest of the poor from a scourge that reappears each year with the beginning of the dry season. Prevention could be achieved by preparing the repellent from local coconuts by the affected individuals themselves with minimal input from the health sector. Taken together, the intermittent application of Zanzarin, twice daily for one week every second week, effectively interrupts transmission of T. penetrans in an area with a high attack rate, and prevents severe morbidity to develop.
|
10.1371/journal.pntd.0007701 | Long-term retrospective assessment of a transmission hotspot for human alveolar echinococcosis in mid-west China | Human alveolar echinococcosis caused by infection with Echinococcus multilocularis is one of the most potentially pathogenic helminthic zoonoses. Transmission occurs involving wildlife cycles typically between fox and small mammal intermediate hosts. In the late 1980s/early 1990s a large focus of human AE was identified in poor upland agricultural communities in south Gansu Province, China. More detailed investigations in 1994–97 expanded community screening and identified key risk factors of dog ownership and landscape type around villages that could support susceptible rodent populations. A crash of the dog population (susceptible domestic definitive host) in the early 1990s appeared to stop transmission.
We subsequently undertook follow-up eco-epidemiological studies based on human population screening and dog survey, in 2005/6 and in 2014/15. Our observations show a decrease in human AE prevalence, especially marked in the 11–30 year old age category. In 2015, although the dog population had recovered and in addition, forest protection and the reforestation of some areas may have favoured red fox (wild definitive host) population growth, there was no evidence of infection in owned dogs.
Those observations suggest that over decades socio-ecological changes resulted in a cascade of factors that exacerbated and then interrupted parasite emergence, with probable elimination of peri-domestic transmission of E. multilocularis in this area, despite the relative proximity of large active transmission foci on the eastern Tibetan Plateau. This study case exemplifies how anthropogenic land use and behavioural changes can modify emergence events and the transmission of endemic zoonotic parasite infections, and subsequently the importance of considering processes over the long-term in a systems approach in order to understand pathogen and disease distribution.
| Human alveolar echinococcosis caused by infection with Echinococcus multilocularis is one of the most potentially pathogenic helminthic zoonoses. Transmission occurs involving wildlife cycles typically between fox and small mammal intermediate hosts. A large focus of human alveolar echinococcosis was identified in the late 1980s in poor upland agricultural communities in south Gansu Province, China, and has been monitored until 2015. Observations suggest that over decades landscape and socio-ecological changes resulted in a cascade of factors that exacerbated and then interrupted parasite emergence, with probable elimination of peri-domestic transmission of E. multilocularis in this area, despite the relative proximity of large active transmission foci on the eastern Tibetan Plateau. This study case exemplifies how anthropogenic land use and behavioural changes can modify emergence events and the long-term transmission of endemic parasitic infections, and subsequently the importance of considering disease ecology transmission socio-ecosystems in order to understand parasite and disease distribution.
| Zoonotic infections that involve domesticated animals and/or wildlife hosts are of increasing concern globally, especially in resource-poor rural communities, and they are usually difficult to monitor, treat and control [1]. Chronic zoonotic parasitic helminthic infections such as trematodiases, cysticercosis and echinococcosis pose additional difficulties caused by long-term pathologies and variable periods of asymptomatology followed by non-specific symptoms. Human alveolar echinococcosis (AE) results from accidental infection with eggs of the canid small tapeworm Echinococcus multilocularis. It is one of the most pathogenic helminthic infections due to development of hepatic multivesiculated metacestode lesions with tissue fibrosis, necrosis and metastatic potential. The life cycle of E. multilocularis involves carnivores as definitive hosts (primarily foxes but also dogs) and small mammal herbivores such as rodents and lagomorphs as intermediate hosts [2]. Despite being globally rare, human AE places a serious burden on affected communities in focal endemic areas and remains difficult and expensive to diagnose and treat. As for many zoonoses, incidence rates of human AE are associated with lifestyle, host ecology and specific transmission ecosystems [3–10]. For example Tibetan pastoral communities of alpine valleys in northwest Sichuan are estimated to lose 0.81 Disability Adjusted Life Years (DALYs) per person due to alveolar and cystic echinococcosis, compared to an average 0.18 DALYs lost in the general Chinese population due to all communicable and non-communicable ailments combined [11]. It is now clear that echinococcosis is a major burden for communities in endemic areas of China and poses a public health problem of primary importance [12].
Since the first hydatid control programme for cystic echinococcosis was implemented in Iceland in the 1860s [13], at least 20 intervention programmes have been undertaken in different world regions targeting mostly E. granulosus and less frequently transmission of E. multilocularis. The first control programme for AE occurred in Reubun Island, Japan, from the 1940s when the fox population was eliminated [14]. Modern control of E. multilocularis transmission can no longer rely on fox elimination considering the large scale at which such programmes should be implemented on continents and the subsequent ethical, ecological and technical issues raised by such targets [15]. However distribution of baits containing praziquantel can have significant impacts on vulpine prevalence of E. multilocularis, but is difficult to maintain over long periods and large geographic areas [16]. Contact with dog definitive hosts can be a major risk factor for human AE infection in communities where dogs live in close vicinity to humans, for example on St Lawrence Island in the Bering sea [17] and rural areas of China [5,18].
A critical aspect of echinococcosis control is the definition of an adequate baseline at the beginning of the intervention and the appropriate surveillance of disease incidence or prevalence in well targeted hosts [14]. Metacestode development in human AE is extremely slow with an asymptomatic period of 5 to 15 years or more [19,20]. A consequence is that the epidemiology of human AE at a given time in a given area might not reflect the current transmission status in the area, but, with a time lag, integrate over years the results of transmission systems that have been variously active decades ago. This is the case when environmental conditions conducive to transmission have changed naturally or as a result of anthropogenic impacts. This is further complicated when chronic diseases, such as AE, often fail to cause hospitalization or hospital records are inadequate, thus the `memory`of the epidemiological baseline can be lost. In ecology, quantified long-term adaptive monitoring [21], although crucial is still often problematic (short-term funding and difficult metrology of wildlife population variables, etc.) [22].
In the absence of a long-term adaptive monitoring framework, post-hoc ‘reactive’ monitoring can be implemented [23]. It is based on collecting available historical data and assembling them in a retro-observatory database in order to rebuild the time series that should have been monitored longitudinally. This can be a way for documenting environmental and epidemiological changes and understanding how transmission patterns can evolve over the long term and help to design more sustainable integrated control strategies and better prevent risks of re-emergence after parasite elimination.
The first comprehensive early studies on human AE and the parasite transmission ecology in China were undertaken in the early 1990s in Han agricultural communities in southern Gansu Province. Prior to that, in the 1980s, a cluster of AE cases were reported from hospitals in Lanzhou (capital city of Gansu) that originated from Zhang county, a poor partly terraced upland region about 250km to the south [24]. A subsequent investigation (in 1991) of six communities in two valleys in Zhang county revealed initially a high seroprevalence (8.8%) for specific E. multilocularis antibodies in a sample population (n = 606). Following that sero-survey confirmation of AE lesions in seropositives was undertaken and a community mass screening program was made using hepatic ultrasound scanning in a larger population (n = 1312). That study indicated a 5% prevalence of hepatic AE with a case age range of 11–73 years and a mean of 40.9 years [25]. Furthermore a necropsy study of unwanted dogs revealed a 10% prevalence of intestinal E. multilocularis infection, however the local red fox (Vulpes vulpes) population was not examined, neither were small mammals, though field mice (Apodemus spp.) and zokors (Eospalax spp.) were commonly trapped by locals close to villages [25].
Three years later, between 1994 and 1997, a much larger study was undertaken in Zhang county and the neighbouring Puma district (Min county), in which 2482 persons were voluntarily screened in 1994–6 by portable ultrasound in their villages (n = 31). That program detected 84 AE cases (3.4% prevalence, but increasing to 4.1% when AE cases identified in the 1991 survey were included) with a mean age of 38.7 years (range 12–70 years) [18]. Village human AE prevalences varied from 0% to >10% and 9 villages had AE prevalences >5%, with main risk factors for an AE case being: female >20 years old, landscape-type around village of domicile (ie.>50% scrub/grassland), presence of free-roaming/scavenging dogs, the number/history of dog ownership and dog carer [18]. In addition small mammal species assemblages were investigated in depth, and two dominant susceptible host species (Microtus limnophilus and Cricetulus longicaudatus) with potential for pluriannual population increases, were shown to occur in the high risk scrub/grassland habitats [4,26]. However, by 1994 the domestic dog population had crashed to almost zero throughout the region, most probably as an indirect result of rodent poisoning campaigns. The dog population only began to recover towards the end of that decade. The red fox population remained extremely small since the mid-1990s according to local farmers’ testimony. Thus, despite the identification of significant numbers of human AE cases in the 1994–97 screenings, that were mostly latent infections from 10–20 years earlier, it was considered that active peri-domestic transmission of E. multilocularis had probably ceased, and furthermore any wildlife cycle would be hardly sustainable. In 2005/2006 a third mass screening program for human AE was implemented in the Zhang/Puma area of south Gansu Province, furthermore by 2015 the dog population had fully recovered which was sampled by coprotesting. Those studies are now reported here and considered in relation to the long-term retrospective view of disease and transmission ecology.
The aim of this article is to attempt to describe the natural history and the fate of E. multilocularis transmission over >25 years in Han farmer communities of south Gansu, and to illustrate how changes in socio-environmental conditions can modify parasite transmission over the long term.
Studies were carried out in the original area of investigation that was undertaken in 1991 and 1994–97, in Zhang and Min counties of south Gansu [18] (Figs 1 and 2). The area (~350 sq. km) is characterized by 31 small discrete villages in valleys and plateau at approximately 2400–2600 m alt (kml file is provided as S1 File). The rural population comprised of greater than 98% Han Chinese, most of whom were subsistence farmers (rape, wheat, potatoes, soya) with a few medicinal herbs such as “dang gui” (当归, Angelica sinensis). The area is characterised by short, wet and warm summers (15–30°C) and cold winters -10 to -20°C, average yearly rainfall is 550–600 mm. Villages ranged in size from 100 to 1700 people with an average population of 350. Livestock, though not abundant, comprised of sheep and goats, pigs, chickens and small herds of cattle-yak cross breeds (“pian niu”, 犏牛).
Mass screening of volunteers (self-selected) by abdominal B-ultrasound (US) scanning, complemented by serology was undertaken over the period 2005–6 (approximately 10 years after the main investigation in 1994–97). A liver scan was voluntarily performed on each person by an experienced sonographer using a portable scanner. Liver lesions/cysts if present were identified as definite, probable or query AE disease and any AE lesions classed according to the PNM classification [30]. Serological antibody testing was performed using a panel of crude native antigen extracts ie. E.granulosus cyst fluid (EgCF), protoscoleces (EgP), antigen B (EgB) and a purified specific E. multilocularis metacestode antigen (Em2) [31]. In addition, a questionnaire was administered relating to knowledge, attitudes and practices, including aspects of dog ownership.
Information about AE cases from the 1994–97 survey was followed-up in May 2014 by interviews with hospital and clinic staff in the endemic zone ie. CaoTan, Han Chuan (now Dong Chuan) and Puma, and with the head person from villages of respective patient domicile.
X2 was used to test the null hypothesis of categorical variable independence in contingency tables. Human AE prevalence was modelled with infection status as response variable (1/0) against independent variables using General linear model (GLM) and General additive model (GAM) with a binomial (logit) link function. Here, GAM uses cubic smoothing splines with various degrees of freedom on independent variables in order to better take possible non-linearity of the response into account [32]. Models were compared and selected using the Akaike Index Criterion [33]. Computing and graphical display were performed using R 3.5.3 [34] and the package gamlss [35].
1994–96 and 2005–6 human screening programs were ethically approved by the relevant authorities of the Lanzhou Medical University. From 1994–96 AE cases were offered a free 6 months course of albendazole (ABZ), referral to the local medical centre or hospital and subsidized surgery for relevant cases [18]. From 2006/2007, the Ministry of Health provided both free ABZ-based treatment and heavily subsidized surgical intervention if deemed appropriate. All adult subjects provided written informed consent, and a parent or guardian of any child participant provided informed consent on the child’s behalf.
All animals were handled in strict accordance with good animal practice according to the Animal Ethics Procedures and Guidelines of the People’s Republic of China, and from 2012 the study was approved by the Animal Ethics Committee of Lanzhou Veterinary Research Institute, Chinese Academy of Agricultural Sciences (No. LVRIAEC2012-007).
Human alveolar echinococcosis caused by infection with Echinococcus multilocularis is one of the most potentially pathogenic helminthic zoonoses. Transmission occurs only in the northern hemisphere primarily involving wildlife cycles typically between fox and small mammal intermediate hosts. In the late 1980s/early 1990s we identified a large focus of human AE in resource-poor upland agricultural communities in south Gansu Province, China [25]. More detailed investigations in 1994–97 expanded community screening and identified key risk factors of dog ownership and landscape type around villages that could support susceptible rodent populations [26,4,18]. We subsequently undertook follow-up eco-epidemiological studies in 2005/6 and in 2014.
The 2005–2006 community mass ultrasound screening in Zhang and Puma areas showed a decrease in AE prevalence in the human population, which was most marked in younger age categories indicating that they were less exposed to AE infection. Some human AE infection post-1995, however was evidenced after community mass screening by at least one case < 10 years of age, indicating that the parasite E. multilocularis was still circulating in the late 1990s or early 2000s. In May 2014, we could not identify in hospital records any “new” AE patients (not included in the 1994–97 screening) that were born after 1997. The youngest of the AE cases was 35 years old at diagnosis, hence had an age compatible with an infection prior to 1994.
In May 2015, in contrast to the period 1994–97, a large number of owned dogs were present in villages in the endemic area with most of them tied, a fraction treated with PZQ, but some were allowed to free-roam and possibly eat small mammals. A large number of small mammal indices (species of the genera Microtus, Cricetulus and Eospalax) could be observed during the 2015 survey, at similar frequencies to those observed in the 1990s (Giraudoux P., personal observations, see Fig 2e and 2g). This supports the notion that a peri-domestic dog–small mammal life-cycle for E. multilocularis could still be potentially completed in the area, at least by a proportion of the dog population. Moreover, the fact that some dogs had been imported from Maqu county (Gannan Tibetan Autonomous Prefecture) increased the risk of possible re-introduction of E. multilocularis infection from distant endemic areas of the eastern Tibetan plateau [27]. Furthermore, since the late 1990s, soil protection and reforestation programmes have largely extended bush and forest areas on slopes too steep to be cultivated without deleterious erosion (Fig 7). This provided more potential habitats for wildlife including red foxes (V. vulpes), the latter being more frequently observed by farmers in 2015 than in the 1990s, when the red fox was believed to be virtually extinct in the area [4]. Nevertheless, we did not find any evidence for E. multilocularis infection in dogs, indicating that a peri-domestic cycle had likely not re-emerged by 2015. The slow progressive recovery of the dog population post 1997, with a probable low or zero infection rate, combined with a small fox population, might have locally interrupted the parasite life-cycle, and kept transmission very low for a period after the early-1990s. This could explain the decrease in human AE prevalence observed in 2005–2006 in younger age categories. In 2005, a National Echinococcosis Control programme was initiated in western China, and included improved surveillance (and treatment access) of human disease and regular deworming of dogs [39].
It is now well recognized that anthropogenic land-use changes drive a range of infectious disease outbreaks and emergence events and modify the transmission of endemic infections. These drivers include agricultural encroachment and deforestation [40]. Patz et al. [41] have proposed a systems model echoing those proposed by WHO [42] and MEA [43]. It includes specific health risk factors, landscape or habitat change, and institutional (economic and behavioural) levels and their health consequences. Various levels of investigation and intervention were described and ranged from specific risk factors and determinants of population vulnerability, to larger institutional and economic activities. Such conceptual models have been specifically considered for E. multilocularis transmission and human AE [4,44,8,10]. They confirm the inextricable linkages between environmental, socio-economic and proximal behavioural factors influencing the transmission dynamics of this helminthic zoonosis. All those models lead to the notion that, not one main factor alone, but a combination of factors, convergent or concurrent, affect the transmission dynamics of E. multilocularis in a nested hierarchy of time and space [44]. In such complex transmission systems, there is little hope for isolating a single aspect that would explain transmission ecology for example in the Zhang/Puma area of western China. However, accepting a form of holistic approach from the start, one can isolate subsystem groupings of predominant factors that altogether help to explain transmission patterns in an area for a given time span and scale, and to identify the tipping points when one shifts from one subsystem to the other.
Combining climatic, land cover and intermediate host species distribution data, Giraudoux et al. [6] have identified and mapped four spatially distinct types of regional transmission ecosystems for E. multilocularis in China, typified by the presence of one of the following small mammal ‘flagship’ species: Ellobius tancrei, Ochotona curzoniae, Lasiopodomys brandtii or Eospalax fontanierii. The Zhang/Puma study area typically belongs to the latter. Furthermore, using data from a community mass-screening screening on the Tibetan plateau, Giraudoux et al. [7] applied general additive linear models and found that human AE was spatially correlated with landscape features and climate which could confirm and predict human AE disease hotspots over a 200,000km2 region of the Tibetan plateau. Notably, four areas of human AE risk that were not in the data set used for training the model were predicted, including the Zhang/Puma AE disease hotspot [7]. This indicated that driving forces relying on climate, landscape and small mammal communities assessed on a large scale, combine to make transmission possible on the local scale (~400 km2). On this scale, in Zhang/Puma, four small mammal assemblages were identified in specific habitats of a deforestation gradient i.e. forest, shrubland and grassland, farmland and village [26]. All of the 10 species forming those small mammal assemblages were potential intermediate hosts for E. multilocularis. However, a significant association between human AE prevalence and land area under shrubland or grassland was found, indicating that, on that scale (~100 km2), the average density of small mammal intermediate hosts was likely a key factor explaining E. multilocularis transmission intensity [4]. The differences between small mammal habitats may be explained by the fact that mature forests, although richer in species biodiversity, cannot sustain high densities of small mammals on a large scale, compared to grassland and shrubby habitats. In these latter habitats, some rodent species prone to cyclicity can thrive at very large population densities (e.g. [45,46,44]), and importantly with comparatively low vegetation, they are more easily accessible to fox and dog predators [47]. Farmland, due to the high productivity of agrosystems can also provide temporary favourable habitats for high densities of small mammals, but tilling and seasonality of resources are often strong limiting factors. Furthermore, transmission of E. multilocularis depends also on the density of definitive hosts [48] and on human behaviour [18]. For instance, differences in living standards can explain why with similar prevalence of E. multilocularis in fox populations, human AE prevalence is more than 40 times lower in high endemicity areas of central Europe [49] compared to the eastern border of the Tibetan plateau [27]. Any change in one of those factors or set of factors can lead to change in transmission intensity.
During this retrospective >25 year overview in the south Gansu endemic zone, a number of economic and environmental changes were observed in the study area. Using Landsat Multispectral Scanner (MSS) and Thematic Mapper TM data, Danson et al. [50] confirmed that there had been an expansion of agricultural land from approximately 41% in 1975 to 58% in 1997. There was also an apparent reduction in secondary forest from 21% to 14%. In addition, there was generally a strong negative correlation between the area of secondary forest and the area of agricultural fields. There was also a strong positive correlation between the area of forest and the area of tree/shrub, showing the close relationship between forest and shrub clearance that is necessary to establish agricultural expansion. This was consistent with reports from local people in the Zhang/Puma area, and in particular the expansion of ploughed fields at the expense of semi-natural vegetation and wildlife. Bears (Ursus thibetanus) were still present in this region of south Gansu in the 1980s as testified by skins found in houses in the early 1990s. One of us (Craig, personal observation) observed one wild takin (Budorcas taxicolor) caught by people close to Han Chuan village (Zhang county) in 1991, and in 1994 local people were reporting that leopard (Panthera pardus) was still found some ten years before in the GuiQing Shan valley close to Cao Tan also in our study area. All those large mammal species were extinct by 1994 when landscape was then characterised by large areas of ploughed fields including on steep slopes (> 30°), by hills capped by bushes and grassland, and by the last timbering sites (Fig 2). However, considering the environmental issues raised by agricultural encroachement everywhere in China, the Chinese government enforced a forest and soil protection policy from the mid-1990s onwards. As a result strict controls led to a ban of tilling on steep slopes, and the setting aside of large areas that were abandoned and then colonized over some years by a succession of natural vegetation from grassland to shrub, or deliberately replanted mostly with larch (Larix sp.) (Fig 7). Another key event in the Zhang/Puma area, as already mentioned, was a dog and fox population crash in 1992–1993. We did not find a clear explanation for this event, the most likely being a side effect of agricultural expansion when massive rodent poisoning campaigns were carried out for crop protection, with deleterious non-intentional impacts on dogs and wildlife.
Considering the complexity of such transmission systems, human AE disease risk and the possible variations of the driving forces over years, it is difficult to assess retrospectively the sequence of events that drove changes in transmission intensity of E. multilocularis since the 1980s. This is further complicated by the generally slow (several years) but varying pathologic development of the parasite in humans, which generally prevents precise estimation of the time of human infection. However, we consider that even if major gaps exist in the decadal data sets for the Zhang/Puma area, we can focus on four probable key periods and tipping points (A-D in Fig 8) described below.
In summary, the current study is one of the few to attempt to explain the long-term transmission ecology and epidemiology of human alveolar echinococcosis in a highly endemic resource-poor region of Eurasia and in western China in particular. Early investigations in the 1990s, in the Zhang/Puma area of south Gansu, identified the dog as the key zoonotic risk and an important peri-domestic definitive host for E. multilocularis. In addition a key role was highlighted in this region for the existence of `risky`or `unhealthy`landscapes [41], characterised by grass and low shrub which was largely created by rapid deforestation and agricultural expansion from the 1970s, which subsequently enabled significant population growth of 2 or 3 susceptible intermediate host rodent species. Due to the long asymptomatic period of human AE the large numbers of cases detected between 1994–96 were probably infected 10–20 years before when high-risk landscapes, large dog populations and an increase in human settlements became a critical tipping point in the local transmission ecology. As further intensification of agriculture progressed in the 1990s the area of ploughed land increased and concurrently areas of grass/shrub habitats decreased. At this time, farmers also started application of newly available rodenticides, which killed rodent pests but also accidentally dogs by secondary poisoning. Together these factors resulted in a reduction in peri-domestic transmission of E. multilocularis and reduced zoonotic risk for many communities so that by 2006 almost all AE cases were older than 30 years. By 2015, although the dog population had recovered the parasite had not re-emerged in dogs, and together with a government dog dosing programme, peri-domestic transmission of E. multilocularis had virtually ceased. Considering the high biotic potential of E. multilocularis [51], this current low transmission situation may however be threatened (i) by reforestation programmes that may lead to appearance of temporary risky landscapes, (ii) by the translocation of dogs from the highly endemic region of the eastern Tibetan plateau, and (iii) by the cessation of PZQ treatments and control policy of dogs in villages [52].
Finally, although we observed that the mean survival period for those AE cases that had died before 2014 was only 8 years, we found that approximately 60% of AE cases were still alive in 2014 after 20 years, most probably in large part due to the application of successful community-based long-term ABZ therapy for AE cases.
|
10.1371/journal.pbio.1000256 | ZBED6, a Novel Transcription Factor Derived from a Domesticated DNA Transposon Regulates IGF2 Expression and Muscle Growth | A single nucleotide substitution in intron 3 of IGF2 in pigs abrogates a binding site for a repressor and leads to a 3-fold up-regulation of IGF2 in skeletal muscle. The mutation has major effects on muscle growth, size of the heart, and fat deposition. Here, we have identified the repressor and find that the protein, named ZBED6, is previously unknown, specific for placental mammals, and derived from an exapted DNA transposon. Silencing of Zbed6 in mouse C2C12 myoblasts affected Igf2 expression, cell proliferation, wound healing, and myotube formation. Chromatin immunoprecipitation (ChIP) sequencing using C2C12 cells identified about 2,500 ZBED6 binding sites in the genome, and the deduced consensus motif gave a perfect match with the established binding site in Igf2. Genes associated with ZBED6 binding sites showed a highly significant enrichment for certain Gene Ontology classifications, including development and transcriptional regulation. The phenotypic effects in mutant pigs and ZBED6-silenced C2C12 myoblasts, the extreme sequence conservation, its nucleolar localization, the broad tissue distribution, and the many target genes with essential biological functions suggest that ZBED6 is an important transcription factor in placental mammals, affecting development, cell proliferation, and growth.
| The molecular identification of genes and mutations affecting complex traits and disorders has proven to be very challenging in humans as well as in model organisms. These so-called quantitative traits arise from interactions between two or more genes and their environment, and can be mapped to their underlying genes via closely linked stretches of DNA called quantitative trait loci (QTL). Previously, we identified a single nucleotide substitution in a noncoding region of the insulin-like growth factor 2 gene (IGF2) in pigs that is underlying a major QTL affecting muscle growth, heart size, and fat deposition. The mutation disrupts interaction with an unknown nuclear protein acting as a repressor of IGF2 transcription. In the present study, we have isolated a zinc finger protein of unknown function and show that it regulates the expression of IGF2. The protein, which we named ZBED6, is encoded by a domesticated DNA transposon that was inserted into the genome prior to the radiation of placental mammals. ZBED6 is exclusive to placental mammals and highly conserved among species. Our functional characterization of ZBED6 shows that it has a broad tissue distribution and may affect the expression of thousands of other genes, besides IGF2, that control fundamental biological processes. We postulate that ZBED6 is an important transcription factor affecting development, cell proliferation, and growth in placental mammals.
| Strong selection for lean growth in the domestic pigs during the last 60 years has resulted in increased muscle growth and reduced fat deposition. Quantitative trait locus (QTL) mapping using an intercross between the European Wild Boar and Large White domestic pigs identified the most important locus that has responded to this selection pressure as a paternally expressed QTL colocalized with the gene for insulin-like growth factor 2 (IGF2) [1]. The allele present in the domestic pig increases muscle growth and heart size, and reduces subcutaneous fat deposition. The causative mutation for this QTL is a single nucleotide substitution in intron 3 of IGF2 [2]. The mutation is located in a CpG island that is well-conserved among mammals, and 16 bp including the mutated site showed 100% sequence identity among eight placental mammals. This quantitative trait nucleotide (QTN) is one of the rare examples in which a single base substitution underlying a complex trait has been identified and the mechanism of action is partially understood [2].
The IGF2 mutation, a G to A transition, disrupts the interaction with an unknown nuclear factor, a repressor, and leads to a 3-fold up-regulation of IGF2 expression in skeletal muscle. Elevated paternal expression from the mutant allele increases skeletal muscle mass and thus meat production by 3%–4%. The favorable allele has undergone a massive selective sweep and is close to fixation in pig populations widely used for meat production. Pigs carrying the favorable allele at the paternal chromosome show higher expression from the IGF2 P2, P3, and P4 promoters in skeletal and cardiac muscle, but not in liver. Importantly, this up-regulated IGF2 expression occurs postnatally, but not in fetal muscle. The mutation also up-regulates expression of an IGF2 antisense noncoding transcript with hitherto unknown function [3]. Thus, the binding of the repressor to its target site represses transcription from at least four promoters spread over a 4-kb region. Furthermore, the repressor binds its target site only when it is unmethylated [2].
Here, we report the identification of the repressor binding the IGF2 QTN site using mass spectrometry analysis after capturing nuclear proteins using a biotinylated oligonucleotide corresponding to the wild-type sequence. The protein, named ZBED6, is previously unknown and is encoded by an exapted DNA transposon. Elucidation of its functional role is shown by small interfering RNA (siRNA) and transient transfection using IGF2 P3 reporters.
Our previous electrophoretic mobility shift assay (EMSA), as well as transient transfection experiments with luciferase reporters, demonstrated that the unknown repressor is expressed in mouse C2C12 myoblasts [2]. To isolate the IGF2 repressor, we employed affinity capture using nuclear extracts from C2C12 cells and biotinylated oligonucleotides corresponding to the wild-type (q) and mutant (Q) sequence where only the former binds the repressor. Two different nuclear extracts were prepared using stable isotope labeling of amino acids in culture (SILAC) technique [4], in which the “heavy” extract proteins contained the stable-isotope–labeled amino acids lysine and arginine, whereas the “light” extract proteins contained the natural versions of these amino acids. The wild-type and mutant oligonucleotides were incubated with heavy and light extracts, respectively. Captured heavy and light proteins were mixed and separated with SDS-PAGE. Gel slices were digested with trypsin, and the resulting peptides were analyzed by liquid chromatography mass spectrometry (LCMS). The acquired spectra were searched against the RefSeq database containing mouse protein sequences to identify proteins present in the sample. Ratios of the amount of each protein enriched by the q and Q constructs were computed by comparing the mass spectral signals from the heavy and light versions of each identified peptide composing the protein.
The protein demonstrating the highest fold enrichment by q (9.0±1.2-fold; Figure 1A) corresponded to a transcript annotated as an alternative splice form of the poorly characterized Zc3h11a gene. ZC3H11A belongs to a large family of zinc finger proteins with 58 known members in mouse [5]. However, a closer examination revealed that the captured peptide is encoded by an intronless gene located in intron 1 of Zc3h11a (Figure 1B). The gene contains an open reading frame of more than 900 codons and encodes a protein with no sequence similarity to ZC3H11A. The encoded protein contains two BED domains and an hATC dimerization domain (Figure 1C). The BED domain was originally identified by a bioinformatic analysis using two chromatin-boundary-element-binding proteins from Drosophila, BEAF and DREF, as seeds for homology search [6]. We named our protein ZBED6 because it is the sixth mammalian protein with one or more BED domains (Figure 1C and 1D). ZBED6 is related to the hAT superfamily of DNA transposons, named after hobo from Drosophila, Activator from maize, and Tam3 from snapdragon [7]. For instance, the active Hermes transposase from the housefly contains an amino-terminal BED domain and a carboxyterminal hATC domain (Figure 1C).
The mammalian ZBED proteins showed high sequence divergence outside the BED and hATC domains and represent divergent members of the hAT superfamily (unpublished data). The two BED domains of ZBED6 are more closely related to each other than to any other mammalian BED sequence, implying an internal duplication after the integration in the genome (Figure S1). The ZBED6 protein is highly conserved among placental mammals, and in particular, the DNA-binding BED domains show 100%, or close to 100%, sequence identity among 26 different species, including the pig (Figure S2); the only sequence differences compared with the consensus sequence for placental mammals occured in species with low-coverage genome sequences and may therefore represent sequence errors.
To address whether ZBED6 is the bona fide repressor binding the QTN region in IGF2 intron 3, we produced a truncated mouse ZBED6 protein including the two DNA-binding BED domains and used this in EMSA with the q and Q oligonucleotides differing only at the QTN position. EMSA revealed a highly specific interaction with the wild-type q oligonucleotide, and a 100-fold excess of mutant Q oligonucleotide could not outcompete the interaction (Figure 1E). A polyclonal anti-ZBED6 antibody was developed by immunizing rabbits with this recombinant protein containing the two BED domains. A supershift was obtained when nuclear extracts from C2C12 cells were incubated with this anti-ZBED6 antibody, providing further biochemical evidence for ZBED6 as the elusive Igf2 repressor (Figure 1E).
Northern blot analysis (Figure 2A) and real-time PCR (RT-PCR) analysis (Figure 2B) showed that Zbed6, like Igf2, has a broad tissue distribution in mouse and is expressed in skeletal muscle, consistent with it being the Igf2 repressor. Northern blot analysis (Figure 2A) and RT-PCR amplification and sequencing (unpublished data) revealed that Zbed6 is coexpressed with Zc3h11a as an ∼13-kb splice variant of Zc3h11a, retaining the genomic region from exon 1 to exon 4 (including introns 1, 2, and 3 with Zbed6 located in intron 1) spliced to the remaining Zc3h11a exons (exons 5–18).
An examination of chromatin immunoprecipitation (ChIP) sequencing data for the human genome provided by the ENCODE consortium showed that there is one major RNA polymerase II binding site located just upstream of exon 1 of ZC3H11A, which apparently constitutes the common promoter for ZC3H11A and ZBED6 (Figure 2C). This is further supported by the perfect colocalization with a 5′ cap analysis gene expression (CAGE) tag. This promoter region contains binding sites for the Max, Myc, Fos, Jun, and NF-E2 transcription factors, all with a central role in regulating cell proliferation and associated with cancer development.
ZBED6 and ZC3H11A proteins were detected in the nuclei of postnatal muscle and brain tissue from mouse by using immunohistochemistry (Figure 3). ZBED6 expression was also confirmed in skeletal muscle (longissimus dorsi) cell nuclei from a 6-mo-old pig (Figure 3K–3N), i.e., at the age at which a highly significant effect of the pig IGF2 mutation is documented [1],[2].
Western blot analysis of proteins from mouse C2C12 cells revealed two different isoforms, denoted ZBED6a and ZBED6b, with apparent molecular weights of 122 and 116 kDa, respectively (Figure 4A). These isoforms most likely correspond to the use of two alternative start codons in the open reading frame of Zbed6, as demonstrated by comigration with recombinant proteins representing the two isoforms (Figure 4A).
Bioinformatic analysis revealed bipartite nuclear localization signals in two regions of ZBED6, amino acid residues 61–80 and 231–248. We used two different GFP-fusion constructs to confirm a nuclear localization in C2C12 cells (Figure 4B). The GFP-BED1/2x protein (residues 47–384) containing both regions with nuclear localization signals was primarily associated with granular structures in the nucleus, suggesting a nucleolar localization. This was confirmed by double immunofluorescence staining of C2C12 cells (Figure 4C) showing colocalization of endogenous ZBED6 and nucleophosmin 1, a well-known marker for the nucleolus [8]. However, ZBED6 did not show a complete localization to the nucleolus since some dispersed staining throughout the nucleus was evident (Figure 4C). Interestingly, the GFP-BED1/2y fusion protein (residues 90–384) also showed nuclear localization, but as an exclusion from the nucleolus (Figure 4B). This implies that the nuclear localization signal comprising residues 61–80 constitutes or contains a nucleolar localization signal. This lysine- and arginine-rich sequence (KKKRKKGLRIKGKRRRKKLI) is highly positively charged and resembles a nucleolar localization signal previously identified in the myogenic regulatory factor Myf5 [9]. The nucleolar localization of ZBED6 is of considerable interest in relation to the phenotype of IGF2-mutant pigs because the function of the nucleolus is associated with regulation of cell growth and proliferation [10].
Zbed6 was silenced in C2C12 cells by using siRNA to obtain further insight into its functional significance. Quantitative PCR revealed a >75% decrease in detected Zbed6 mRNA, and immunocytochemistry further confirmed a highly efficient silencing at the protein level (Figure 5A). Zbed6-silenced and control C2C12 cells were used to repeat our previously described luciferase assay including a reporter construct containing either the wild-type or mutant sequence of the QTN region fused with the IGF2 P3 promoter [2]. An assay based on C2C12 control cells transfected with scrambled oligonucleotide replicated our previous results since a construct containing the wild-type QTN region repressed luciferase expression in comparison with a construct containing P3 alone, whereas a construct including the mutant QTN region was associated with no or only minor repression (Figure 5B). In contrast, transfection experiments using three different silencing oligonucleotides directed against Zbed6 mRNA completely abolished repression with the wild-type q construct (Figure 5B). The interaction between ZBED6 and the QTN site in Igf2 intron 3 in mouse C2C12 cells was also validated by ChIP analysis using our anti-ZBED6 antibody, as a significant reduction in the enrichment of this region was observed after Zbed6 silencing (Figure 5C).
ZBED6 function was further investigated by specific gene silencing in C2C12 cells that were induced to differentiate by changing from growth to differentiation medium. Igf2 mRNA expression was low in both control and Zbed6-silenced cells the first days after differentiation was induced (Figure 5D). However, at day 6, Igf2 mRNA expression was significantly increased in silenced cells compared with controls (Figure 5D). This result is consistent with the increased IGF2 expression in skeletal muscle of pigs carrying the mutation at the ZBED6 target site in IGF2 intron 3 [2].
Silencing of Zbed6 was also accompanied by increased cell proliferation (Figure 5E), a faster formation of myotubes (Figure S3), and a faster wound healing process after scratching the surface of growing C2C12 cells (Figure 5F and 5G). A faster wound healing may reflect increased cell proliferation and/or increased cell migration. The fact that increased cell proliferation and faster wound healing were observed at day 3 after silencing, when there was not yet any significant effect on Igf2 expression (Figure 5D), implies that these effects are mediated by other target genes controlled by ZBED6.
A ChIP-sequencing experiment using our anti-ZBED6 antibody was performed to search for other downstream targets of ZBED6 besides Igf2. Mouse C2C12 cells were used for this experiment as ZBED6 is expressed in this cell line and interacts with the Igf2 QTN site, providing a positive control. The AB SOLiD technology was used to sequence the ChIP DNA fragments which resulted in the generation of 24 million reads aligned to the mouse genome. An analysis of these data revealed 2,499 peaks with a minimum of 15 overlapping extended reads (Table S1). As expected, the region in Igf2 corresponding to the porcine QTN site was among the most highly enriched regions (Figure 6A). De novo motif searches on both the full dataset and on subsets divided by enrichment levels gave a perfect match to the QTN site in Igf2 (Figure 6B). The result implies that the majority of the 2,499 peaks represent authentic target sites interacting with ZBED6 in C2C12 cells. Functional support for this is that the ZBED6 peaks often occurred in the vicinity of known transcription start sites (TSS), with approximately 50% of the peaks located within 5 kb of TSS (Figure 6C). Interestingly, there was a clear bias of binding sites to be located downstream of TSS, and a large proportion was found in intron 1, suggesting a role in transcriptional silencing. The IGF2 site in pigs is within a CpG island [2], and in this study, we found an enrichment of peaks close to CpG islands (Figure 6D), with as many as 28% of all peak maxima within a CpG island. In comparison only 16% of the locations of the 5′-GCTCGC-3′ consensus sequence occur in CpG islands in the mouse genome, indicating that other sequence elements are required for binding.
About 1,200 annotated genes were associated with one or more putative ZBED6 binding sites located within 5 kb of the gene, and 255 genes were associated to peaks of similar or higher enrichment as was seen for Igf2. We used this list of ∼1,200 genes to search for an enrichment of specific Gene Ontology classifications. The analysis showed that genes associated with development, regulation of biological processes, transcriptional regulation, cell differentiation, morphogenesis, neurogenesis, cell–cell signaling, and muscle development were all highly enriched in this list (Figure 6E). The list included 262 genes encoding transcription factors and certain families of transcription factors were particularly abundant among the putative ZBED6 targets (Figure 6F).
The ChIP-sequencing results indicated that ZBED6 takes part in the regulation of genes associated with basal functions in placental mammals. Therefore, we decided to perform an Ingenuity Pathways Analysis (Ingenuity Systems, http://www.ingenuity.com) to test whether putative ZBED6 targets are overrepresented among disease-associated genes in humans. This was accomplished by first downloading a table with all established mouse–human orthologs (Mouse Genome Informatics, http://www.informatics.jax.org/orthology.shtml) and then this list of human orthologs was used for the analysis. The results showed a highly significant association between our putative ZBED6 target genes and a number of diseases (Figure 6G). The most significant association was observed for developmental disorders, consistent with the Gene Ontology analysis, followed by cancer, cardiovascular disease, and neurological disease.
Our quest for the nuclear factor binding the QTN site in the porcine IGF2 gene has been driven by the vision that this factor must be important, since disruption of the interaction with one of its target sites alters body composition and promotes cardiac growth in pigs. Our previous experiments revealed a highly specific interaction between the factor and its target site in IGF2 [2], but it was not until we used the ultrasensitive SILAC technology that we could take advantage of this specificity and isolate ZBED6. The results presented in this study have conclusively demonstrated that ZBED6 is the bona fide repressor binding the QTN site in pig IGF2. This conclusion is based on (i) EMSA with recombinant ZBED6 protein, (ii) supershift of EMSA complex using an anti-ZBED6 antibody, (iii) abolishment of the repressor function in a luciferase assay after siRNA silencing, and (iv) ChIP data. The biological significance of ZBED6 was underscored in this study as siRNA silencing in C2C12 cells led to faster myotube formation and wound healing, and increased cell proliferation.
The difference between less complex eukaryotes like Caenorhabditis elegans and more complex eukaryotes, such as human, is related not to the number of protein-coding genes, but rather to the complexity of the gene regulatory networks. A large proportion of vertebrate genomes is composed of transposable elements, and their integration in the genome has contributed to the evolution of regulatory networks [11]. The majority of these transposable elements are retrotransposons, but 5%–10% are derived from DNA transposons. In the initial analysis of the first human genome assembly, Lander et al. [12] identified 47 human genes derived from transposable elements, as many as 43 of these are derived from DNA transposons, and in fact, one of the genes listed in Table 13 of the human genome paper corresponds to ZBED6. However, ZBED6 has never been appropriately annotated in any mammalian genome, despite the fact that it constitutes an ∼2,900-bp open reading frame and that part of the ZBED6 protein is extremely well conserved among placental mammals. A bioinformatic analysis of other vertebrate genomes did not reveal the presence of a functional ZBED6 gene outside the placental mammals. We found evidence for a nonfunctional ZBED6 sequence at the orthologous positions in the Platypus and opossum genomes, but these genomes did not contain an extended open reading frame for ZBED6 (unpublished data). This implies that the integration of ZBED6 happened before the divergence of the monotremes from the other mammals, but that the gene has been inactivated or lost in monotremes and marsupials. Thus, ZBED6 must have evolved its essential function in the time span after the split between marsupial and placental mammals, but before the radiation of different orders of placental mammals. An interesting topic for future research will be to reveal what advantage the development of ZBED6 as a new regulatory protein has provided to the placental mammals.
ZBED6 is an apparent example of a domesticated transposon that has lost its ability to transpose, because it occurs as a single copy gene at the same location in intron 1 of ZC3H11A in all placental mammals for which at least a partial genome sequence is available. ZBED6 has evolved an essential function in this group as implicated by the observation that the two DNA-binding BED domains (about 100 amino acids together) show near 100% amino acid identity across 26 placental mammals (Figure S2). The two BED domains in ZBED6 have apparently evolved by internal duplication because the two copies are more similar to each other than to any other mammalian BED sequence. The mechanism by which ZBED6 acts as a repressor remains to be determined. Chromatin remodeling is an obvious possibility since other members of the ZBED family have this function. For instance, the Drosophila Dref protein, a BED domain protein, is found in complex with the NURF chromatin remodeling complex and its human ortholog ZBED1 interacts with MI2, a chromatin remodeling factor, and PC2, a Polycomb group protein involved in heterochromatin formation [13]. The ability of ZBED6 to interact with chromatin and affect transcriptional regulation is most likely a function derived from the ancestral transposase. The nucleolar localization of ZBED6 (Figure 4C) suggests that it may mediate transcriptional silencing by moving the IGF2 locus and other targets to the nucleolus.
Our ChIP-sequencing experiment using mouse C2C12 myoblasts revealed more than 1,000 genes putatively regulated by ZBED6 in the mouse. We assume that a majority of these binding sites are true positives, because (i) we were able to generate a consensus binding motif (Figure 6B) with a perfect match with the established Igf2 binding site using both peaks with high and low enrichment levels, (ii) the majority of the binding sites occurred in the vicinity of TSS (Figure 6C), (iii) most of the binding sites occurred within or near CpG islands (Figure 6D), in line with the established binding site in Igf2, and (iv) the highly significant enrichment of certain Gene Ontology terms (Figure 6E). Thus, although we are certain that ZBED6 interacts with a majority of the genes listed in Table S1, transcriptome analysis will be required to assess the importance of ZBED6 for transcriptional regulation of these putative targets. In this context, it is worth emphasizing that disruption of the interaction between ZBED6 and the IGF2 QTN in pigs leads to a 3-fold up-regulation of IGF2 mRNA in skeletal muscle and altered body composition. Interestingly, our data indicated that the Zbed6 gene itself was bound by ZBED6 (Table S1), implying autoregulation of its expression.
About 1,200 of the ZBED6 binding sites in C2C12 cells occurred within 5 kb of the TSS of an annotated gene. The analysis of Gene Ontology terms associated with these genes revealed a highly significant enrichment for a number of important biological processes such as development, transcriptional regulation, and cell differentiation (Figure 6E). As many as 262 of the putative target genes encode transcription factors, 36 containing the homeobox domain, 26 members of the basic helix-loop-helix (bHLH) family, ten belonging to the FOX family, eight nuclear receptors, and seven members of the SOX family (Figure 6F). Many of these putative ZBED6 targets have a crucial role during development, and the results suggest that ZBED6 is an important regulator of development, cell proliferation, and growth. The binding of ZBED6 to its target sites in IGF2 leads to repression of IGF2 expression both in pig skeletal muscle [2] and in mouse C2C12 cells (this study). It may appear surprising that genes associated with neurogenesis were much more overrepresented in our peak list than genes associated with muscle development (Figure 6E), given the fact that we used mouse C2C12 myoblasts in this experiment. However, this pattern is expected if ZBED6 is primarily a repressor that silence genes not being part of the developmental program of a certain cell type. Another intriguing observation was the clear trend that ZBED6 preferentially binds downstream of the transcription start site which appears logical for a repressor (Figure 6C).
Igf2 is an imprinted gene, but our list of top hits did not indicate any overrepresentation of imprinted genes. In this respect, it is noteworthy that the QTN mutation in pigs does not result in loss of imprinting, but rather exclusively increases the transcription from the paternal Igf2 allele [2]. Thus, ZBED6 is unlikely to be a regulator of imprinting. However, one of the identified ZBED6 targets is the gene for growth factor receptor-bound protein (Grb10), also denoted Meg1 (maternally expressed gene 1), that is maternally expressed and a potent growth inhibitor [14]. GRB10 binds to the insulin receptor (INSR) and the IGF1 receptor (IGF1R), and inhibits the growth-promoting activities of insulin (INS), IGF1, and IGF2.
The list of genes associated with ZBED6 binding sites (Table S1) includes additional members, besides Igf2, of the IGF-signaling pathway, namely the genes for the IGF1 receptor (Igf1r), IGF2 binding protein 2 (Igf2bp2), IGF binding protein 3 (Igfbp3), and IGFBP-like protein 1 (Igfbpl1), suggesting that ZBED6 is an important regulator of IGF signaling. Furthermore, Grb10, as mentioned above, also takes part in the regulation of IGF signaling [14].
Genome Wide Association (GWA) studies have revealed a number of loci in the human genome associated with multifactorial disorders (Office of Population Genomics; http://www.genome.gov/26525384). An examination of this database showed that the region harboring ZBED6 is not one of the associated regions in any of the studies published so far. This means that the current GWA screens for different multifactorial disorders have not revealed any common ZBED6 variants associated with disease. This does not exclude the possibility of rare sequence polymorphism in ZBED6 affecting disease susceptibility in certain families. However, the ChIP-sequencing data indicated that ZBED6 has a fundamental role in regulating several biological processes. Mutations altering ZBED6 function or expression may therefore have severe pleiotropic effects through the many downstream targets. This notion is consistent with the near 100% conservation of the BED domains among placental mammals.
Our current model for ZBED6 function is summarized in Figure 7. Our data on the IGF2 locus indicate that ZBED6 acts primarily as a repressor, likely with a modulating effect, although it is fully possible that it acts as a transcriptional activator under some circumstances.
First, germline or somatic mutations at target sites may lead to transcriptional up-regulation as demonstrated for the IGF2 locus in pigs [2]. Our findings that the mammalian genome contains thousands of putative ZBED6 targets and that these are enriched among genes associated with disease suggest that sequence polymorphism at ZBED6 target sites may contribute significantly to variation in disease susceptibility in humans. Furthermore, the ZBED6 binding motif contains a CpG dinucleotide so we expect to find genetic polymorphisms as CpG sites are associated with a high rate of C→T and G→A transitions [15], as exemplified by the pig IGF2 QTN. Gain or loss of ZBED6 binding sites may also have contributed to phenotypic evolution in placental mammals.
Second, our data suggest that ZBED6 targets can be released from repression by epigenetic activation. This is implied by the finding that EMSA using an oligonucleotide with a methylated CpG site was not bound by ZBED6 [2]. Interestingly, the pig QTN had no effect on IGF2 transcription in liver, and the QTN region was shown to be methylated in this tissue, whereas it was undermethylated in skeletal muscle where the QTN had a drastic effect on IGF2 expression [2]. Thus, epigenetic regulation of the access of ZBED6 to its target sites may play an important role during development and cell differentiation.
Third, ZBED6 targets can be released from repression by down-regulation of ZBED6 expression, as demonstrated by siRNA experiments in the present study. Finally, loss-of-function mutations in ZBED6 are expected to up-regulate many target genes. Our finding that Zbed6 silencing in C2C12 cells leads to faster cell proliferation and wound healing combined with the identification of a large number of cancer-associated downstream targets by ChIP sequencing implies that further studies of ZBED6 function is of considerable interest for tumor biology.
Data reported here suggest that ZBED6 has an essential role in a number of crucial gene regulatory networks. Thus, the discovery of ZBED6 opens up many avenues for research that may have profound implications for human medicine.
Two sets of murine C2C12 myoblast cells (ATCC CRL-1772) were used. Cultures were grown according to standard cell-culture procedures in SILAC-light and SILAC-heavy labeled Dulbecco's Modified Eagle Medium (DMEM, ThermoFisher) containing 10% dialyzed fetal bovine serum (FBS) and 13C6-l-Arg and 13C6,15N2-l-Lys in the “heavy” formulation, see [16] for more details. “Light” and “heavy” nuclear extracts were prepared using a commercially available kit (ActiveMotif). The (+)-strand sequences of the wild-type q and mutant Q oligonucleotides were as follows (QTN underscored):
IGF2-q: 5′-Biotin-GATCCTTCGCCTAGGCTCGCAGCGCGGGAGCGA-3′
IGF2-Q: 5′-Biotin-GATCCTTCGCCTAGGCTCACAGCGCGGGAGCGA-3′
The complementary (−)-strands were also synthesized, and the pairs were annealed prior to use. One picomole of ds-q oligonucleotide was mixed with 1.4 mg of heavy nuclear extract and 1 pmol of ds-Q oligonucleotide was mixed with 1.4 mg of light nuclear extract in binding buffer (50 mM Tris [pH 8], 150 mM NaCl, 0.25 mM EDTA, 0.5 mM DTT, 0.1% Tween-20, 0.5 mg/ml BSA, 200 µg/ml poly-dI/dC) in a total volume of 700 µl. Mixtures were incubated for 45 min at room temperature (RT) on a rotator. Ten microliters of streptavidin-coated magnetic beads (Dynal) was added to each tube, and the mixture was further incubated for 30 min at RT on a rotator. Beads were spun for 5 s at 1,000g and then captured using a magnetic pull-down system. Beads were washed 3×700 µl in binding buffer without poly-dI/dC and then 4×700 µl in binding buffer without poly-dI/dC or BSA. The supernatant was discarded, and proteins were eluted by boiling in Laemmli buffer (+10 mM DTT). Protein eluates were mixed 1∶1 by volume.
Proteins were separated on an SDS-PAGE gel (4%–12% NuPage), stained with Coomassie Blue, and the entire lane was cut into ∼20 bands. Each band was reduced, alkylated, digested with trypsin according to standard proteomics practices [17], and the resulting peptides were analyzed by LCMS on an Orbitrap (ThermoFisher) mass spectrometer as described [18]. Database searching was performed by Mascot against the REFSEQ database of mouse proteins as of June 2006. SILAC quantification was performed using msQuant [19].
RT-PCR, using poly-A–enriched RNA extracted from C2C12 cells, was used to obtain nucleotides 1–2,943 and 139–2,943 of mouse Zbed6 transcripts (named Zbed6a and Zbed6b, respectively; EMBL Bank accession number FM882123); the two constructs begin at the two alternative start codons. These constructs, containing a Kozak sequence for efficient initiation of translation, were cloned into pcDNA3 (Invitrogen) and verified by DNA sequencing.
PCR was used to subclone a fragment encoding the two BED domains, amino acid residues 90–384 of ZBED6a, into pGEX-5X-3 (GE Healthcare). GST and GST-BED1/2 fusion protein was purified from BL21(DE3)pLysS bacteria using gluthathione Sepharose 4B or GSTrap FF columns (GE Healthcare), according to the manufacturer's instructions.
Polyclonal antibody production was performed by Agrisera AB. Shortly, GST-BED1/2 was used to immunize one rabbit. Polyclonal anti-ZBED6 antibodies were affinity purified by first passing serum over a HiTrap NHS-activated HP column (GE Healthcare) coupled with GST, whereafter the flow-through was applied to a column coupled with GST-BED1/2. Anti-ZBED6 antibodies were eluted with 0.2 M glycine (pH 2.5) and dialyzed against 20 mM HEPES (pH 7.4) and 150 mM NaCl.
Nuclear extracts from C2C12 cells were prepared using the NucBuster Protein Extraction kit (Novagen). EMSAs were performed as previously described with minor modifications [2]. The following oligonucleotides were annealed in 1× NEB2 buffer (NEB) q/Q-fwd AGATCCTTCGCCTAGGCTC(G/A)CAGCGCGGGAGCGA and q/Q-rev AGATCTCGCTCCCGCGCTG(C/T)GAGCCTAGGCGAAG. A total of 20 ng of purified GST-BED1/2 protein or 10 µg of C2C12 nuclear extracts were preincubated on ice for 20 min in binding buffer (15 mM Hepes-KOH [pH 7.65] at RT, 30.1 mM KCl, 2 mM MgCl2, 0.1 mM EDTA, 0.063% NP-40, 7.5% Glycerol, 1.3 mM dithiothreitol, 2 mM spermidine, 0.1 µg/µl Poly(dI-dC)•Poly(dI-dC)). Competition reactions were supplemented with 4 pmol (100-fold molar excess) unlabelled ds-oligonucleotide. After the addition of 40 fmol end-labeled 32P-dCTP ds-oligonucleotide, reactions were incubated at RT for 30 min. In EMSAs including supershift reaction, incubation for 20 min at RT preceded the addition of 2 µl of purified polyclonal anti-ZBED6 antibody (0.13 µg/µl) and incubation then continued for an additional 20 min at RT. Complexes were separated on a 1.5-mm 5% native 29∶1 polyacrylamide gel in 0.5× TBE at 70V for 3–4 h.
A mouse multiple-tissue Northern blot panel 4 (MN-MT-1) and a mouse developmental tissue skeletal muscle panel (MN-102-D from Zyagen) was used. Partial Zbed6 and Zc3h11a coding sequences were cloned into a vector and probe template amplified by PCR including either SP6 or T7 sequence from the vector. A purified probe template (200 ng) was used for 32P-labeled RNA probe synthesis using the MAXIscript Kit (Ambion). Mouse ß-actin (Actb) was amplified by PCR from C2C12 cDNA, sequenced, and used as template for 32P-labeled DNA probe synthesis using the Amersham Megaprime DNA labeling system (GE Healthcare). Hybridizations were done at 68°C (RNA probe) or 42°C (DNA probe) using the ULTRAhyb buffer (Ambion) followed by washes in 2× SSC+0.1% SDS and 0.1× SSC+0.1% SDS at 68°C (RNA probes) or 60°C (DNA probe). Autoradiographs were exposed for a few hours to several days.
Mouse and pig tissue were fixed in 4% paraformaldehyde (PFA) in phosphate buffered saline (PBS) for 2–4 h on ice, cryoprotected in 30% sucrose, followed by embedding in Tissue-Tek O.C.T. compound (A/S Chemi-Teknikk) and cryostat sectioning. Fluorescent immunohistochemistry was performed on cryosections from mouse hind limb muscle postnatal day 3 (p3), mouse brain (p22), and pig muscle (longissimus dorsi) (6 mo). Rabbit antibodies against ZC3H11A (A301-525A, Bethyl Laboratories) and ZBED6 were diluted to 0.2 µg/ml in PBS containing 1× blocking reagent (Roche), 0.3% Triton X-100 (Sigma), and incubated at 4°C overnight. In the case of the anti-ZC3H11A antibody, a commercially available blocking peptide (Bethyl Laboratories) was used at 2.0 µg/ml and incubated with the anti-ZC3H11A antibody for 2 h at RT prior to application on tissue. The following day, slides were washed in PBS and incubated with a secondary anti-rabbit IgG antibody conjugated to Alexa 594 (Invitrogen) and DAPI (Sigma) in PBS for 1 h at RT. Slides were washed in PBS and mounted with 2.5% DABCO (Sigma) in glycerol containing 0.1M Tris (pH 8.6). For pig muscle tissue, sections were also incubated with Alexa 488–conjugated α-bungarotoxin (1∶1,000, Invitrogen). Staining was analyzed with a fluorescence microscope (Olympus BX61WI) or a confocal microscope (Zeiss LSM 510 META). Images were adjusted for contrast and brightness in Adobe Photoshop (Adobe).
The C2C12 mouse myoblast cell line (ATCC-CLR-1772) was cultured at 37°C in a humidified atmosphere of 5% CO2 using DMEM (ATCC-30-2002) supplemented with 10% FBS (Invitrogen) and 1× Antibiotic-Antimycotic solution (Invitrogen). The cultures were split every 2 to 3 d. C2C12 cells were differentiated by growing the cells in differentiation media, DMEM with 2% horse serum.
BED1/2x and BED1/2y (corresponding to amino acid residues 47–384 and 90–384 of ZBED6, respectively) were cloned into pcDNA3 containing an N-terminal enhanced green fluorescent protein (eGFP) coding sequence. A total of 5×104 cells were plated the day before transfection in 12-well plates. Two µg of DNA of either GFP, GFP-BED1/2x, or GFP-BED1/2y was transfected using 6 µl of Lipofectamine 2000 reagent (Invitrogen) in Opti-MEM (Gibco). Medium was changed to growth medium without antibiotics 4 h posttransfection, and cells were analyzed the following day. Photographs were captured using an epifluorescence-equipped Nikon Eclipse TS100 microscope, a superplan fluor 60× objective, and a Nikon D300 body.
Cells were seeded on coverslips, in six-well plates, overnight to be around 50%–70% confluent the following day. The coverslips were fixed by 4% formaldehyde for 10 min at RT in PBS (pH 7.4), followed by permeabilization of the cellular membrane with 0.2% Triton X-100 on ice for 10 min, blocked with 5% FBS for 30 min at RT in PBS, and treated with mixture (2 µg/ml each) of rabbit anti-ZBED6 and mouse anti-nucleophosmin 1 (SIGMA, catalog no. B0556) for 1 h at RT. The cells were then washed four times with PBS to remove unbound antibodies and then treated with mixture (20 µg/ml each) of Alexa Flour 488–labeled goat anti-mouse and Texas Red-labeled goat anti-rabbit secondary antibody for 1h at RT. Cells were washed four times with PBS and mounted with Fluoromount G on an objective slide. The fluorescence was analyzed using LSM 510 confocal microscopy (Zeiss).
A total of 5×104 C2C12 cells or 30%–50% confluent cells were transfected with 50 pmol of negative control siRNAs (Ambion) or ZBED6 siRNAs (Ambion) with 6 µl of Lipofectamine 2000 Reagent (Invitrogen) in 1.5 ml of Opti-MEM I (Invitrogen) per well in 6-well plates. The following pooled Silencer Select siRNA sequences (Ambion) were used to silence Zbed6 expression in C2C12 cells: duplex 1 sense, 5′-CUUCAACACUUCAACGACAtt-3′; duplex 2 sense, 5′-UGUGGUACAUGCAAUCAAAtt-3′; duplex 3 sense, 5′-GGGCUGUUGCCAACAAAGAtt-3′; the dinucleotide tt (indicated in lower case letters) was added to all oligonucleotides to improve the stability of siRNA after transfection. After 24 h, medium was changed to fresh DMEM with 10% FBS. Biological triplicates were used for each siRNA treatment.
Silencing was performed 2 d prior to transient transfection with luciferase reporters. Previously described [2] constructs containing the porcine IGF2 QTN region, the porcine IGF2 P3 promoter and the firefly luciferase reporter gene (P3, q+P3 and Q+P3) were used. Zbed6-silenced C2C12 cells or negative control siRNA treated cells grown in 12-well plates were transfected with a total of 2 µg of DNA and 6 µl of Lipofectamine 2000 Reagent (Invitrogen). One microgram of firefly luciferase construct and 20 ng of Renilla luciferase vector as control (ph-RG, Promega) and empty pcDNA3 vector (Invitrogen) up to 2 µg of DNA were used. Transfections were performed in opti-MEM (Gibco), and medium changed to growth medium (DMEM supplemented with 10% FBS) after 4 h. Cells were harvested 24 h posttransfection, and firefly and Renilla luciferase activities were measured using the Dual-Luciferase Reporter Assay System (Promega) and an Infinite M200 luminometer (Tecan).
At 48 h post-siRNA transfection, the cells were incubated 3–4 h with medium containing 10% Alamar blue (Invitrogen). The reduction of Alamar blue was measured on a Tecan Sunrise Absorbance Plate Reader (Oxidized/Reduced: 600/570 nm).
At 72 h post-siRNA transfection, cells reached almost confluence, and medium was replaced with fresh DMEM supplemented with 0.1% FBS. A surface wound was created by scraping a pipette tip across the confluent cell monolayer. Twenty-four hours after scraping, the number of cells in the scratch was counted and cells treated with negative control siRNAs and Zbed6 siRNAs were compared. Statistical analysis was performed using a Student t-test.
RNA was isolated from tissue samples from six C57BL/6 mice (three males and three females) either by RNeasy mini kit (Qiagen) or acidic phenol extraction as described [20]. RNA from C2C12 cell samples was isolated using the RNeasy Mini Kit (Qiagen), then all samples were subjected to reverse transcription using cDNA high capacity kit (Applied Biosystems). mRNA transcripts were measured by quantitative PCR analysis using TaqMan Gene Expression master mix (Applied Biosystems) on a 7900HT Fast RT-PCR System (Applied Biosystems); probes and primers used are given in Table S2. Data were analyzed with a threshold set in the linear range of amplification, based on a standard curve of serial 10-fold dilutions for each primer set. The Zbed6 data was normalized to the level of cDNA from two endogenous housekeeping genes (GAPDH and 18S rRNA) and plotted as mean fold change (±s.e.m.). Statistical analysis was performed using a Student t-test.
Cells were trypsinized by 0.25% trypsin EDTA (Invitrogen). After blocking with 10% FBS and washing twice in PBS (Invitrogen), 2 to 3 million cells were resuspended in 1 ml of PBS with 20% FBS and centrifuged as cytospins for 3 min at 800 rpm. The cytospins were dried at RT overnight and then fixed with acetone for 10 min and hybridized with 250 µl of anti-ZBED6 antibody (0.2 µg/ml) for 20 min, followed by staining protocol described by Human Protein Atlas (http://www.proteinatlas.org).
Normal ChIP was performed as previously described [21], whereas ChIP sequencing was performed as follows. Chromatin was immunoprecipitated from approximately 107 subconfluent C2C12 cells. Protein-DNA cross-links were made in RT for 10 min with 0.37% formaldehyde, and cells were lysed in RIPA buffer (25 mM Tris-HCl [pH 7.6], 150 mM NaCl, 1% NP-40, 1% sodium deoxycholate, 0.1% SDS). Sonication was done to generate DNA fragments mainly in the 100–300-bp range using a BioRuptor (Diagenode) at highest settings for 30 cycles of 30 s. Incubation with 5 µg of antibody was done overnight, and protein-G agarose beads (Roche) were used to pull down the protein-DNA complexes. After washing and elution, proteins were degraded using proteinase K, and DNA was extracted with phenol-chloroform and precipitated using EtOH with the addition of glycogen. DNA from six replicates was pooled and purified using Qiagen MinElute columns. In order to recover as much as possible of the enriched DNA for sequencing, additional sonication of the ChIP DNA was performed using a Covaris instrument to get optimal fragment sizes. Library construction was done according to the manufacturer's protocol (AB SOLiD v3.0 fragment library) with eight rounds of amplification. Sequencing was done on a quarter of a slide and gave 58 million 50-bp reads. Alignment was done in two steps, first the AB pipeline (mapreads) was used to align full-length reads with up to four mismatches, and subsequently, the remaining reads was truncated to 35 bases and realigned with four mismatches using ZOOM! [22]. This gave 18+6.5 million uniquely placed alignments at 23 million unique positions. Each read was extended to 200 bases and overlap profiles were calculated to identify regions of enrichment. FindPeaks 3.1.92 [23] was used to estimate the false discovery rate (FDR), giving a probability of <0.001 at 15 overlapping fragments. Peaks falling within 100 kb of centromeric gaps or overlapping with Satellite and rRNA repeats were removed to reduce nonrandom false-positive peak calls. This gave 2,499 peaks, which were then associated with the nearest TSS and CpG island (Table S1). We noted that the additional sonication done before sequencing gave larger regions with enrichment compared to what is normally seen in size-selected ChIP-seq dataset, therefore, we additionally filtered the peaks to contain only the highest scoring peak within 5 kb before performing de novo motif finding using BioProspector [24]. We searched for motifs of length 8 bp both in the full list of peaks and in subsets of peaks divided by enrichment and location, and used mouse sequences 1 kb upstream of transcription start sites (TSS) as background. We used the DAVID software for Gene Ontology analysis [25],[26].
Information on genomes is available at http://www.genome.ucsc.edu.
|
10.1371/journal.pgen.1003996 | NCYM, a Cis-Antisense Gene of MYCN, Encodes a De Novo Evolved Protein That Inhibits GSK3β Resulting in the Stabilization of MYCN in Human Neuroblastomas | The rearrangement of pre-existing genes has long been thought of as the major mode of new gene generation. Recently, de novo gene birth from non-genic DNA was found to be an alternative mechanism to generate novel protein-coding genes. However, its functional role in human disease remains largely unknown. Here we show that NCYM, a cis-antisense gene of the MYCN oncogene, initially thought to be a large non-coding RNA, encodes a de novo evolved protein regulating the pathogenesis of human cancers, particularly neuroblastoma. The NCYM gene is evolutionally conserved only in the taxonomic group containing humans and chimpanzees. In primary human neuroblastomas, NCYM is 100% co-amplified and co-expressed with MYCN, and NCYM mRNA expression is associated with poor clinical outcome. MYCN directly transactivates both NCYM and MYCN mRNA, whereas NCYM stabilizes MYCN protein by inhibiting the activity of GSK3β, a kinase that promotes MYCN degradation. In contrast to MYCN transgenic mice, neuroblastomas in MYCN/NCYM double transgenic mice were frequently accompanied by distant metastases, behavior reminiscent of human neuroblastomas with MYCN amplification. The NCYM protein also interacts with GSK3β, thereby stabilizing the MYCN protein in the tumors of the MYCN/NCYM double transgenic mice. Thus, these results suggest that GSK3β inhibition by NCYM stabilizes the MYCN protein both in vitro and in vivo. Furthermore, the survival of MYCN transgenic mice bearing neuroblastoma was improved by treatment with NVP-BEZ235, a dual PI3K/mTOR inhibitor shown to destabilize MYCN via GSK3β activation. In contrast, tumors caused in MYCN/NCYM double transgenic mice showed chemo-resistance to the drug. Collectively, our results show that NCYM is the first de novo evolved protein known to act as an oncopromoting factor in human cancer, and suggest that de novo evolved proteins may functionally characterize human disease.
| The MYCN oncogene has a central role in the genesis and progression of neuroblastomas, and its amplification is associated with an unfavorable prognosis. We have found that NCYM, a MYCN cis-antisense RNA, is translated in humans to a de novo evolved protein. NCYM inhibits GSK3β to stabilize MYCN, whereas MYCN induces NCYM transcription. The positive feedback regulation formed in the MYCN/NCYM-amplified tumors promotes the aggressive nature of human neuroblastoma. MYCN transgenic mice, which express human MYCN in sympathoadrenal tissues, spontaneously develop neuroblastomas. However, unlike human neuroblastoma, distant metastasis is infrequent. We established MYCN/NCYM double transgenic mice as a new animal model for studying neuroblastoma pathogenesis. We found that NCYM expression promoted both the metastasis and chemo-resistance of the neuroblastomas formed in the double transgenic mice. These results demonstrate that NCYM may be a potential target for developing novel therapeutic tools against high-risk neuroblastomas in humans, and that the MYCN/NCYM double transgenic mouse may be a suitable model for the screening of these new drugs.
| Gene evolution has long been thought to arise from pre-existing genes through duplication or rearrangement followed by rapid divergence [1]–[5]. De novo gene birth from non-coding genomic regions has been generally believed to be exceptionally rare [1]. However, recent studies using genome-wide analyses have suggested the presence of a large number of de novo evolved genes in some species [3], [5]–[11], including primates [12]–[17]. Studies in yeast revealed that the proteins produced from de novo genes were not insignificant polypeptides but functional proteins [6], [7] and that de novo gene birth could be a major mechanism of new gene generation [6]. In multicellular organisms, however, the functions of de novo evolved proteins have been poorly characterized [3], [15], and thus their pathophysiological significance has remained elusive. Therefore, it is still unclear whether de novo gene birth is a general mechanism throughout evolution for the creation of functional protein-coding genes.
Neuroblastoma is one of the most common solid tumors in children. It originates from the neuronal precursor cells of the sympathoadrenal lineage of the neural crest [18]. Its clinical behavior is enigmatic; the tumors in patients of less than one year of age often regress spontaneously, whereas the tumors detected in patients over one year of age are usually aggressive and eventually cause the patient's death despite intensive multimodality therapies [18]. The MYCN oncogene is frequently amplified in those tumors that occur in patients who are over one year of age at diagnosis [19], [20]. Transgenic mice expressing human MYCN in sympathoadrenal tissues spontaneously develop neuroblastomas [21], suggesting that MYCN alone can initiate tumorigenesis and promote tumor growth. However, unlike human neuroblastoma, its distant metastasis is infrequent. Furthermore, in human neuroblastomas without MYCN amplification, MYCN mRNA expression levels do not correlate with the prognosis of the patients [22], [23], suggesting that additional events might contribute to the acquisition of increased aggressiveness. We focused on NCYM as a candidate gene that promotes the aggressiveness of MYCN-amplified neuroblastomas. NCYM is a cis-antisense gene of MYCN [24], [25] and is co-amplified with MYCN in human neuroblastoma cells. NCYM is transcribed in the opposite direction to MYCN, starting from intron 1 of the MYCN gene (Figure 1A), and it has remained unclear for a long time whether the gene encodes a functional protein [24], [26]. In this study, we have found that NCYM is indeed a functional protein that regulates MYCN function in human, but not mouse, neuroblastoma.
We first analyzed the genomic sequence of NCYM in various species and found that in humans and chimpanzees the potential NCYM protein is composed of 109 amino acids (Figure 1B, Figure S1). We next searched for paralogs and orthologs of the human NCYM protein among other animals using the Basic Local Alignment Search Tool (BLAST) databases with an E-value threshold of 10−3. We did not find any paralogs, but identified orthologs for a probable NCYM protein in olive baboons, chimpanzees and pigmy chimpanzees. From here on, we focused on the NCYM gene of the hominidae to investigate the function of the protein product. The evolutionary rates between the indicated species suggest that the coding sequence of NCYM gene was exposed to positive selection in humans and chimpanzees (Figure 1C), and the amino acid frequencies in these species were significantly different from the uniform usage of amino acids (P<0.001; Figure S2). We next raised an antibody against the putative human NCYM protein, and identified a 12 to 15 kDa protein in human neuroblastoma cells which mainly localized to nuclei in MYCN-amplified neuroblastoma cells (Figure S3, Figure S4). The NCYM protein was expressed in a variety of normal human tissues, including the neuronal cells of the cerebrum and cerebellum, spermatocytes of the testis, pancreatic cells and also the heart (Figure S5). NCYM was also localized in both the nucleus and cytoplasm in these cells (Figure S5A–D). NCYM was expressed in both primary and metastatic human neuroblastomas (Figure 1D, Figure S5E and F), and was co-expressed with the MYCN protein in cells of human neuroblastomas (Figure 1D and E) and the neuronal cells of the human cerebrum (Figure 1F and G). It was also co-expressed with the MYCN protein in some primary human cancers, including thyroid cancer (Figure S6). Thus, the NCYM protein is a de novo evolved gene product and is endogenously expressed in both normal human tissues and cancers.
We next examined the prognostic significance of NCYM mRNA expression in human neuroblastoma. The NCYM gene was co-amplified with the MYCN gene in all the cell lines and primary neuroblastomas we examined (Figure S7). NCYM expression levels were significantly correlated with that of MYCN in primary neuroblastomas (n = 106, P = 4.69×10−16; Figure 2A) and in the tumors with a single copy of MYCN (n = 86, P = 1.11×10−13; Figure 2B). In addition, high levels of NCYM mRNA expression were significantly associated with unfavorable prognostic factors (P<0.05, Table S1) and a poor outcome (P = 3.70×10−5; Figure 2C), similar to that for MYCN mRNA expression (P<0.05; Table S1 and P = 2.31×10−5; Figure 2D). Interestingly, high levels of NCYM mRNA expression were also significantly correlated with poor outcome in those patients diagnosed at over one year of age without MYCN amplification (n = 45, P = 0.0375; Figure S8A) whereas those of MYCN did not correlate with the prognoses (n = 45, P = 0.144; Figure S8B). Multivariate analysis of 106 primary neuroblastomas showed, as expected, that NCYM mRNA expression is not an independent prognostic factor from expression and amplification of MYCN (Table S2). However, it is an independent prognostic factor from age at diagnosis, stage and TrkA expression.
The co-amplification and co-expression of NCYM and MYCN in human primary neuroblastomas prompted us to investigate the functional interaction between NCYM and MYCN. Previously we have reported that MYCN directly targets its own expression in neuroblastoma cell lines [27]. Because the promoter region of the NCYM gene is localized within intron 1 of MYCN (Figure S9A), we examined whether MYCN regulates NCYM transcription. Overexpression of MYCN in human neuroblastoma cells induced NCYM mRNA expression (Figure 3A), whereas shRNA-mediated knockdown of MYCN downregulated endogenous NCYM mRNA levels (Figure 3B). MYC overexpression did not induce either MYCN or NCYM expression (Figure S9B). However, MYCN overexpression did enhance NCYM promoter activity in a dose-dependent manner (Figure 3C), suggesting that MYCN, but not MYC, activates the transcription of NCYM. Putative E-boxes exist in intron 1 of the MYCN gene; however, it is unclear whether they are responsible for this feedback regulation. We therefore generated constructs containing different lengths of the MYCN intron 1 region and performed luciferase assays to identify the MYCN-responsive region (Figure S9C). MYCN enhances its own promoter activity in a dose-dependent manner when co-transfected with reporter plasmids containing the NCYM promoter region (from +1073 to +1312). However, when co-transfected with plasmids without this NCYM promoter region, MYCN positive autoregulation was diminished. Within this region, there is a putative E-box located just 2 base pairs upstream from the transcription start site of the NCYM gene (Figure S10A). We generated constructs containing the NCYM promoter region comprising either a wild-type or a mutant E-box. Overexpression of MYCN enhanced NCYM wild-type promoter activity, but mutation of the E-box diminished its activation (Figure S10C). MYC overexpression did not activate either of the NCYM promoter constructs (Figure S10B and C). Therefore, these results indicate that MYCN enhances NCYM promoter activity in an E-box-dependent manner. MYC, however, is not involved in NCYM transcription.
We next investigated the function of NCYM in neuroblastoma cells. NCYM overexpression induced MYCN protein levels (Figure 3D, left panel; Figure S11A), but had no effect on the mRNA levels of MYCN (Figure 3D, right panel; Figure S11A). Consistent with these results, shRNA-mediated knockdown of NCYM significantly downregulated the amount of MYCN protein without affecting the level of MYCN mRNA expression (Figure 3E). In addition, knockdown of NCYM decreased the stability of the MYCN protein (Figure S11B). This NCYM knockdown-mediated destabilization of MYCN could be inhibited using the proteasome inhibitor MG132 (Figure S11C). It is known that the stability of the MYCN protein is regulated by a series of phosphorylation and ubiquitination events that are required for its recognition by the proteasome [28]. CDK1/Cyclin B1 phosphorylates MYCN at serine 62: the mono-phosphorylated MYCN is then recognized by GSK3β and subsequently phosphorylated at threonine 58, leading to its proteasome-dependent protein degradation after an E3-mediated polyubiquitination [28], [29]. Therefore, using immunoprecipitation, we next searched for factors interacting with NCYM that are able to induce MYCN stabilization, and found that NCYM forms a complex with MYCN and GSK3β in CHP134 cells (Figure 3F and G). In addition, purified NCYM was capable of interacting with purified GSK3β and MYCN in vitro (Figure 3H). To examine the effect of NCYM on GSK3β-mediated phosphorylation of MYCN, we performed an in vitro kinase assay (Figure 3I). NCYM protein inhibited the phosphorylation of MYCN. Because the purified NCYM protein is not a substrate of GSK3β (Figure S12), it is unlikely that NCYM competes with MYCN for GSK3β as a substrate. Taken together these results suggest that the NCYM protein inhibits GSK3β-mediated MYCN phosphorylation and stabilizes the MYCN protein in vitro.
It has been reported that MYCN knockdown decreases cell proliferation and induces apoptosis and/or differentiation in MYCN-amplified neuroblastoma cells [30]. Therefore, we next investigated the functional role of NCYM in these cells (Figure S13 and S14). We performed NCYM knockdown in BE (2)-C, CHP134, SK-N-AS and SH-SY5Y human neuroblastoma cells. SK-N-AS and SH-SY5Y cells are MYCN-single copy but have a high expression of MYC, while BE (2)-C and CHP134 are cell lines with MYCN-amplification and hence have a high expression of MYCN and NCYM (Figure S13A). NCYM knockdown did not affect the survival of the MYCN-single neuroblastoma cell lines, but promoted massive apoptosis of the MYCN-amplified neuroblastoma cells (Figure S13B and C). In addition, in BE (2)-C cells, NCYM knockdown was found to inhibit cell proliferation and invasion (Figure S14B and D). These results suggest that NCYM promotes the survival and aggressiveness of MYCN-amplified neuroblastoma cells.
To assess the function of NCYM in vivo, we generated transgenic mice expressing the human NCYM gene under the control of the rat tyrosine hydroxylase (TH) promoter (Figure S15A and B). In addition, we made double transgenic mice carrying both the human MYCN and NCYM genes. NCYM Tg/+ mice were mated with MYCN Tg/+ NCYM Tg/+ mice, and 83 descendants were observed for 200 days (Figure S15C and D). None of the NCYM transgenic mice of the 129+ter/SVJ background developed neuroblastoma (Figure S15D), suggesting that NCYM overexpression alone is not sufficient to initiate neuroblastoma in vivo. Although tumor formation was not accelerated in the MYCN/NCYM double transgenic mice (Figure S15E), the incidence of neuroblastomas with distant metastases was significantly increased in the MYCN/NCYM double transgenic mice (Figure 4, Figure S16, Table S3). The overexpression of the MYCN and NCYM proteins in primary and metastatic tumor cells was confirmed by immunohistochemistry (Figure 4B). In the neuroblastoma tissue of the double transgenic mice, GSK3β was significantly inactivated by phosphorylation at serine 9 (Figure 5A). To investigate the mechanism by which NCYM promotes the phosphorylation of GSK3β, we analyzed the phosphorylation status of the known upstream kinases for GSK3β, AKT [28] and S6K [31]. S6K was highly phosphorylated in the MYCN/NCYM double transgenic mice, whereas AKT was not noticeably activated. The phosphorylation levels of S6K in neuroblastomas from the MYCN/NCYM double transgenic mice were correlated with the expression levels of MYCN and NCYM (Figure 5A, M7-M11). These results suggest that NCYM promotes the phosphorylation of GSK3β via the activation of mTOR-S6K signaling. Furthermore, NCYM co-immunoprecipitated with GSK3β (Figure 5B) and substrates of GSK3β such as MYCN and β-catenin were stabilized in the neuroblastoma tissues induced in MYCN/NCYM transgenic mice (Figure 5A). We next examined the number of apoptotic cells in neuroblastomas from MYCN transgenic mice and MYCN/NCYM double transgenic mice by staining for cleaved caspase-3 (Figure S17). The number of apoptotic tumor cells was significantly decreased in the primary tumors of MYCN/NCYM double transgenic mice, suggesting that NCYM promotes the survival of neuroblastoma cells in vivo.
To examine whether the overexpression of NCYM contributes to the chemosensitivity of neuroblastomas via GSK3β inhibition, we tested the effect of NVP-BEZ235 on the survival of the MYCN/NCYM double transgenic mice. NVP-BEZ235 is a dual inhibitor of both PI3K and mTOR and promotes the degradation of MYCN to effectively reduce tumor burden in the MYCN transgenic mouse via GSK3β activation [32]. As reported, NVP-BEZ235 treatment significantly prolonged the survival duration of the MYCN transgenic mice (P<0.01; Figure 5C). In contrast NVP-BEZ235 did not prolong the survival of the MYCN/NCYM double transgenic mice (P = 0.648; Figure 5D). Thus, the expression of NCYM reduced the efficiency of this drug in vivo.
Our results reveal that NCYM, which was initially thought to be a large non-coding RNA transcribed from a cis-antisense gene of human MYCN [26], is actually translated into a functional protein in humans. MYCN is a highly conserved, major oncogene in human cancer. The newly evolved cis-antisense NCYM gene product targets the sense MYCN gene product, influencing its stabilization, which in turn enhances transcription of the NCYM gene. This positive autoregulatory loop may function in primary human neuroblastomas to enhance metastasis as well as drug resistance through stabilization of MYCN and even β-catenin, which are mediated by inhibition of GSK3β (Figure S18). Thus, NCYM is the first de novo evolved gene product shown to function in the development of human neuroblastoma.
NCYM promoted phosphorylation of GSK3β at serine 9 possibly via the activation of mTOR-S6K signaling, that might have led to the constitutive inactivation of GSK3β in vivo. Recently, Schramm et al. reported that MYCN transcriptionally regulates the mTOR pathway, promoting its activation [33]. Thus, MYCN might have enhanced S6K phosphorylation by activating the mTOR pathway in neuroblastomas caused in the double transgenic mice. Previous reports have suggested that neuroblastoma cell lines expressing high levels of MYCN were significantly more sensitive to mTOR inhibitors compared with cell lines expressing low MYCN levels [34]. Furthermore, our study showed that NCYM knockdown significantly induces apoptosis in MYCN-amplified neuroblastoma cells, whereas the effects were marginal in MYCN-single neuroblastoma cells. Therefore, the feedback regulation between mTOR-S6K signaling and MYCN/NCYM may contribute to the survival of MYCN-amplified neuroblastoma cells (Figure S18).
Although NCYM inhibits GSK3β-mediated MYCN phosphorylation in vitro, our data does not rule out the possibility that NCYM may stabilize MYCN in a GSK3β-independent manner. Because NCYM binds directly to MYCN both in vitro and in neuroblastoma cells, this may affect the recruitment of the regulators of MYCN stability. Indeed, we have recently found that the tumor suppressor protein Runx3 directly binds to MYCN in neuroblastoma cells and promotes degradation of MYCN in the ubiquitin–proteasome system dependent manner [35]. Therefore, the binding of NCYM to MYCN itself could affect the interaction of Runx3, or other regulators such as Aurora A [36] with MYCN to induce its stability. Further studies are required to evaluate the role of NCYM-mediated inhibition of GSK3β activity on MYCN stability.
Recent reports have suggested that both mutant ALK [37], [38] and Lin28B [39] promote the growth of neuroblastomas in transgenic mouse models by targeting MYCN for stabilization [37], [38] or overexpression [39]. Among the known regulators of MYCN, NCYM is the only gene that shows 100% co-amplification with MYCN in human primary neuroblastomas. Overexpressed NCYM stabilizes both MYCN and β-catenin, and enhances the generation of neuroblastomas with increased aggressive behavior such as distant metastasis and/or drug resistance, which are characteristics reminiscent of human neuroblastoma. Recently, Valentijn et al. suggested that the activation of the MYCN pathway is a more significant prognostic factor than the expression or amplification of MYCN in primary neuroblastomas [40]. Consistent with this, our results indicate that NCYM expression is associated with poor outcomes in human neuroblastoma regardless of genomic status of the MYCN/NCYM locus. Therefore, we anticipate that the positive auto-regulatory loop formed by MYCN and NCYM may be a promising target for developing novel therapeutic tools against high-risk neuroblastoma.
As suggested in the recent report [37], the concomitant inhibition of apoptosis and/or activation of survival signals may be required for MYCN to induce multiple tumors or metastases in vivo. In this study, we found that NCYM maintains the survival of MYCN-amplified neuroblastoma cells, and that the apoptotic cell number, indicated by cleaved caspase-3, was downregulated in MYCN/NCYM transgenic mice. In addition, GSK3β inhibition contributes to the inhibition of apoptosis in response to treatment with DNA-damaging drugs in neuroblastoma cells [41]. Therefore, the concomitant activation of other GSK3β substrates, such as β-catenin, and mTOR-S6K signaling by NCYM may be involved in the inhibition of apoptosis (Figure S18).
Since the proposals of Ohno and Jacob, the birth of a new gene has been believed to be caused by the duplication or rearrangement of pre-existing genes [1], [2]. The recent advances in whole genome sequencing technology and bioinformatics have identified the presence of de novo proteins; however, their physiological or pathological significance have largely remained unclear [3], [15]. In 2010, Li et al. reported that MDF1 originated de novo from a DNA sequence previously thought to be non-coding in Saccharomyces cerevisiae [7]. MDF1 inhibits mating efficiency by binding MATα2 and promoting vegetative growth. Therefore, while MDF1 was the first reported de novo gene whose protein product function was unveiled in a monad, NCYM may be the first de novo protein whose precise function has been clarified in multicellular organisms, specifically in humans.
In conclusion, NCYM is a de novo evolved protein which acts as an oncopromoting factor in human neuroblastoma. Our results suggest that de novo evolved new gene products may be involved in the functional regulation of human cancers and even other diseases.
DNA sequences of all species were extracted from the UCSC genome browser on the basis of conservation. From the protein-coding regions, we took the conserved block that was annotated as the region corresponding to the NCYM coding sequence, located in exon 3. For intron sequences, we used BLAT [42] to align the NCYM mRNA sequence (NR_026766) to the genome of each species and extracted the unmapped regions in the alignment. We found exactly two unmapped regions for each species except for mouse (and thus did not use the mouse sequence). For intergenic regions, we used multiz [43] alignment across 48 species in the browser and cut out 1000-bp sequences that corresponded to human intergenic regions. The sequences of common ancestors were estimated based on the maximum parsimony principle that led to the minimum number of nucleotide-base changes along the already-known phylogenetic tree of the five primates and mice [16]. For multiple possibilities with the same minimum number, we broke the tie by selecting the nucleotide base of the closest outgroup (e.g., when we had A for human, T for chimpanzee, and T for orangutan, we chose T for the common ancestor of human and chimpanzee). When multiple possibilities still remained (as in common ancestor 1), we considered all the possibilities to be equally probable. We estimated common ancestor sequences only within close species (human, chimpanzee, orangutan, and rhesus macaque). We used BLAST [44] to make an alignment between two translated amino-acid sequences ending at the first terminal codon, and calculated Ka and Ks using the KaKs_Calculator [45] with the ‘gMYN’ method, where Ka and Ks are the rates of non-synonymous and synonymous amino-acid changes, respectively. All pairs of sequences were aligned entirely from the start codon to the terminal codon and did not include any indels, except for the alignment between common ancestor 2 and common ancestor 3, for which we noted ‘frameshift’ instead of the Ka and Ks values.
We measured a bias in the codon frequencies (or amino acid frequencies) through the deviation from the uniform usage of each codon, using the Chi-squared statistic normalized to the number of codons:where n1, …, nk (ni≠0) are the observed number of codon 1, …, and that of codon k, respectively. N is n1+…+nk. We used R for the calculations and computed the P-values using a Monte-Carlo simulation with 10,000 replicates.
A polyclonal anti-NCYM antibody was raised in rabbits against a 14-amino acid stretch at the C-terminal region of NCYM (84-LGTRPLDVSSFKLK-97) (Medical and Biological Laboratories, Nagoya, Japan). The specificity of the purified antibody's affinity was assessed by immunoblotting.
Neuroblastoma tissues obtained from mice were fixed in 4% paraformaldehyde and paraffin-embedded for histological studies. Tissue sections were stained with hematoxylin and eosin (H&E) and examined histologically by pathologists for confirmation of the tumor type. Tissue arrays (FDA808a-1 and FDA808a-2, US Biomax, Rockville, MD, USA) were used for the analyses of NCYM or MYCN expression in normal and tumorous human tissues. For immunohistochemistry, tissue sections were stained with the polyclonal anti-NCYM antibody we generated, an anti-MYCN antibody (Calbiochem, San Diego, CA, USA), and cleaved Caspase-3 (Cell Signaling Technology).
MYCN-amplified human neuroblastoma TGW cells grown on coverslips were fixed with 4% paraformaldehyde in 1× PBS for 20 min at 4°C, permeabilized with 0.1% Triton-X for 20 min at room temperature, and then incubated with 2% BSA and 3% goat serum in PBS for 1 h to reduce nonspecific binding. Immunostaining was performed by incubating cells with the polyclonal anti-NCYM antibody and a monoclonal anti-MYCN antibody (Calbiochem) for 2 h at room temperature in a humidified chamber, followed by incubation with fluorescent-conjugated goat anti-rabbit IgG (diluted 1∶400) or fluorescent-conjugated goat anti-mouse IgG (diluted 1∶400), respectively. The coverslips were washed extensively with PBS, mounted in VECTASHIELD mounting medium with DAPI (Vector Laboratories, Burlingame, CA, USA) and images were captured using a confocal microscope (DMI 4000B, Leica).
We previously made a MYCN-luc (+1312) plasmid that contains the region of MYCN promoter region spanning from −221 to +1312 (where +1 represents the transcription start site) [27]. Luciferase reporter plasmids containing different lengths of the MYCN promoter were generated from MYCN-Luc (+1312) by partial removals of the MYCN promoter region with appropriate restriction enzymes. The MYCN promoter region in MYCN-luc (+1312) was subcloned into the pGL3basic vector or pGL4.17ΔEcoRV EcoRI vector in the opposite direction for generation of NCYM-luc vectors. pGL4.17 ΔEcoRV EcoRI was the luciferase reporter plasmid, where an EcoRV site in pGL4.17 (Promega, Southampton, UK) is replaced with an EcoRI site. NCYM-luc E-box WT and NCYM-luc E-box MT were generated by PCR-based amplification using MYCN-luc (+1312) as a template. Oligonucleotide primers used were as follows: 5′-AACCAGGTTCCCCAATCTTC-3′ (forward) and 5′-ACCACCCCCTGCATCTGCAT-3′ (reverse, NCYM-luc E-box WT) or 5′-ACCACCCCCTGCATCCGCAT-3′ (reverse, NCYM-luc E-box MT). Underlined sequences in the reverse primers indicate the wild-type or mutant E-boxes. The NCYM complementary DNA was introduced into a pcDNA3 expression vector, comprising a FLAG-tag at the 5′ locus of NCYM to generate pcDNA3-FLAG-NCYM. The sequence of the entire NCYM open reading frame was confirmed by sequencing. The FLAG-NCYM cDNA was ligated downstream of the rat TH promoter in the pGEM7z(f+) expression plasmid, which was originally made from a MYCN transgenic construct [21] by excision of the MYCN gene, to generate pGEM7z(f+)-FLAG-NCYM.
Human neuroblastoma cell lines SH-SY5Y, SK-N-AS, NLF, IMR32, CHP134, and SK-N-BE were maintained in RPMI-1640 medium supplemented with 10% fetal bovine serum (FBS) and antibiotics. Human neuroblastoma cell line BE (2)-C was maintained in a 1∶1 mixture of minimal essential medium (MEM, Gibco by Life technologies, Carlsbad, CA, USA) and Ham's Nutrient Mixture F12 (Gibco) supplemented with 15% heat inactivated fetal bovine serum (FBS) (Gibco) with MEM non-essential amino acids (Gibco) and antibiotics. Mouse neuroblastoma cell line Neuro 2a was maintained in DMEM supplemented with 10% FBS and antibiotics. NLF, IMR32, CHP134, SK-N-BE, and BE (2)-C have amplified MYCN, whereas SH-SY5Y, SK-N-AS, and Neuro 2a are cell lines with a single copy of MYCN. The cells or tissues with a single copy of MYCN have one copy of MYCN gene in a haploid genome. Lentivirus was produced by co-transfecting cDNA or shRNA expression plasmids with pCMVR and pMDG plasmids into HEK293T cells using FuGENE HD reagent (Roche, Mannheim, Germany). The MYCN and NCYM shRNA expression plasmids contained pLKO.1-puro as the backbone (Sigma, St Louis, MO, USA). At 24 and 48 h after transfection, the viral supernatants were collected and mixed with neuroblastoma cells. Other plasmid transfections were done using Lipofectamine 2000 transfection reagent (Invitrogen, Karlsruhe, Germany) according to the manufacturer's instructions. The target sequences of the shRNAs used were as follows: NCYM sh-1 (N-cym1 custom shRNA, Sigma) 5′-tggcaattgcttgtcattaaa-3′, NCYM sh-2 (N-cym 2 custom shRNA, Sigma) 5′-gaggttgctcctgtgtaatta-3′, NCYM sh-3 (N-cym 3 custom shRNA, Sigma) 5′-tcctgtgtaattacgaaagaa-3′, MYCN sh-1 (TRCN0000020694, Sigma) 5′-gccagtattagactggaagtt-3′, MYCN sh-2 (TRCN0000020695, Sigma) 5′-cagcagcagttgctaaagaaa-3′. The control shRNA (SHC002) was purchased from Sigma.
Total RNA was isolated from the frozen tumor samples and adrenal tissues of transgenic mice with ISOGEN (NIPPON GENE, Tokyo, Japan), and treated with RNase-free DNase I. Total RNA from neuroblastoma cells (CHP134 and SK-N-AS) was prepared using an RNeasy Mini kit (Qiagen, Valencia, CA) following the manufacturer's instruction. cDNA was synthesized using SuperScript II with random primers (Invitrogen). Quantitative real-time RT-PCR (qRT-PCR) using an ABI PRISM 7500 System (Applied Biosystems, Foster City, CA) was carried out using a SYBR green PCR reaction. The primer sets used were as follows: (for clinical experiments using primary neuroblastomas) human MYCN, 5′-ggacaccctgagcgattcag-3′, and 5′-aggaggaacgccgcttct-3′, human NCYM 5′-ccgacagctcaaacacagaca-3′ and 5′- gtaatggcttctgcgaaaagaaa-3′; (for cellular experiments) human MYCN, 5′-tccatgacagcgctaaacgtt-3′ and 5′- ggaacacacaaggtgacttcaaca-3′, human NCYM, 5′-cgcccccttaggaacaagac-3′ and 5′- gcgcccctcttctttcaatt-3′, mouse MYCN, 5′-tcgggacactaaggagcttca-3′ and 5′-ggaatcttggaccggaacaa-3′, mouse GAPDH, 5′-gggaagcccatcaccatct-3′ and 5′-cggcctcaccccatttg-3′. The mRNA levels of each of the genes were standardized by β-actin or GAPDH.
SK-N-AS cells were co-transfected with the indicated reporter constructs and the pRL-TK Renilla luciferase cDNA together with increasing amounts of the expression plasmid for MYCN or MYC. Total DNA per transfection was kept constant (510 ng) by adding pcDNA3 (Invitrogen). Forty-eight hours after transfection, firefly and Renilla luciferase activities were measured with a dual-luciferase reporter assay system according to the manufacturer's instructions (Promega).
We resolved cell proteins by SDS-PAGE before electro-blotting onto PVDF membranes. We incubated the membranes with the following primary antibodies overnight: anti-NCYM (1∶1000 dilution), anti-MYCN antibody (1∶1000 dilution; Calbiochem and Cell Signaling), anti-Lamin B (1∶1000 dilution; Calbiochem), anti-α-tubulin (1∶1000 dilution; Santa Cruz, CA, USA), anti-GST (1∶1000; Santa Cruz), anti-GSK3β (1∶1000 dilution; Cell Signaling), anti-phospho-GSK3β (S9) (1∶1000 dilution; Cell Signaling), anti-β-catenin (1∶1000 dilution; Cell Signaling), anti-phospho-AKT (S473) (1∶1000 dilution; Cell Signaling), anti-phospho-AKT (S308) (1∶1000 dilution; Cell Signaling), anti-AKT (1∶1000 dilution; Cell Signaling), anti-S6K (1∶1000 dilution; Cell Signaling), anti-phospho-S6K (T389) (1∶1000 dilution; Cell Signaling), and anti-actin (1∶4000 dilution; Sigma). The membranes were then incubated with a horseradish peroxidase-conjugated secondary antibody (anti-rabbit IgG at 1∶2000–1∶4000 dilution or anti-mouse IgG at 1∶2000 dilution; both from Cell Signaling Technology) and the bound proteins were visualized using a chemiluminescence-based detection kit (ECL and ECL pro kit, Amersham, Piscataway, NJ, USA; ImmunoStar LD, Wako).
Whole lysates prepared from CHP134 cells or tumor tissues were pre-cleared by incubation with protein G-Sepharose beads (Amersham Pharmacia Biotech) for 1 h at 4°C. The supernatant was collected after a brief centrifugation, and incubated with the indicated primary antibodies at 4°C overnight. The immune complexes were precipitated with protein G-Sepharose beads for 1 h at 4°C, and the non-specific bound proteins were removed by washing the beads with lysis buffer five times at 4°C. Different lysis buffers were used for the cell-based experiments (50 mM Tris-HCl pH 8.0, 137 mM NaCl, 2.7 mM KCl, and 1% Triton X) and for the tumor tissues (50 mM Tris-HCl pH 8.0, 1 mM EDTA, 0.2% DOC and 0.2% SDS). The immunoprecipitated proteins were eluted by boiling in Laemmli sample buffer and analyzed by immunoblotting.
CHP134 cells were cultured with 50% lentiviral supernatant for transfection of the indicated shRNA. Forty-eight hours after the transfection, cycloheximide (Sigma) was added to the culture medium at a final concentration of 50 µg/ml and cells were harvested at the indicated time points. For MG132 treatment, 44 h after the transfection, cells were treated with DMSO or 10 µM MG132 for 4 h.
DH5α cells were transformed with pGEX-4T-NCYM plasmid and cultured in Luria Broth (LB) at 37°C. The expression of the GST-NCYM fusion protein was induced by culturing the cells with 1 mM IPTG for 10 h at 25°C. Cells were collected by centrifugation, dissolved in cell lysis buffer (PBS, 1% TritonX-100, 5 mM EDTA and protease inhibitors), and stored at −80°C. Cell extracts were obtained by thawing the frozen cells, followed by sonication and ultra-centrifugation. After a pulldown with glutathione sepharose 4B beads, the beads were washed five times in cell lysis buffer and once in thrombin buffer containing 50 mM Tris-HCl pH 8.0, 150 mM NaCl, 2.5 mM CaCl2, 5 mM MgCl2, 1 mM DTT. GST-Tag cleavage mediated by thrombin released the full-length NCYM protein from the beads and the thrombin was removed by adding p-aminobenzamidine agarose beads according to the standard protocol. The full length NCYM protein was further purified by filtration using Amicon Ultra-4 (Millipore, Temecula, CA, USA), and dissolved in stock buffer (50 mM Tris-HCl pH 8.0, 150 mM NaCl, 5 mM EDTA, 0.25 mM DTT, 10% sodium azide, 50% glycerol and protease inhibitors) and stored at −20°C. Complete PIC (Roche) was used for protease inhibition.
For GST-pulldown assay, 0.5 µg of purified NCYM proteins were incubated with 0.5 µg of GST protein or GST-fused CDK1/Cyclin B1 (Signal Chem, Richmond, Canada), GSK3β (Promega) and MYCN (Abnova, Taipei, Taiwan) for 2 h at 4°C. Bound complexes were recovered on the glutathione-sepharose beads, washed with the binding buffer (50 mM Tris-HCl, pH 8.0, 1 mM EDTA, 150 mM NaCl, 0.1% Nonidet P-40 and Complete PIC), boiled in in Laemmli sample buffer and analyzed by immunoblotting.
For MYCN phosphorylation, two kinase reactions were performed sequentially. The first kinase reactions were performed for 1 h in kinase buffer (40 mM Tris-HCl pH 7.5, 20 mM MgCl2, 0.1 mg/ml BSA, 50 µM DTT) in the presence of 50 µM Ultrapure ATP (Promega), 50 ng of purified MYCN (Abnova), and 40 ng of purified CDK1/Cyclin B1 (Signal Chem) at room temperature. At 1 h, the first reaction solution was mixed with the same volume of kinase buffer containing 100 nM CDK1 inhibitor (CGP74514A, Calbiochem), 4 µCi of [γ-32P] ATP (PerkinElmer), and 20 ng of purified GSK3β with the indicated amounts of purified NCYM or purified GST. The second reaction was performed for 1 h at room temperature. The amount of phosphorylated MYCN was quantified using standard autoradiography. The total amount of MYCN was quantified by using an Oriole Fluorescent Gel stain (Bio-Rad). We also examined whether purified NCYM could be a substrate of GSK3β using the ADP-Glo system (Promega) according to manufacturer's instructions. Reactions were performed for 1 h in kinase buffer (40 mM Tris-HCl pH 7.5, 20 mM MgCl2, 0.1 mg/ml BSA, 50 µM DDT) in the presence of 25 µM Ultrapure ATP (Promega) and 25 ng of purified GSK3β with increasing amounts of NCYM or GST at room temperature. The peptide of human muscle glycogen synthase-1 (YRRAAVPPSPSLSRHSSPHQ(pS)EDEEE) was used as a positive control for the GSK3β substrate. At 1 h, the reaction solutions were mixed and incubated with ADP-Glo reagent for 40 min at room temperature, and the mixture was combined with a kinase detection reagent and allowed to stand for 30 min. The kinase activities were detected using a luminometer (PerkinElmer ARVOX3).
The indicated neuroblastoma cells were transfected with the indicated shRNA with 50% lentiviral supernatant. Seventy-two hours after transfection, all cells were collected by centrifugation, attached onto the coverslips by CYTOSPIN 4 (Thermo Fisher Scientific, Wilmington, DE, USA), and fixed in 4% paraformaldehyde for 1 h. Apoptotic cells were detected by using an in situ cell death detection kit (Roche) according to the manufacturer's protocol. The coverslips were mounted with DAPI-containing mounting medium (Vector Laboratories) and observed under a confocal microscope.
Cell viability was quantified by the 3-(4, 5-dimethylthiazol-2-yl)-2, 5-diphenyltetrazolium bromide (MTT) method. Cells were collected and seeded in 96-well plates at 1×104 cells/ml. After addition of 10 µl of MTT tetrazolium salt (Sigma) solution to each well, the plates were incubated in a CO2 incubator for 60 min. The absorbance of each well was measured using a Dynatech MR5000 plate reader with a test wavelength of 450 nm and a reference wavelength of 630 nm.
The invasive potential of BE (2)-C cells in vitro was measured by evaluating the number of invading cells using Matrigel-coated trans-well inserts (BD Biosciences) according to the manufacturer's instructions. BE (2)-C cells transfected with the indicated shRNA were seeded onto an insert containing 8 µm pores (BD Biosciences) in a 24-well plate at 1×105 cells/ml. Cells on the lower side of the membrane were fixed with 4% paraformaldehyde and stained using a Diff Quick Staining Kit (Sysmex).
All animal experimental procedures used in this study were reviewed and approved by the Committee on the Ethics of Animal Experiments of the Chiba Cancer Center (Permit Number: 12–13). Linearized and purified pGEM7z (f+)-FLAG-NCYM was injected into the pronuclei of fertilized eggs derived from 129/SvJ×C57BL/6J mice. We selected four lines of NCYM transgenic mice according to the level of NCYM expression in adrenal tissues (Figure S15B), and the transgenic mice were backcrossed to 129/SvJ at least 10 times to generate NCYM transgenic mice. To generate MYCN/NCYM double transgenic mice, the NCYM transgenic mice were crossed with MYCN transgenic mice of the 129/SvJ strain. On the basis of breeding schemas, all mice carrying the MYCN transgene were hemizygous. Tail DNA was analyzed for MYCN and NCYM transgenes, and the NCYM transgene copy number was quantified by quantitative genomic PCR. The primer sets used for genotyping were as follows: NCYM, 5′-cgcccccttaggaacaagac-3′ and 5′- gcgcccctcttctttcaatt-3′, MYCN, 5′-tggaaagcttcttattggtagaaacaa-3′ and 5′-agggatcctttccgccccgttcgttttaa-3′.
If more than one tumor over 2 mm in a diameter separately developed in a different organ, we defined this as the mouse having macroscopic metastatic tumors. In Figure 4C, only the number of mice with macroscopic metastatic tumors was counted. As a preliminary experiment, we used microscopy to detect tumors in the brain, pancreas, spleen, heart, lungs, kidneys and liver in nine mice (MYCN/NCYM double transgenic mice; n = 6, MYCN transgenic mice; n = 3). In addition to macroscopic metastases in the brain, heart, ovary and uterus, we found microscopic metastases in the lungs of MYCN/NCYM double transgenic mice, but the mass of these tumor cells was not large enough to be visible by eye. We also microscopically analyzed the HE-stained bone marrow from the hind legs of 19 mice (MYCN/NCYM double transgenic mice, n = 10; MYCN transgenic mice, n = 9). However, no metastatic tumor cells were found in the bone marrow.
All mice were genotyped to detect the presence of human MYCN or NCYM transgenes. After weaning, at about 30 days old, MYCN transgenic mice or MYCN/NCYM double transgenic mice were palpated for intra-abdominal tumors every day. Mice of either genotype found with palpable tumors were treated with NVP-BEZ235 (Cayman Chemical, Ann Arbor, MI, USA) (35 mg/kg in PEG300) or vehicle (PEG300, Wako) once daily for 30 days by oral gavage. All mice were monitored until euthanasia was required in accordance with the institutional animal committee.
The 106 human neuroblastoma specimens used in the present study were kindly provided by various institutions and hospitals in Japan to the Chiba Cancer Center Neuroblastoma Tissue Bank. Written informed consent was obtained at each institution or hospital. This study was approved by the Chiba Cancer Center Institutional Review Board. Tumors were classified according to the International Neuroblastoma Staging System (INSS): 27 Stage 1, 15 Stage 2, 34 Stage 3, 23 Stage 4, and 7 Stage 4 s. Clinical information including age at diagnosis, tumor origin, Shimada's histology, prognosis and survival duration of each patient was obtained. The patients were treated following the protocols proposed by the Japanese Infantile Neuroblastoma Cooperative Study and the Group for the Treatment of Advanced Neuroblastoma and subjected to survival analysis. Cytogenetic and molecular biological analysis of all tumors was also performed by assessing DNA ploidy, MYCN amplification and TrkA expression, as previously described [46].
Array CGH analysis was conducted using the Human Genome CGH 244K Oligo Microarray Kit (G4411B, Agilent Technologies, Santa Clara, CA, USA). Genomic DNA prepared from primary neuroblastoma tissues or cell lines was labeled with Cy3-dye using a QuickAmp labeling kit. Human placental DNA was labeled with Cy5-dye and used as a reference control. Labeling, hybridization and subsequent data processing by FeatureExtraction and CGH-Analytics software were performed according to the manufacturer's instructions. Relative copy number of the probes surrounding the MYCN and NCYM genomic locus (from DDX1 to FAM49A) were compared in each primary tumor or cell line.
Statistical significance was tested as follows: two-group comparison of survival by log-rank test, correlation of gene expression by Pearson's correlation coefficient test or Student's t-test, multivariate analysis for survival by Cox regression model, and the rate of mouse genotype and metastatic tumor occurrence in line 6 was calculated by Chi-square independence test and Mann–Whitney U test, respectively.
|
10.1371/journal.pbio.2006537 | Vocal practice regulates singing activity–dependent genes underlying age-independent vocal learning in songbirds | The development of highly complex vocal skill, like human language and bird songs, is underlain by learning. Vocal learning, even when occurring in adulthood, is thought to largely depend on a sensitive/critical period during postnatal development, and learned vocal patterns emerge gradually as the long-term consequence of vocal practice during this critical period. In this scenario, it is presumed that the effect of vocal practice is thus mainly limited by the intrinsic timing of age-dependent maturation factors that close the critical period and reduce neural plasticity. However, an alternative, as-yet untested hypothesis is that vocal practice itself, independently of age, regulates vocal learning plasticity. Here, we explicitly discriminate between the influences of age and vocal practice using a songbird model system. We prevented zebra finches from singing during the critical period of sensorimotor learning by reversible postural manipulation. This enabled to us to separate lifelong vocal experience from the effects of age. The singing-prevented birds produced juvenile-like immature song and retained sufficient ability to acquire a tutored song even at adulthood when allowed to sing freely. Genome-wide gene expression network analysis revealed that this adult vocal plasticity was accompanied by an intense induction of singing activity-dependent genes, similar to that observed in juvenile birds, rather than of age-dependent genes. The transcriptional changes of activity-dependent genes occurred in the vocal motor robust nucleus of the arcopallium (RA) projection neurons that play a critical role in the production of song phonology. These gene expression changes were accompanied by neuroanatomical changes: dendritic spine pruning in RA projection neurons. These results show that self-motivated practice itself changes the expression dynamics of activity-dependent genes associated with vocal learning plasticity and that this process is not tightly linked to age-dependent maturational factors.
| How is plasticity associated with vocal learning regulated during a critical period? Although there are abundant studies on the critical period in sensory systems, which are passively regulated by the external environment, few studies have manipulated the sensorimotor experience through the entire critical period. Thus, it is a commonly held belief that age or intrinsic maturation is a crucial factor for the closure of the critical period of vocal learning. Contrary to this idea, our study using songbirds provides a new insight that self-motivated vocal practice, not age, regulates vocal learning plasticity during the critical period. To examine the effects of vocal practice on vocal learning, we prevented juvenile zebra finches from singing during the critical period by postural manipulation, which separated the contribution of lifelong vocal experience from that of age. When these birds were allowed to freely sing as adults, they generated highly plastic songs and maintained the ability to mimic tutored songs, as normal juveniles did. Genome-wide transcriptome analysis revealed that both juveniles and singing-prevented adults, but not normally reared adults, expressed a similar set of singing-dependent genes in a song nucleus in the brain that regulates syllable acoustics. However, age-dependent genes were still similarly expressed in both singing-prevented and normally reared adult birds. These results exhibit that vocal learning plasticity is actively controlled by self-motivated vocal practice.
| Both human speech and birdsong are acquired through vocal learning [1,2]. This learning process is achieved through sensory learning to memorize model sounds and sensorimotor learning based on matching auditory inputs and motor output to the model sounds by iterative self-motivated practice of vocalization. However, neither humans nor songbirds maintain their vocal learning ability equally well during all phases of life; the ability is circumscribed by critical/sensitive periods for vocal learning. Although there exist abundant studies on the critical period of sensory system development regulated by environmental stimuli [3–5], the neural mechanisms underlying the critical period for sensorimotor learning, especially for vocal learning, are not well understood.
The songbird is highly advantageous for studying the neural substrate of vocal learning and its critical period. The critical period of song learning includes 2 phases, the sensory and sensorimotor learning phases (Fig 1A). In the sensory learning phase, juveniles acquire sensory memories of song by listening to adult birds’ songs, which serve as a template to imitate. The sensorimotor learning phase starts with the generation of soft and highly variable syllables, called “subsong.” Thereafter, birds start producing “plastic song,” characterized by the gradual inclusion of recognizable yet variable syllables. At the end of the learning process, the song is crystallized with acoustically and sequentially stable syllable patterns (“crystallized song”). In a closed-ended vocal learner songbird, like the zebra finch, the time window of the sensorimotor learning phase lasts 2 months, beginning in juveniles at 30–45 post hatching day (phd) and ending in adulthood at 90–100 phd with the production of crystallized motif song patterns that are then maintained throughout life (Fig 1A). During the sensorimotor learning phase, zebra finches produce approximately 1,000 song renditions in a day through self-motivated vocal practice [6–8].
Neuronal activity itself causes genetic responses in the brain. These activity-dependent genes either directly or indirectly influence the physiological function and structural maturation of neural circuits as genetic regulators for long-term neuronal plasticity [9–11]. Singing behavior also induces a set of activity-dependent genes in specialized brain regions, called the song nuclei (Fig 2A) [12–15]. The song nuclei are interconnected to form neural pathways for vocal learning and production [1,16]. Some of the singing activity–dependent genes are differentially regulated in the song nuclei between juvenile and adult stages [13,17–19]. However, a large variety of genes are developmentally regulated as age-related genes during the critical period of vocal learning in the song nuclei [20–22]. These studies suggest that the differentially regulated genes during the sensorimotor learning phase could be crucial molecules for modulation of vocal learning plasticity.
In this study, we investigated first how self-motivated vocal practice influences song maturation and second whether such practice modulates vocal learning plasticity during the sensorimotor learning phase in the zebra finch. Using a reversible singing prevention paradigm, we found that cumulative singing practice itself is an essential regulator of vocal learning plasticity: singing-prevented (SP) birds retained the ability to imitate a memorized tutored song into adulthood, well beyond the critical period of the sensorimotor learning phase. This age-independent sensorimotor learning was accompanied by significant changes in the expression of singing activity–dependent genes—but not age-related genes—in the projection neurons of song nuclei RA. Furthermore, the number of dendritic spines of the robust nucleus of the arcopallium (RA) was also affected by cumulative singing practice. These results show that self-motivated practice itself changes the expression dynamics of activity-dependent genes associated with vocal learning plasticity and that this process is not completely determined by age-dependent maturational factors.
The zebra finch (Taeniopygia guttata) is termed a closed-ended learner songbird because they can only learn during a critical period and subsequently produce a stereotyped song (Fig 1A)[23,24]. To elucidate the importance of self-motivated vocal practice to song development during the sensorimotor learning phase in contrast to age itself, we prevented juvenile zebra finches from singing before initiation of their first song (approximately 30 phd) until adulthood (91–133 phd, mean = 101.6 phd) by postural manipulation (Fig 1B and S1 Fig). The postural manipulation was performed by attaching a custom-made weight on the neck of juvenile birds only during daylight hours, based on a modified method that uses weights to manipulate singing [25]. The weight shifted their posture slightly toward an inferior position (approximately at 0.5–1.5 cm lower than normal height); the weight was supported by the floor of the cage (i.e., not carried by the bird’s neck). Although this manipulation prevented singing, the birds were exposed to a tutor song produced by their biological fathers and could still freely generate daily behaviors, such as drinking, eating, and producing non-song-related vocalization, like calls. The weight was detachable and daily adjusted for each bird (up to 16.5–24.0 g) so that it was at the threshold of preventing singing without adversely affecting the bird’s health or bodily growth. To account for the overall health of the animals, body weights were regularly monitored. Body weight and acoustic features of call vocalizations were not significantly different between normally reared and SP birds (P > 0.05, 1-way ANOVA with Bonferroni correction for body weight; P > 0.05, Welch’s t test for call acoustics) (S1 Fig), demonstrating no adverse effects on the growth of the peripheral vocal organs, such as the syrinx.
Singing prevention was highly effective: normal birds produce over 60,000 song bouts during the sensorimotor learning period (approximately 1,000 bouts/day × 60 days) [6], while the singing prevention birds produced less than 0.1% of this output (24–485 song bouts, mean = 279.0 bouts). When the birds were released from singing prevention at adulthood, they produced “age-unmatched” immature songs with highly variable acoustics and sequence of syllables, i.e., subsong or early plastic-like song (Fig 1B). To quantify the immaturity of song quality in SP birds, we calculated the values of pitch, pitch goodness, and amplitude modulation (AM) and the entropy variance of syllables as acoustic parameters [26,27] and motif consistency as a parameter of song sequence [20]. In all parameters, songs of the SP birds were similar to the subsong/early plastic song produced by normal juveniles (Fig 1C). Despite the manipulation, in 3 of 24 birds, the weight-based postural manipulation did not affect singing practice, i.e., the birds persistently continued singing (more than 10,000 bouts of total singing). Even under the postural manipulation condition, the persistent singers developed crystallized songs with the typical motif structure and copied song traits from their tutors (S1 Fig), indicating that iterative singing experience per se but not the experimental handling influences song development and learning.
In adult SP birds producing immature-like plastic songs, we investigated whether or not they also retained vocal learning plasticity to mimic the tutor songs that they had already memorized. We found that SP birds quickly crystallized structured songs within 4 weeks after release at adulthood (Fig 1D and S2 Fig), which was less than half the duration of the normal sensorimotor learning period in the zebra finch. Moreover, the syllables produced by the SP birds, despite being crystallized so rapidly, had the same acoustic traits that were observed in unmanipulated adults (Fig 1C). In addition, the SP adult birds did not only develop species-typical crystalized songs, but they also mimicked their tutor songs at the levels of both syllable acoustics and sequence (motif) features. Comparison between the songs produced at 1–2 days after release (phd 101–103) and ones after 3–4 weeks (phd 120–126) revealed a significant increase of syllable and motif similarity scores toward their tutor songs (paired t test: syllable, t(4) = 7.7, P = 0.0015; motif, t(4) = 5.8, P = 0.0044) (Fig 1E and 1F). A subset of the SP adults near-perfectly mimicked the phonology and sequence order of all syllables of the memorized tutor song (S2 Fig). Consistently, the SP birds showed a similar imitation accuracy of their acquired songs against tutor songs as did the birds that persistently continued singing (Welch’s t test: syllable, t(6) = 0.68, P = 0.52; motif, t(6) = 1.2, P = 0.27) (Fig 1E and 1F). These results indicate that cumulative singing experience (i.e., vocal practice) acts as an age-independent sensorimotor learning mechanism in the zebra finch.
We conducted a genome-wide gene expression network analysis comparing juvenile, adult, and SP adult (1–2 days after release from the postural manipulation) birds to elucidate the transcriptional impacts of cumulative singing experience versus age in the brain regions. The song nuclei are interconnected to form 2 functional circuits: one being the vocal motor pathway (VMP) and another forming the anterior forebrain pathway (AFP), a cortical–basal ganglia–thalamic loop (Fig 2A) [28–30]. For this purpose, we sampled laser microdissected tissues of 2 song nuclei HVC and RA in the VMP, which regulate syllable sequence and acoustics, respectively (Fig 2A and 2B). In 12,156 genes expressed in the telencephalon of the zebra finch, 3,214 and 1,811 genes were identified in HVC and RA, respectively, as genes that were differentially regulated by age and/or singing. For these genes, a weighted gene coexpression network analysis (WGCNA) identified 5 and 4 “Gene Clusters” in HVC and RA, respectively, as the coexpressed genes correlated with singing experience, age, and/or singing induction (Fig 2C–2E). Only one of the gene clusters, RA Gene Cluster I, met criteria that were significantly regulated by cumulative singing experience instead of age. This gene cluster contained 119 genes, including a novel set of singing activity–dependent genes (transcription factors/regulators: Atf3, Crem, Nr4a1, and Irf8; subunits of histone deacetylase complex: Fam60a; serine/threonine kinases: Sik1 and Sgk1; and mitogen-activated protein (MAP) kinase phosphatases: Dusp5 and 6). In general, the expression of neuronal activity–dependent genes in song nuclei is regulated by singing rather than hearing [12,20]. RA Gene Cluster I was induced by diurnal acute singing. However, the singing-driven induction response was gradually attenuated as singing experience accumulated over time (Fig 2E and 2F), suggesting that the long-term cumulative experience of vocal practice progressively dampens the expression of singing activity–dependent genes as the sensorimotor learning phase progresses.
WGCNA further indicated that other gene clusters (Cluster II and III) were regulated by both singing experience and age (Fig 2C–2E). However, when comparing each individual gene expression of the clusters in the expression heat map, most of the genes in Cluster II and III were similarly expressed in both normally reared and SP birds as age-regulated genes (Fig 2F and S3 Fig), indicating that SP birds normally retain the developmental expression dynamics of age-regulated genes. In line with this, a cis-enrichment analysis revealed that differentially unique sets of transcription factors presumably bound the promoter regions of the coexpressed genes between different clusters. For an example, transcription factor families, including Nf-κB, Creb/Atf, Sp, Rfx, and Pax, were significantly enriched in putative proximal promoter regions (less than 1 kb from the transcription start site [TSS]) of the RA Cluster I genes but were not enriched in Cluster II genes (both Fisher’s exact test and G test, p < 0.05) (Fig 2G). These results suggest cumulative singing experience regulates the transcriptional of activity-dependent genes in RA.
In addition, to ensure that the expression of RA Cluster I genes reflects cumulative singing experience but not age, we further performed direct comparisons of the RA transcriptomes between the SP adult singing versus the normal adult singing groups and between the SP adult singing versus the juvenile singing groups (Fig 3). As the result, a total 57 of 119 RA Cluster I genes were sorted into the differentially regulated genes between the SP adult versus the normal adult groups (Fig 3B; green) but few into the ones between the SP adult versus juvenile groups (Fig 3B; purple). This result reconfirms that the expression of singing activity–dependent genes in the RA Cluster I is modified by cumulative singing experience instead of age.
To reveal the expression pattern of the cumulative singing experience–regulated genes in the entire song nuclei for vocal learning and production, we compared gene expression between silent and singing conditions in juveniles, adults, and SP adults (1–2 days after their release from the postural manipulation). We chose 13 genes: Arc, Crem, Nr4a1, Sik1, Dusp5, Fam60a, Atf3, c-fos, Egr1, H3.3b, Gadd45b, Dusp6, and Odc, which were independently identified as the cumulative singing experience–regulated genes by WGCNA (RA Cluster I genes in Fig 2) and the direct transcriptome comparison between normal and SP singing birds (Fig 3). All 13 tested genes were induced by singing and had a unique expression pattern in song nuclei between the 3 groups (Fig 4A, S4 and S5 Figs). In addition, singing activity–dependent gene expressions were differently regulated among song nuclei and also between the juvenile, adult, and SP adult groups. For example, Arc was consistently and intensely induced by singing in HVC, lateral magnocellular nucleus (LMAN), and Area X in all 3 groups. However, in RA, although juveniles and SP adult birds showed strong response of Arc expression after singing, the singing-driven induction was attenuated in normal adults (Fig 4A and S4 Fig). We therefore compared each gene induction after singing in each song nucleus between the 3 groups and then identified that RA is a major region that differently regulated the singing-driven induction response between the 3 groups (Fig 4B, S4 and S5 Figs). SP adults showed a striking resemblance to normal juveniles in the expression dynamics of the cumulative singing experience–regulated genes throughout song nuclei that was unlike normally reared adults (Fig 4B). Induction of 5 of 13 tested genes (Arc, Fam60a, Dusp6, Odc, and Gadd45b) was almost fully repressed in RA during adult singing, although singing evokes robust neuronal activity in RA in both juvenile and adult stages [31]. In contrast, a set of age-regulated genes identified as RA Gene Cluster II (including Gabra5, Evl, Dpysl3, and Il1rapl2) showed similar expression levels and patterns between SP and normal adults (Fig 4A and 4B and S6 Fig). These results indicate that SP birds selectively maintain juvenile-like expression for singing activity–dependent genes but express age-regulated genes in the entire song circuits similarly to untreated adults.
Precise and reliable neural activity driven by the functional neural connectivity between excitatory and inhibitory neurons within premotor circuits is critical for the production of structured song patterns [32–35]. We therefore examined which types of RA neuron expressed the cumulative singing experience–regulated genes. By colabeling with a glutamatergic excitatory neuron maker Vglut2 or a GABAergic inhibitory neuron maker Gad2 [36], we identified that the tested cumulative singing experience–regulated genes—Arc, Nr4a1, Sik1, and Dusp5—were coinduced in the glutamatergic excitatory neurons, not GABAergic interneurons, of RA after singing (Fig 5A). Furthermore, by colabeling with DiI retrograded from nXII, we confirmed neurons expressing the singing experience–regulated genes projected to nXII, innervating vocal and respiratory musculature [37] (Fig 5B).
Dendritic spine density in RA projection neurons was associated with vocal plasticity and reduced through the critical period of song learning [38,39]. Therefore, we measured whether singing experience regulated dendritic spine density in the RA projection neurons. We found that both SP adults and normal juveniles retained a higher density of dendritic spines in the RA projection neurons compared with normal adults (Bonferroni-corrected unpaired t test: juvenile:adult, t(31) = 9.14, p = 2.6e-10; SP adult:adult, t(28) = 6.59, p = 3.8e-7) (Fig 5C and S7 Fig). In contrast, the arcopallial region surrounding RA, which is a nonvocalization-related area, did not show significant differences in the number of dendritic spines between the 3 groups, supporting the idea that the singing prevention treatment has selective effects on the song-control regions. Although we could identify only a few of the RA interneurons given technical limitation of Golgi staining, these results suggest that cumulative singing experience results in dendritic spine pruning of RA projection neurons that express singing activity–dependent genes.
We revealed here that cumulative singing experience regulates song development, the expression dynamics of activity-dependent genes, and dendritic spine density of RA projection neurons. These results demonstrate that singing practice, rather than age, acts as a nongenetic factor to regulate vocal learning plasticity. A number of neural circuits mature during the critical period of heightened neuronal plasticity early in life [3]. Past studies have shed light on the effects of “passive” sensory experience from the external environment on the regulation of critical periods of sensory systems [40–42]. Sensory input induces gene expression for the functional and structural plasticity of synapses [43,44]. However, the regulatory mechanisms of the critical period of sensorimotor learning for the “active” acquisition of sequential motor skills, such as human language, playing instruments, or birdsong learning, remain unclear. In general, such complex and structured motor patterns do not suddenly emerge through the short-term experience of practice. They rather gradually develop after the cumulative experience of longer periods of self-motivated practice. Therefore, separating the effects of age (intrinsic developmental maturation) and practice (self-motivated behavior) is crucial for precise understanding of the neural mechanisms underlying the critical period of sensorimotor learning.
Compared with birds isolated from sensory learning (tutoring) or auditory feedback during the sensorimotor learning, SP birds produced immature subsong/early plastic songs when they were released from manipulations at adulthood (>100 phd). In contrast to SP birds, both nontutored and auditory feedback–prevented birds can produce singing behavior during the sensorimotor learning phase and develop a certain degree of structured songs by adulthood even without referencing tutor song memories [45–47]. When nontutored birds in social isolation first hear tutor songs at 120 phd, they thereafter change a few syllables of their own already existing song to mimic those from their tutor’s song. However, they do not retain the vocal plasticity necessary to mimic their tutor’s song structures [45,46]. This decreased song plasticity in early isolated birds could be caused by partial closing of sensory and sensorimotor learning ability in nontutored birds at adulthood. In parallel, birds prevented from hearing their own song production through noise exposure can change syllable acoustics but not copy the sequence order of memorized tutor songs when they are released from noise exposure as adults [47]. In contrast to adult birds prevented from matching their song output to memories, SP adult birds retain sufficient vocal learning plasticity to copy both the syllable acoustics and sequence order of memorized tutor songs. This suggests that singing practice itself, rather than other age-related factors, regulates the critical phase of sensorimotor learning. The high degree of vocal learning plasticity in SP birds was associated with producing structurally immature songs when they are allowed to sing freely. Therefore, although interruption of normal tutoring and hearing experiences can retain a certain degree of vocal plasticity until the adult stage, preventing cumulative singing experience induces a more intensive effect on vocal learning plasticity by almost stopping song maturation. This song immaturity could be a direct cause of the later vocal learning plasticity, and it enabled age-independent vocal learning after the sensitive period in the SP birds. However, because of the limited sample size in this study, we could not fully examine whether the amount of singing practice during the prevention period affects individual’s song imitation accuracy after release.
We previously reported that early-deafened zebra finches produce a normal amount of singing during development and still maintain highly variable songs in terms of both syllable acoustics and sequence even as adults (140–180 phd) [20]. Therefore, cumulative singing experience itself is not sufficient to decrease vocal plasticity. Rather, the cumulative experience of sensorimotor integration—i.e., self-motivated vocalization with normal auditory feedback (tutoring)—is an essential behavioral factor to promote song stabilization—i.e., closing of the sensorimotor learning period. However, this study does not examine a potential contribution of cumulative singing experience to closing the sensory learning phase versus the sensorimotor learning phase. To elucidate this point, it would be necessary to track the song development of SP birds that are exposed to a tutor only in adulthood. In addition, singing prevention longer than the 100 days used in the present study would allow us to examine the time limitation of retaining sensorimotor learning plasticity, which might be genetically constrained.
RA has a crucial role in the song circuits as the telencephalic output nucleus receiving 2 premotor inputs: a motor exploration signal from the LMAN as the cortical-basal ganglia-thalamic circuit output and a time-locked sequence signal from HVC of the motor circuit (Fig 2A) [31,48]. However, the 2 circuits do not equally contribute to song production during the critical period of sensorimotor learning. Early in song development, vocal output is dominated by LMAN input to RA [49]. With accumulation of singing experience, the LMAN’s effectiveness is curtailed by pruning and strengthening of the HVC to RA synapses [50,51]. In addition, synaptic connectivity from interneurons to projection neurons in RA is initially dense and nonspecific and then pruned with specific reciprocal patterns during sensorimotor learning [33], suggesting that song development engages multiple processes to reduce shared synaptic inputs to RA projection neurons. If so, our finding that the SP birds retained higher density of dendritic spines in the RA projection neurons like juveniles may indicate a singing experience—but not age-dependent mechanism for sculpting of functional connectivity of RA. Activity-dependent synaptic potentiation and depression are induced at the synapses of RA projection neurons [52]. Notably, zebra finch juveniles (45–60 phd) possess distinct capacity for long-term depression (LTD) in RA neurons compared with adult birds [53]. Arc is a critical regulator that modulates the synaptic plasticity underlying LTD induction [54,55]. This evidence supports a functional linkage between activity-dependent synaptic plasticity and activity-dependent gene induction in RA for regulation of vocal plasticity through the singing experience–dependent sculpting of functional connectivity.
Most of the singing experience–regulated genes were previously known as immediate-early genes, which regulate downstream effector proteins for activity-dependent synaptic plasticity [11,56–58]. Singing behavior generates robust neuronal activity in the song system, including RA and other song nuclei, throughout a bird’s life from juvenile to adult stages [31,48,59–61]. However, some immediate-early genes are more highly induced by singing in the song nuclei (e.g., egr1[zenk] and Arc in RA and penk in HVC) in juvenile than adult stages, although the causal reason of the developmental different induction is not known [13,17–19]. In this study, we found that RA shows a reduction of singing activity–dependent gene induction during the accumulation of singing experience. Which region-selective neuronal mechanisms regulate change in the expression dynamics of singing activity–dependent genes in RA? One possibility is neuronal activity–mediated epigenetic regulation—i.e., induction and activation of epigenetic regulators by neuronal firing to change epigenetic states at the regulatory regions of other activity-dependent genes [62]. In the RA Cluster I genes, we identified the epigenetic regulators that were induced by singing: a DNA methylation regulator Gadd45b [63,64], replacement histone H3.3b [65], and a subunit of the Sin3–histone deacetylase 1 (HDAC1) complex Fam60a [66,67]. These singing-driven epigenetic regulators could directly change the epigenetic state of the regulatory regions of other RA Cluster I genes, but this hypothesis remains to be fully evaluated.
Vocal learning has evolved convergently in a few lineages of birds and mammals. Recently, there is accumulating evidence that marmoset monkeys possess the ability for production-related vocal plasticity, especially to develop acoustic features via feedback from their parents [68–70]. Production-related vocal plasticity in marmosets could be a good mammalian model system to examine the cumulative vocal practice–related expression dynamics of activity-dependent genes that we found in songbirds. Avian vocal learners (songbirds, parrots, and hummingbirds) and humans possess analogous neural networks connecting anatomically similar cortical and subcortical brain regions as a form of convergent neural circuit evolution [16]. In addition, songbirds and humans share convergent transcriptional specializations in the brain regions for learned vocalization [71]. Therefore, language acquisition in humans could be mediated by the expression dynamics of neuronal activity–dependent genes in specific neural populations, such as cortical layer V projection neurons in laryngeal motor regions, analogous to songbirds’ RA [71]. Although many challenges remain in elucidating the neural basis of vocal learning, insights from songbirds may lead to a better understanding of the molecular mechanisms underlying learned vocal communication.
All experiments were conducted under the guidelines and approval of the Committee on Animal Experiments of Hokkaido University (Approved No. 13–0061). These guidelines are based on the national regulations for animal welfare in Japan (Law for the Humane Treatment and Management of Animals with partial amendment No. 105, 2011). For brain sampling, the birds were humanely killed by decapitation after overdose pentobarbital injection.
Zebra finch adult males were obtained from our breeding colonies at Hokkaido University. Birds were kept in breeding cages under a 13:11 hour light/dark cycle. During song-recording sessions, each bird was individually housed in a cage inside a sound-attenuating box.
For vocal practice restriction, singing prevention was performed from the initiation of first singing (at around 30 phd) until adult stage (n = 24 birds, 91–133 phd, mean ± SD = 101.6 ± 8.8 phd). During light-on time, juveniles were prevented from singing by a custom-made weight on the neck that shifted their posture slightly toward an inferior position (approximately at 0.5–1.5 cm lower than normal height). The weight was detachable and daily adjusted for each bird (up to 16.5–24.0 g). Note that weight was usually supported by the floor, not carried by the bird’s neck, during the days. Therefore, the birds could freely generate daily behaviors, such as drinking, eating, grooming, and calling. The weight was removed from birds during light-off time and for at least 1 hour during light-on time to reduce potential stress. During the weight-free time each day, the real-time singing behaviors were monitored with Sound Analysis Pro (SAP) and interrupted by light tapping or by opening the sound-attenuating box. No singing behavior was observed during light-off time. Body weight was carefully monitored every 2–3 days. For hearing experience to a tutor song, birds were kept with their biological fathers after hatching until 30 phd and subsequently exposed to their father every 2–4 days until adult stage.
Songs were recorded using a unidirectional microphone (SM57, Shure) connected to a computer with SAP (v1.04) [72]. Singing duration was defined as the total amount of singing during the last 30 minutes before euthanasia for brain sampling. A song bout was defined as the continuous production of syllables followed by at least 200 ms of silence.
Song motif consistency was measured as the motif similarity score within each day and calculated with the default setting in the SAP software using “time-course” and “symmetric” comparison modes. We randomly selected 20 bouts of songs produced after 3 PM in 1 day. The similarity scores between any 2 of 20 bouts were compared by the round-robin comparison—i.e., a total of 190 similarity scores were calculated at each developmental stage. The similarity score, which represents a global measure of percent similarity, was calculated by comparing syllable acoustic features (e.g., pitch, FM, AM, Wiener entropy, and goodness of pitch) within 9 ms sliding time windows. P values for comparisons of motif consistency between developmental stages or conditions were obtained using an unpaired t test for different conditions and a paired t test among similar conditions with Bonferroni correction.
For the motif-based song similarity analysis, 20 bouts of songs were randomly selected and analyzed at each developmental time point. Song similarity scores were calculated by whole-motif comparison against each pupil’s tutor songs using the of the SAP software. For the syllable-based song similarity analysis, we used 50 syllables from the same song bouts, which were analyzed for the motif-based song similarity. Introductory notes in a song were not included for analyses. The series of separated syllable files of songs were transferred to the CORRELATOR program of Avisoft SASLab pro (Avisoft Bioacoustics, Berlin, Germany) for calculating the similarity scores between the syllables from pupils’ and tutors’ songs by the round-robin comparison [73]. The highest similarity score for each syllable of pupil songs against tutor syllables was averaged as the similarity score of total syllables for each individual.
Male zebra finch juveniles (n = total 29, 40–55 phd), adults (n = total 26, 101–338 phd), and SP adults (n = total 12, 91–133 phd at 1–2 days after release from singing prevention) were used for Quartz RNA-seq and in situ hybridization. Each bird was individually placed in a sound-attenuating box overnight, and singing behavior (undirected singing) was recorded during the next morning after lights on. Similarly to above, for brain sampling of silent conditions, birds were prevented from singing (but allowed to produce calls) by light tapping on cages when the birds started singing after lights on. After each session of singing behavior observation, the birds were humanely killed by decapitation. Brains were embedded in OCT compound (Sakura Fine Technical) and stored at −80 °C until use.
For sampling of song nuclei and RNA extraction, male zebra finch juveniles after 45 minutes silent (n = 3, 47–48 phd), juveniles with 45 minutes singing (n = 3, 40–50 phd), adults with 45 minutes silent (n = 2, 101–104 phd), adults with 45 minutes singing (n = 4, 110–338 phd), and SP adults with 45 minutes singing (n = 3, 96–101 phd at 1–2 days after release from singing prevention) were used (S1 Table). For identification of clear RA and HVC boundaries against surrounding nonvocal areas under microscope observation, a fluorescent-retrograde tracer, Cholera Toxin B subunit conjugated with AlexaFluor555 (Invitrogen, 1 mg/μl in 1× PBS, 100 nl/hemisphere), was injected into RA 10 days before euthanasia. After behavioral observations, birds were decapitated, and brains were removed and stored at −80°C until sectioning. Brain sections were cut at a 20 μm thickness in the sagittal plane and mounted onto glass slides with a handmade membrane system for laser microdissection. Fluorescent-labeled RA and HVC tissues were microdissected from 14 to 20 brain slices using a laser capture microscope ArcturusXT (Arcturus Bioscience). The collected tissue was dissolved in Qiagen RLT buffer. Total RNA was purified using a column-based method (RNeasy Micro kit; Qiagen) and treated with DNase in the column to avoid contamination of genomic DNA. RNA integrity number (RIN) and concentration were measured with a Bioanalyzer2100 (Agilent Technologies) to confirm RNA quality (RIN: 6.4–7.4, RNA concentration: 3.5–10 ng/μl).
For cDNA synthesis, amplification, and library preparation, cDNA was amplified from purified total RNA using previously reported methods [74]. Total RNA (10 ng) was used for synthesis of first-strand cDNA. The following PCR amplification was performed with 14 PCR cycles at 98 °C for 10 seconds, 65 °C for 15 seconds, and 68 °C for 5 minutes. The amplified cDNA samples were purified using a PCR purification column (MiniElute PCR Purification Kit; Qiagen). To check the quality of amplified cDNA, concentrations and smearing patterns of cDNA samples were measured with a Bioanalyzer 2100 (cDNA amount, 72–434 ng) (S1 Table). Amplified cDNA samples (10 ng) were fragmented to 100–300 bp in size using a DNA Shearing System LE220 (peak incident power 450 W, duty factor 30%, cycle/burst 200, and treatment time of 700 seconds) (Covaris) and then purified by a Zymo DNA 5 column. After end repair of DNA fragments, adaptors were ligated and amplified using a ligation-based Illumina multiplex library preparation method (LIMprep) with a KAPA Hyper Prep Kit (Nippon genetics) and 7 PCR cycles. All libraries were then sequenced using a Hiseq 2500 Sequencer (Illumina) for 100 bp single-end sequencing. Library preparation was performed in the Bioinformatics Research Unit at RIKEN Advanced Center for Computing and Communication under supervision by Drs. Y. Sasagawa and A. Nikaido. All Quartz RNA-seq data from zebra finches were deposited in the DDBJ Sequence Read Archive (submission number DRA005559).
The previous gene annotation file from Ensemble (Taeniopygia_guttata taeGut3.2.4.76.gtf) did not include 3′ UTR information. For annotation of read sequences obtained from the RNA-seq data, the lack of 3′ UTR information decreases the chances of accurate estimations of gene expression. Therefore, we elongated the annotation information with our RNA-seq data from zebra finch whole-brain samples (S1 Table). Total RNA was isolated from the pallium and pallidum regions of adult male zebra finches under silent and dark conditions (n = 5, 234–786 phd) using TRIzol Reagent (Invitrogen) according to manufacturer’s protocol (Invitrogen) and then column purified using a RNeasy Micro kit (Qiagen). Samples were treated with RNase-free DNase. The total RNA samples were used for library synthesis with TruSeq DNA Sample Prep Kits (Illumina). All libraries were then sequenced using the Illumina Hiseq 2500 Sequencer for 100 bp paired ends. These experimental steps were performed in Dr. Y. Suzuki’s laboratory in the Department of Computational Biology at the University of Tokyo. The 33.5–47.0 M reads for each telencephalon brain sample were output from the Illumina Hiseq 2500. Sequencing reads were mapped onto the ZF reference genome obtained from Ensemble (Taeniopygia_guttata taeGut3.2.4.dna.fa) with the Tophat2 program and assembled to predicted transcripts with the Cufflinks program. By comparison with the previous annotation file using the cuffcompare program, 12,156 transcripts were identified as predicted RNA transcripts expressed in the zebra finch telencephalon. The RNA-seq data from the zebra finch telencephalon were deposited in the DDBJ Sequence Read Archive (submission number DRA005548 and DRA005559).
Total RNA-seq reads (9.7–20.9 M) from zebra finch juveniles after 45 minutes silent, juveniles with 45 minutes singing, adults after 45 minutes silent, and adults with 45 minutes singing were used. First, RNA-seq reads were mapped onto the zebra finch reference genome with the Tophat2 program, and then the fragments per kilobase of exon per million mapped fragments (FPKM) of each transcript (12,156 genes) was calculated using the Cufflinks program. Principal component analysis (PCA) using the prcomp package in R was used to check whether there were outliers of quality of RNA-seq. A Bonferroni-corrected DEseq2 was used to identify the differentially expressed genes between juveniles and adults (P < 0.05; 1,540 genes in HVC and 1,352 genes in RA), singing and silent conditions (P < 0.05; 385 genes in HVC and 266 genes in RA), and juvenile singing and adult singing conditions (P < 0.05; 937 genes in RA and 2,443 genes in HVC) in normal birds. A total of 3,214 genes in HVC and 1,811 genes in RA were detected as differentially expressed genes.
WGCNA is a biologically meaningful technique for quantifying similarity of expression patterns among all pairs of probes across all treatment conditions [75,76]. WGCNA identifies modules of densely interconnected probes by hierarchical clustering based on topological overlap and by assigning each probe to a “Cluster (module)” based on shared expression patterns. A WGCNA was performed using the WGCNA R package to further investigate the relationship between biological traits (singing experience, age, and/or singing induction) and identify coregulated gene clusters. General information about network analysis methodology and WGCNA software is available at http://labs.genetics.ucla.edu/horvath/htdocs/CoexpressionNetwork/. Pairwise Pearson correlation coefficients were calculated for all detected genes. The resulting Pearson correlation matrix was transformed into a matrix of connection strength (an adjacency matrix) using the power function [(1 + correlation) / 2 × soft threshold power], which was then converted to a topological overlap matrix. A preliminary network was built to assess overall connectivity. From this network, 3,214 and 1,811genes in HVC and RA, respectively, with the highest connectivity were retained for subsequent WGCNA (soft threshold power = 10, corType = pearson, minModuleSize = 30, detectCutHeight = 0.98, and merge CutHeight = 0.4). Clusters were defined as branches of the dendrogram obtained from clustering and were labeled with colors beneath the dendrograms. To study the relationship between expression variability within the clusters and behavioral trait variability, correlations were computed between the principal components of each module and traits. P values were computed for each correlation.
To further confirm the WGCNA results, we separately compared the expression profiles of the differentially expressed genes (1,811 genes) in RA under singing condition. DESeq2 (P < 0.05) was performed to compare the FPKM values of each transcript between adult singing (n = 4) and SP adult singing (n = 3) and between juvenile singing (n = 3) and SP adult singing (n = 3). We then counted the overlap between the Cluster I gene and genes detected by DESeq2.
Information on transcription factors (TFs) and their binding motifs was obtained from the JASPAR 2018 database (http://jaspar.genereg.net, CORE Vertebrata). We then used 349 genes in the CIS-BP database as TFs existing in the zebra finch genome (T. guttata taeGut3.2.4.dna.fa). The upstream 1 kb sequences of the TSSs of the 12,156 genes that were expressed in the telencephalon were extracted from the zebra finch genomic sequence. A total of 7,052 genes, including 61 and 519 genes in RA Gene Cluster I and II, respectively, were used for further cis-enrichment analysis, because the other 5,104 genes possessed a sequence gap (nonsequenced region) in their upstream 1 kb from TSS.
The Find Individual Motif Occurrences (FIMO) program was used to search for TF binding sites in the upstream 1 kb of the TSS (p < 104, for each binding site). For each TF, we calculated how many RA Gene Cluster I and II genes had the TF binding sites in the upstream 1 kb. To predict TFs of the cluster genes, the frequency appearance of binding sites was compared with the background frequency of 7,052 genes for each TF (both Fisher’s test and G test, p < 0.05).
cDNA fragments used for the synthesis of in situ hybridization probes were cloned from a whole-brain cDNA mixture of a male zebra finch. Total RNA was transcribed to cDNA using Superscript Reverse Transcriptase (Invitrogen) with oligo dT primers. The cDNAs were amplified by PCR using oligo DNA primers directed to conserved regions of the coding sequence from the NCBI cDNA database (S2 Table). PCR products were ligated into the pGEM-T Easy plasmid (Promega). The cloned sequences were searched using NCBI BLAST/BLASTX to compare with homologous genes to other species and identified genome loci using BLAT of UCSC Genome Browser.
For radioisotope in situ hybridization, male zebra finch juveniles (n = total 23, 45–55 phd), adults (n = total 20, 103–227 phd), and SP adults (n = total 12, 91–133 phd at 1–2 days after release from singing prevention) were split into 6 experimental groups: (I) juvenile 30 minutes silent, (II) adult 30 minutes silent, (III) juvenile 30 minutes singing, (IV) adult 30 minutes singing, (V) SP adults with 30 minutes silent, and (VI) SP adults with 30 minutes singing. Frozen sections (12 μm thick) were cut in the sagittal plane. Brain sections for a given experiment were simultaneously fixed in 3% paraformaldehyde/1× PBS (pH 7.4), washed in 1× PBS, acetylated, dehydrated in an ascending ethanol series, air dried, and processed for in situ hybridization with antisense 35S-UTP-labeled riboprobes of genes. To generate the riboprobes, gene inserts in the pGEM-T Easy vector were PCR amplified with plasmid M13 forward and reverse primers and then gel purified. The amplified DNA fragments and SP6 or T7 RNA polymerase was used to transcribe the antisense 35S-riboprobes. A total of 1 × 106 cpm of the 35S-probe was added to a hybridization solution (50% formamide, 10% dextran, 1× Denhardt’s solution, 12 mM EDTA [pH 8.0], 10 mM Tris-HCl [pH 8.0], 300 mM NaCl, 0.5 mg/mL yeast tRNA, and 10 mM dithiothreitol). Hybridization was performed at 65 °C for 12–14 h. The slides were washed in 2× SSPE and 0.1% β-mercaptoethanol at RT for 1 h, 2× SSPE, 50% formamide, and 0.1% β-mercaptoethanol at 65 °C for 1 h, and 0.1× SSPE twice at 65 °C for 30 minutes each. Slides were dehydrated in an ascending ethanol series and exposed to X-ray film (Biomax MR, Kodak) for 1–14 days. We carefully attended not to overexpose X-ray films to S35-riboprobe hybridized brain sections. The slides were then dipped in an autoradiographic emulsion (NTB2, Kodak), incubated for 1–8 weeks, and processed with D-19 developer (Kodak) and fixer (Kodak). Developed slides were Nissl-stained with a cresyl violet acetate solution (Sigma) for the capture of high-resolution images. For quantification of mRNA signal, exposed X-ray films of brain images were digitally scanned under a microscope (Leica, Z16 APO) connected to a CCD camera (Leica, DFC490) with Application Suite V3 imaging software (Leica), as previously described [12,13,19,20]. To minimize handling bias for signal detection among experimental groups, we performed in situ hybridization using multiple brain sections at once for each probe and exposed S35-riboprobe hybridized brain sections on the same sheet of X-ray films. The same light settings were used for all images. Photoshop (Adobe Systems) was used to measure the mean pixel intensities in the brain areas of interest from sections after conversion to 256 grayscale images. For statistical analysis of the expression of each gene, we performed analysis of covariance (ANCOVA) to examine the homoscedasticity from the regression line of the gene induction ratio between singing duration and expression level.
For fluorescence in situ hybridization, dinitrophenyl (DNP)—and digoxigenin (DIG)-labeled riboprobes were used. A total of 100–200 ng of the DNP/DIG-labeled riboprobe was mixed with the hybridization solution (50% formamide, 10% dextran, 1× Denhardt’s solution, 1 mM EDTA [pH 8.0], 33 mM Tris-HCl [pH 8.0], 600 mM NaCl, 0.2 mg/mL yeast tRNA, 80 mM dithiothreitol, and 0.1% N-lauroylsarcosine). Hybridization was performed at 68 °C for 6–13 h. Washing steps were performed as follows: 5× SSC solution at 68 °C for 30 minutes, formamide-I solution (4× SSC, 50% formamide, and 0.005% Tween20) at 68 °C for 40 minutes, formamide-II solution (2× SSC, 50% formamide, and 0.005% Tween20) at 68 °C for 40 minutes, 0.1× SSC 68 °C 15 minutes × 3, NTE buffer at RT for 20 minutes, and TNT buffer × 3, and TNB buffer (0.5% blocking reagent [Perkin Elmer]/1× TNT buffer) at RT for 30 minutes. DNP-labeled probes were detected with an anti-DNP horseradish peroxidase (HRP)-conjugated antibody using a TSA DNP system (Perkin Elmer) and anti-DNP KLH AlexaFluor488 (Molecular Probes, cat#A-11097). Following treatment with 1% H2O2/1× PBS for 30 minutes, DIG-labeled probes were detected with anti-DIG HRP-conjugated antibody (Jackson Laboratory, cat#200-032-156) and a TSA Plus Cy3 system (Perkin Elmer). Signal images were obtained by fluorescence microscopy (EVOS FL; Thermo Fisher Science).
nXIIts of juvenile birds (30–34 phd) was targeted with stereotaxic coordinates in mm: −0.8 rostral, 0.2 lateral, and 5.9–6.0 ventral from the bifurcation of the central sinus at the border of the forebrain and cerebellum. The retrograde tracer DiI (SIGMA, 70 mg/ml dissolved in N, N-dimethylformamide; 100 nl) was injected into the nXIIts 10 days before euthanasia. Birds were humanely killed after 30 minutes of singing, and brains were processed for fluorescence in situ hybridization.
Zebra finch male juveniles (n = 6, 46–55 phd), adults (n = 5, 106–796 phd), and SP adults (n = 5, 100–101 phd) were used for Golgi staining. Brain tissues were sampled under silent and dark conditions and incubated in the impregnation solution from an FD Rapid GolgiStain kit (FD NeuroTechnologies) for 2 weeks in the dark, incubated in a replacement solution for 3 days, embedded in OCT compound (Sakura Fine Technical), and stored at –80 °C until sectioning. Brain sections with a thickness of 100 μm were cut in the sagittal plane. Sections were dried at RT, rinsed with water, and immersed in a staining solution (FD NeuroTechnologies). After staining, sections were dehydrated in an EtOH series, immersed in xylene, and then mounted with Permount. Dendritic spines were counted using photo images taken at 100× magnification with a BZ-X710 Microscope (Keyence). Golgi-stained cell images were obtained by Z-stacking images with 100 planes positioned at a distance of every 0.2 μm. Dendritic spine density was calculated for 3 RA projection neurons and 3 surrounding neurons in the arcopallium for each bird (n = 18 neurons from 6 juveniles, n = 15 neurons from 5 adults, and n = 15 neurons from 5 SP adults). Although only a few of the RA interneurons could be identified by the Golgi staining, RA projection neurons could be distinguished from interneurons by their characteristic mossy dendritic and axonal morphologies (S7B Fig) [36].
Data for song motif consistency were analyzed using an unpaired t test for different conditions and a paired t test with Bonferroni correction for multiple comparisons (Fig 1E). Data for differentially regulated genes in RNA-seq were obtained using DEseq2 with Bonferroni correction and subsequently analyzed using a WGCNA (Fig 2C–2F). Data for induction of singing activity–regulated genes were analyzed using ANCOVA with Bonferroni correction (Fig 4A and 4B and S3–S5 Figs). Dendritic spine density data were compared using an unpaired t test with Bonferroni correction (Fig 5C).
|
10.1371/journal.pntd.0000617 | Serotype-Specific Differences in the Risk of Dengue Hemorrhagic Fever: An Analysis of Data Collected in Bangkok, Thailand from 1994 to 2006 | It is unclear whether dengue serotypes differ in their propensity to cause severe disease. We analyzed differences in serotype-specific disease severity in children presenting for medical attention in Bangkok, Thailand.
Prospective studies were conducted from 1994 to 2006. Univariate and multivariate logistic and multinomial logistic regressions were used to determine if dengue hemorrhagic fever (DHF) and signs of severe clinical disease (pleural effusion, ascites, thrombocytopenia, hemoconcentration) were associated with serotype. Crude and adjusted odds ratios were calculated. There were 162 (36%) cases with DENV-1, 102 (23%) with DENV-2, 123 (27%) with DENV-3, and 64 (14%) with DENV-4. There was no significant difference in the rates of DHF by serotype: DENV-2 (43%), DENV-3 (39%), DENV-1 (34%), DENV-4 (31%). DENV-2 was significantly associated with increased odds of DHF grade I compared to DF (OR 2.9 95% CI 1.1, 8.0), when using DENV-1 as the reference. Though not statistically significant, DENV-2 had an increased odds of total DHF and DHF grades II, III, and IV. Secondary serologic response was significantly associated with DHF (OR 6.2) and increased when considering more severe grades of DHF. DENV-2 (9%) and -4 (3%) were significantly less often associated with primary disease than DENV-1 (28%) and -3 (33%). Restricting analysis to secondary cases, we found DENV-2 and DENV-3 to be twice as likely to result in DHF as DEN-4 (p = 0.05). Comparing study years, we found the rate of DHF to be significantly less in 1999, 2000, 2004, and 2005 than in 1994, the study year with the highest percentage of DHF cases, even when controlling for other variables.
As in other studies, we find secondary disease to be strongly associated with DHF and with more severe grades of DHF. DENV-2 appears to be marginally associated with more severe dengue disease as evidenced by a significant association with DHF grade I when compared to DENV-1. In addition, we found non-significant trends with other grades of DHF. Restricting the analysis to secondary disease we found DENV-2 and -3 to be twice as likely to result in DHF as DEN-4. Differences in severity by study year may suggest that other factors besides serotype play a role in disease severity.
| The four dengue viruses (DENV) represent the most common human arbovirus infections in the world and are currently a challenging problem, particularly in the tropical and subtropical regions of Asia and the Americas. Infection with DENV may produce symptoms of varying severity. While access to care, appropriate interventions, host genetic factors, and previous exposure to DENV are all known to affect the outcome of the infection, it is not entirely understood why some individuals develop more severe disease. It has been hypothesized that the four dengue serotypes differ in disease severity and clinical manifestations. This analysis assessed whether there were significant differences in severity of disease caused by the dengue serotypes in a pediatric population in Thailand. We found significant and non-significant correlations between dengue serotype 2 infection and more severe dengue disease. We also found that individual serotypes varied in disease severity between study years, perhaps supporting the hypothesis that the particular sequences of primary and secondary DENV infections influence disease severity.
| Dengue virus (DENV) is an increasing problem in tropical and sub-tropical countries, where Aedes spp mosquitoes transmit the virus primarily in urban or semi-urban settings. Infection with DENV may result in a sub-clinical infection, undifferentiated fever, dengue fever (DF), dengue hemorrhagic fever (DHF), or dengue shock syndrome (DSS) [1]. Clinical manifestations of DF commonly include fever, rash, hemorrhagic symptoms, headache, ocular pain, arthralgia, myalgia, nausea, and vomiting [2]–[4]. DHF is difficult to differentiate from DF in the early stages of infection and illness [4],[5]. The criteria that differentiate DHF from DF are plasma leakage confirmed by pleural effusion, ascites, and/or hemoconcentration (>20% above patient's baseline), and thrombocytopenia (<100,000/mm3) [2].
While access to care, quality of interventions [6], host genetic factors [7],[8], and previous exposure to DENV [9],[10] are all known to affect outcome, it is not entirely understood why some individuals develop more severe disease. It has been well established that secondary infections and infections in infants with non-neutralizing maternal antibody to dengue are at increased risk of resulting in DHF [11]–[14]. Host genetic determinants of disease severity have been reported, including evidence that black patients may have a lower incidence of DHF compared to other patients [15]–[17]. Some research has indicated that children are more susceptible to developing DHF than adults [2],[12],[18],[19].
It has been observed that different dengue serotypes, and even strains of the same serotype, differ in their propensity to cause severe disease [20],[21]. However, there is no clear consensus on the association of DHF and severe disease with serotype. Earlier analyses of data collected during various DF and DHF outbreaks worldwide have shown associations of DENV-2 [10], [11], [22]–[24] and DENV-3 [13],[21] with increased risk of hospitalization and severe disease. Small sample sizes, lack of all four serotypes, inconsistencies in findings, and short study durations are weaknesses of these studies. Rico-Hesse et al. proposed that virulence of genotypes within serotypes differs, as evidenced by the failure of American DENV-2 in Latin America to produce DHF and the subsequent increase in DHF with the introduction of a Southeast Asian strain of DENV-2 [20]. However, neutralization of American DENV-2 by DENV-1 antibodies may have accounted for the absence of DHF in these cases [25],[26].
We conducted an analysis of data collected at the Queen Sirikit Institute of Child Health (QSNICH) in Bangkok, Thailand to assess whether there were significant differences in severity of disease between the four DENV serotypes in a pediatric population after adjusting for other known risk factors. All four serotypes were encountered, there was consistent application of diagnostic criteria over multiple years, and care was guided by WHO recommendations.
A prospective observational study of dengue disease in children at the QSNICH was conducted from 1994–1997, 1999–2002, and 2004–2006 collaboratively by QSNICH, the Armed Forces Research Institute of Medical Science (AFRIMS), and the University of Massachusetts Medical School. QSNICH is a Thai Ministry of Health facility with 538 beds, including a DHF unit with 30 beds and a dengue holding unit with approximately 20 beds. The hospital is a World Health Organization Collaborating Center for Case Management of Dengue.
The clinical study design has been reported previously [27]. Briefly, children who presented to the outpatient department or who had already been admitted to the hemorrhagic fever ward were considered for study enrollment based on the following criteria: oral temperature ≥37.5°C or rectal temperature ≥38.5°C; history of fever for <72 hours; no obvious source of infection; weight >6 kg; and age 6 months through 15 years. Exclusion criteria included presentation with clinical signs of shock; or serious chronic disease (thalassemia, nephrosis or nephrotic syndrome, cirrhosis, or malignancy). Due to staffing limitations, only six cases per week could be enrolled. The sampling methods were the same throughout the entire period. We perform all the dengue diagnostic testing for the hospital and found the percentages of each serotype (over the same time period) within 3% of those in the study suggesting no bias in serotype inclusion from the sampling.
Children were monitored as in-patients on the dengue ward, where vital signs and weight were routinely measured. Blood for serologic and clinical studies was drawn each morning for a maximum of five consecutive days while the patient was febrile and on the day following defervescence. A follow-up sample for serological diagnosis was obtained from each child five to ten days after discharge. Plasma leakage was assessed for each patient with chest x-ray, physical exam, and hematocrit changes [28]; ultrasound has been studied in a subset of this population, but was not used in this study to classify patients as having ascites, pleural effusion, or DHF. The clinical management of study subjects was directed by the same physician for the study duration.
This study was approved by Human Use Review Committee of the Walter Reed Army Institute of Research, Human Subjects Research Review Board for the Commanding General of the U.S. Army Medical Research and Material Command, University of Massachusetts Medical School IRB, the Thai Ministry of Public Health's Ethical Review Committee for Research in Human Subjects, and the QSNICH IRB. Written informed consent was obtained from the legal guardian of each participant. Secondary data analysis at Mahidol Oxford Tropical Medicine Research Unit and Johns Hopkins University was performed using de-identified data, hence was not deemed human subjects research.
The outcomes of interest were DHF, DF, and DHF severity grades (I–IV). Although outcomes were directed by WHO guidelines, the physician considered the patient's clinical course and treatment in determining final clinical outcomes. Additional outcomes included in the analysis were pleural effusion [measured as both a binary variable and a continuous variable, the pleural effusion index (“PEI” = 100× (maximum width of right pleural effusion)/(maximum width of right hemithorax))] [28], ascites, thrombocytopenia, and hemoconcentration. Explanatory variables used were age, sex, study year, blood type, serological response profile (primary vs. secondary infection), and viral serotype. Dengue EIA results were classified as primary if anti-dengue IgM was greater than or equal to 40 units and anti-dengue IgM∶anti-IgG ratio was greater than or equal to 1.8. Results were classified as secondary if anti-dengue IgM was greater than or equal to 40 units and anti-dengue IgM∶anti-IgG ratio was less than 1.8 [29].
The case definition for DHF included acute febrile illness; hemorrhagic manifestations (positive tourniquet test, petechiae, epistaxis, gum bleeding or gastrointestinal bleeding); thrombocytopenia (platelet count of <100,000/mm3); and plasma leakage evidenced by hemoconcentration ≥20% increase above the patient's baseline, pleural effusion and/or ascites. Cases presenting with the above criteria as well as either narrow pulse pressure (<20 mmHg) or profound hypotension were diagnosed with DSS.
Clinical tests, diagnostic serology, and virologic tests were performed on all study subjects. These measurements included: complete blood count, hematocrit, ALT, AST, blood group, dengue IgG/IgM antibody, hemagglutination inhibition, dengue RT-PCR [30], and virus isolation by mosquito inoculation. Viruses isolated in mosquitoes were serotyped using an antigen capture EIA [29]. Differences between the serotype results from RT-PCR and EIA were extremely rare and resolved with repeat testing.
Only patients with confirmed dengue illness were included in the analysis. Data were entered using FoxPro for Windows software and analysis was performed using Stata/IC 10.0 for Macintosh [31]. Crude odds ratios (OR) and their 95% confidence intervals (CI) were calculated. Student's t test was used to compare mean values of variables and Pearson's x2 test was used to determine significance. Unadjusted and adjusted analyses were performed using univariate and multivariate logistic regression to assess the association of DHF (of any grade) with serotype while controlling for sex, age, blood type, and serological response profile. Multivariate logistic regression was performed by building progressively larger models including Model A (serotype), Model B (serotype, sex), Model C (serotype, sex, age), Model D (serotype, sex, age, primary/secondary infection), and Model E (serotype, sex, age, primary/secondary infection, blood type). Models including interactions between serotypes and all blood types were also assessed [32]. Multinomial logistic regression was used to assess the association between the covariates in models A–E and the multinomial outcome of each case as either DF, DHF I, DHF II, DHF III, or DHF IV (5 possible outcomes).
Additional regressions were performed to investigate the association between severe clinical manifestations and serotype. The parameters included in the analysis were pleural effusion/PEI, ascites, hemoconcentration, and plasma leakage, a summary variable considered positive if any of ascites, pleural effusion, or hemoconcentration was present. A two-tailed P-value <0.05 was considered statistically significant.
Among the 457 children included in this study, ages ranged from 18 months to 15 years with a mean (SD) age of 8.6 (3.0) years (Table 1). The mean (SD) age of children diagnosed with DF was 8.5 (3.0) years, while the mean (SD) age for children with DHF was 8.7 (3.1) years (p = ns). Based on results from graphical (Figure 1) and exploratory data analyses, we included in our candidate models an indicator variable to control for whether subjects were greater or less than 5.4 years of age. The indicator variable allowed us to estimate different effects of increasing age among those less than 5.4 years of age compared to those older. Figure 1 suggests that, before 5.4 years of age, the risk of DHF declines with increasing age, while after 5.4 years of age the risk increases with increasing age. Males and females were represented with comparable frequency (55.5% male).
There were statistically significant differences in disease severity among patients with primary and secondary infections. Of the 457 cases with serological data, 14% of patients with primary infections had DHF, whereas 43% with secondary infections had DHF (OR 4.71 95% CI 2.6, 8.6; Table 1). We also found that the severity of DHF was associated with secondary infection. The odds ratio estimated for secondary infection increased with the severity of DHF; secondary infections were more strongly associated with DHF grades III and IV (15.7 95% CI 2.1, 118.3), than either DHF grade I (OR 2.8 95% CI 0.9, 8.5) or DHF grade II (OR 4.9 95% CI 2.3, 10.4).
DENV-1 was the most commonly found serotype in study subjects, followed by DENV-3, DENV-2, and DENV-4; the eight cases for which viral serotype was unknown were not included in this analysis (Table 2). When adjusting for primary/secondary serological response profile, there was no association found between DHF and serotype (p = 0.295 by Pearson's x2 test). When considering a multinomial response (DF, and the four grades of DHF), DENV-2 was found to be statistically significantly associated with increased odds of DHF grade I compared to DF (OR 2.9 95% CI 1.1, 8.0) when compared to DENV-1. None of the DHF grades were associated with DENV-2 when compared to either DENV-3 or DENV-4. Though not statistically significant, DENV-2 was consistently associated with an increased odds of DHF total (OR 1.5 95% CI 0.9, 2.5), DHF grade II (OR 1.3 95% CI 0.7, 2.4) and DHF grades III and IV (OR 1.4 95% CI 0.5, 3.6) compared to DENV-1. Controlling for a primary or secondary serological response, age, season and sex, DHF grade I was again associated with DENV-2 (OR 3.1 95% CI 1.0, 9.4) compared to DENV-1.
DENV-1 (28%) and DENV-3 (33%) were more often associated with primary DENV infections than DENV-2 (9%, OR 0.2 95% CI 0.08, 0.5) and DENV-4 (3%, OR 0.06 95% CI 0.01, 0.3). Restricting analysis to secondary cases, we found DENV-2 and DENV-3 to be twice as likely to result in DHF rather than DF compared to DEN-4 (p = 0.05, p = 0.04, respectively).
In aggregate across all serotypes, disease was significantly less severe during 1999 (OR 0.3 95% CI 0.1, 0.8), 2000 (OR 0.2 95% CI 0.1, 0.7), 2004 (OR 0.4 95% CI 0.2, 0.1), and 2005 (OR 0.3 95% CI 0.1,0.8) compared to 1994, the year with the highest percentage of severe cases (Table 2). This is not associated with differences in the serotype distribution in these years compared to others.
As was shown previously with a subset of these data, AB blood type showed a significant association with DHF (OR 2.8 95% CI 1.2, 6.3) [32]. There were no significant modification effects when analysis was performed to investigate the interactions between serotypes and blood types.
Secondary DENV infection was associated with severe manifestations of pleural effusion/PEI, ascites, and thrombocytopenia; it was not statistically significantly associated with hemoconcentration (Table 3). Univariate analyses showed correlations of both ascites and pleural effusion index (PEI) with DENV-2 (OR 2.1 95% CI 1.0, 4.1; mean PEI was 1.8 times greater for DENV-2; 95% CI 1.1, 3.1); however, multivariate analyses, when controlling for sex, age, study year, and serological response profile, did not (OR 1.5 95% CI 0.7, 3.1; OR 1.5 95% CI 0.8, 2.6). There were no statistically significant associations of the presence of pleural effusion, hemoconcentration, or thrombocytopenia with serotype of infection in univariate or multivariate analyses.
The results of this prospective observational, single-site study of pediatric dengue showed a significant association between DENV-2 and DHF grade I and nonsignificant associations with other grades of DHF. DENV-2 was also associated with the presence of ascites and larger pleural effusions index. Additionally, we confirmed the previously reported associations between secondary serologic response and DHF as well as a more severe grade of DHF.
Other reports have suggested a correlation between serotype and disease severity. Studies in Thailand and Taiwan found a significant correlation between DENV-2 and greater disease severity [5],[9]. While we found that DENV-2 was most likely to cause DHF (44%) and DENV-4 was less likely (31%) this finding was not statistically significant.
Statistical analysis showed that clinical manifestations of ascites and larger pleural effusions were significantly associated with DENV-2. However, upon adjustment for primary or secondary infection, this association became non-significant. We had limited ability to assess the independent effects of DENV-2 and primary/secondary serological response on the risk of ascites and pleural effusion because such a high percentage of DENV-2 infections were secondary (92%). Several reports have suggested differences in clinical manifestations with distinct dengue serotypes. Vaughn et al., using a subset of data included in this analysis, reported that from 1994–1996, there were more cases of pleural effusion in DENV-2 infections than those infected with the other three serotypes [9]. This was not found to be statistically significant in this larger dataset using multivariate analysis incorporating other factors including secondary infection, but the volume of effusions was still greater with DENV-2 as in the earlier study. Additionally, Balmaseda et al., found an increased incidence of plasma leakage and thrombocytopenia among those with DENV-1 as compared to DENV-2 infections in Nicaragua from 1999–2001 and during an outbreak in 2003 [33]; however, these findings are limited by the short observation period.
Data from this study support the finding that, among those who presented for medical attention, dengue cases caused by DENV-2 and DENV-4 are overwhelmingly secondary infections [34], and there were no cases of DHF caused by primary DENV-2 and DENV-4. This suggests that DENV-1 and DENV-3 are more pathogenic without immune priming from other serotypes. It has been reported that DENV-4 causes more mild disease in primary DENV infections [21],[35]. We found that DENV-2 and DENV-3 were more likely to cause DHF as secondary infections than DENV-4.
Importantly, this analysis found significantly less severe disease in the study population during 1999, 2000, 2004, and 2005, regardless of viral serotype. (Table 2) Several mechanisms could be responsible for this observation. It is hypothesized that the sequence of serotype infections (both first and second infections) influences disease severity. These years may have been less severe because fewer people had some specific dengue virus immunity that placed them at risk of severe disease upon infection with the predominant viruses circulating those years. It is possible that cross-immunity to viruses circulating in prior years affected disease severity. Finally, there may be significant variation in the risk of severe illness upon infection between viruses within serotypes.
The greatest limitation in this study was that all subjects presented for medical attention, but many who would have been treated as outpatients were admitted for the purposes of this study. By including only patients who seek medical attention, those with milder disease or asymptomatic infections are missed. Additionally, with a limited sample size, estimating models with large numbers of covariates is difficult. Further, only a Thai population was captured; as this study group was ethnographically homogenous, it may not be appropriate to apply the study findings in clinical settings to ethnically diverse populations. Another consideration is that QSNICH is in the urban center of Bangkok and may represent only one of several ecological niches for the dengue vector. However, pediatricians caring for children with suspected dengue infections at the QSNICH are recognized as experts in the management of clinical dengue; consequently, parents from Bangkok and the surrounding area are likely to seek care at QSNICH.
Strengths of the study include the long 11–year data collection period, with virus isolation and serotype identification in 98.5% of cases. This extensive study period, spanning various outbreaks, allowed for the representation of all four serotypes, thus providing data for a more complete analysis. Furthermore, the care and evaluations were performed by the same attending physician throughout the study, reducing the possibility of recording and diagnostic errors and inconsistencies. Lastly, the diagnoses were evaluated in greater depth than standard clinical practice.
In this analysis we found some significant correlations between DENV-2 infection and more severe dengue disease, although other associations were not statistically significant. The increased odds of DHF grade I illnesses among DENV-2 infections compared to DENV-1 is difficult to explain, given we did not observe an increase among higher grades of DHF. While this result might be spurious, it was observed in several models, adjusting for slightly different sets of covariates. Overall, the evidence for DENV-2 infections being more severe is a suggestive trend rather than strong evidence. We did find that individual serotypes varied in disease severity between study years, perhaps supporting the hypothesis that the sequence of serotype infections influences the disease severity.
|
10.1371/journal.pntd.0002702 | Determinants of Heterogeneous Blood Feeding Patterns by Aedes aegypti in Iquitos, Peru | Heterogeneous mosquito biting results in different individuals in a population receiving an uneven number of bites. This is a feature of many vector-borne disease systems that, if understood, could guide preventative control efforts toward individuals who are expected to contribute most to pathogen transmission. We aimed to characterize factors determining biting patterns of Aedes aegypti, the principal mosquito vector of dengue virus.
Engorged female Ae. aegypti and human cheek swabs were collected from 19 houses in Iquitos, Peru. We recorded the body size, age, and sex of 275 consenting residents. Movement in and out of the house over a week (time in house) and mosquito abundance were recorded on eight separate occasions in each household over twelve months. We identified the individuals bitten by 96 engorged mosquitoes over this period by amplifying specific human microsatellite markers in mosquito blood meals and human cheek swabs. Using a multinomial model assuming a saturating relationship (power), we found that, relative to other residents of a home, an individual's likelihood of being bitten in the home was directly proportional to time spent in their home and body surface area (p<0.05). A linear function fit the relationship equally well (ΔAIC<1).
Our results indicate that larger people and those who spend more time at home are more likely to receive Ae. aegypti bites in their homes than other household residents. These findings are consistent with the idea that measurable characteristics of individuals can inform predictions of the extent to which different people will be bitten. This has implications for an improved understanding of heterogeneity in different people's contributions to pathogen transmission, and enhanced interventions that include the people and places that contribute most to pathogen amplification and spread.
| We studied the biting habits of Aedes aegypti, the principal vector of dengue virus, to determine why certain people are bitten more often by this day-active mosquito. Over one year in dengue-endemic Iquitos, Peru, we collected blood fed mosquitoes from 19 households. Mosquito blood meals were then matched to household residents using genetic fingerprinting. We found that within a household, larger individuals and those spending more time in the home were bitten more often than other household residents. Importantly, our results show that one's probability of being bitten is dependent on the characteristics of other household residents and visitors. These results indicate that measurable characteristics of individuals do predict who is most exposed to mosquito-borne pathogens, which contributes to our understanding of pathogen transmission processes, informs development of mathematical disease models, and can enhance the design of targeted control programs.
| Mosquito blood feeding behavior is epidemiologically important because of its central role in determining which vertebrate hosts and mosquitoes are exposed to a pathogen. Aedes aegypti, the principal mosquito vector of dengue (DENV) and urban yellow fever viruses [1] is highly anthropophilic, feeding predominantly on people during daylight hours and tending to travel short distances to obtain its blood meals [2], [3], [4], [5]. Females often take more than one blood meal per gonotrophic cycle [6], increasing their probability of (1) imbibing an infected blood meal and (2) after surviving an extrinsic incubation period, becoming infectious, and transmitting virus to an uninfected person [7]. These behaviors lead to the assumption that the risk of DENV infection is highest at the scale of individual locations; the places where female Ae. aegypti feed and people live or visit [8], [9], [10], [11], [12]. Even at this fine scale, however, predicting infection risk remains difficult because some individuals are bitten more often than others for reasons that are poorly understood [13], [14], [15], [16], [17], [18], [19], [20].
A better understanding of who gets bitten more often and why would be useful for designing targeted methods of dengue prevention as well as for developing mathematical models of virus transmission. Although models have traditionally assumed that mosquitoes bite people randomly [21], growing empirical evidence indicates that mosquito biting patterns are heterogeneous and theoretical work indicates that this can have important impacts on transmission dynamics [22], [23], [24]. In particular, people who receive many more mosquito bites than others could act as superspreaders of a pathogen, infecting a disproportionate number of vectors and thus playing a central role in pathogen transmission dynamics [25]. Identifying these people is, therefore, key for effective, targeted disease control strategies [20]. A number of factors have been identified that may make some people more likely to be bitten than others: host body size (larger people being bitten more often), infection with parasites, body temperature, age (perhaps as a proxy for other biological factors), sex, semiochemicals, microflora on the skin, and host movement and defensive behavior [10], [12], [14], [16], [18], [19], [26], [27], [28], [29]. In the case of Ae. aegypti, results from a study conducted in Puerto Rico indicated that people under 20 years of age received fewer bites than those 20 years and older, regardless of gender [15]. There are several plausible explanations for the detected differences, including variation in individual body size and host movement patterns [10].
Our understanding of why some hosts are bitten more often by Ae. aegypti is incomplete, in part, because most studies do not account for the many potentially important differences among human hosts that could influence the chance of receiving a mosquito bite. Variation in biting patterns could be due to differences in inherent attractiveness to mosquitoes, determined by body size or smell, or some other characteristic that has yet to be identified. Observed variation in biting could also be due to the amount of time an individual spends in the same house as biting mosquitoes. We suspect that the most likely explanation combines individual characteristics and exposure time as principal determinants governing which individuals mosquitoes tend to bite most often. For instance, children may receive fewer bites than adults because they are smaller, exposed to fewer mosquitoes during the day or more active than adults. In Iquitos, Peru, for instance, mosquito abundances were found to be very low in schools compared to households [30]. During major portions of the day, when they are at school, children in Iquitos may be physically removed from biting mosquitoes. It is also important to consider the other individuals available at a particular location for mosquitoes to bite. Although mosquitoes may find a given individual suitable for biting, he or she may not be bitten if there are other people in the home that spend more time there or are more attractive to biting mosquitoes. Likewise, if mosquitoes only ever encounter a single individual, they will likely bite that person regardless of how attractive or unattractive they are. Making inferences about the factors that contribute to one's risk of being bitten requires simultaneously accounting for the characteristics of other potential blood meal hosts in the locations where mosquito encounters take place.
In this study, we sought to isolate individual-level factors driving Ae. aegypti biting patterns by identifying which people living in 19 houses in Iquitos, Peru were bitten most often over a 12-month period. The person bitten was determined by DNA profiling of blood in engorged mosquitoes collected inside each house. We then assessed how a number of factors affected each participant's probability of receiving a bite. Our analysis revealed that some individuals are indeed bitten more often than others and that human exposure time and body surface area are associated factors with this heterogeneity.
There was a high correlation between body surface area and age (until adulthood), and between the number of entrances and total weekly time in house. Additionally, the fitted relationship between biting score and either age, entrances, or both was weaker than those with surface area and time in house (age and gender were not significant predictors). Our primary analysis thus only includes time in house and surface area. The aggregated data on time in house and surface area (Figure 2) indicate that the majority of smaller individuals (children) spent more than half of their time in the home. There was no indication that smaller individuals (less than 1 m2 body surface area) that spent less than 50 hours in a week in their home were bitten by an engorged mosquito. Ultimately, however, whether an individual received a bite depended not only on their attributes, but also on the attributes of other residents in their house. In many of the houses in which blood meals were positively identified, larger people and those who spent more time at home tended to be the ones who were bitten (Figure 3).
When weekly time-in-house alone was included in the model as a linear predictor of biting score, the fit was poor (Table 3; LRT: p = 0.247; Fig. 4A). The shallow slope of the fitted curve indicates that individuals who spent little time in a house did not have significantly lower biting scores than those who spent more time in the same house. Surface area, by contrast, was highly significant by itself (LRT: p<0.001; Figure 4B). Combining time in house and surface area improved the model's fit (Figure 4C; lower AIC), but this was not significantly better than the model with surface area by itself (LRT: p = 0.066; Table 3).
Models using power functions gave similar qualitative results to the linear models (Table 4), but somewhat different quantitative results (Figure 5). Weekly time-in-house was still a poor predictor by itself, and with the power functional form there was a sub-linear response (fitted power term 0.322). In other words, as time in house doubles, the biting score less than doubles. As with the linear models, a power function of surface area by itself also did a good job of explaining heterogeneous biting patterns in the data (Table 4; Figure 5B; p<0.001). In contrast to time in house, surface area had a super-linear relationship (fitted power term 1.4), indicating that incremental increases in surface area result in more than equivalent increases in biting score. Combining surface area with time in house again had the best AIC of all power models, and significantly improved model fit (p = 0.038; Table 4; Figure 5C). Biting probabilities predicted by the power model with time in house and surface area are shown in Figure 3 for each house in which a human source of a blood meal was positively identified.
Neither linear nor power functions provided better fits to the data, both having similar optimal AIC scores (321.84 vs 321.286).
Understanding how female Ae. aegypti distribute their bites among human hosts is necessary to develop accurate models that ultimately assist in the design and implementation of more efficacious surveillance and disease control strategies. Our results indicate that, within a given household in Iquitos, Ae. aegypti more often bit larger people and those spending more time in the house, highlighting the importance of human movement behavior in determining individual risk of exposure to the viruses Ae. aegypti transmit. These factors predispose some individuals to receive more bites than others, with potentially important epidemiological effects. For instance, we expect the role of children in transmission to be less during the invasion of a new serotype, when immunologically naïve adults can become infected with, amplify, and transmit the virus. Under endemic transmission, however, infective bites are likely to fall on previously infected and thus immune adults, dampening transmission potential. Although future studies may elaborate on the determinants of heterogeneous biting, our results present a methodological advance in the analysis of DNA profiling data and empirical insight into the causal factors of Ae. aegypti biting and, by extension, DENV transmission.
Previous studies identified human body size as a potentially important predictor of who receives the most bites from anopheline mosquitoes [18]. Explanations include more surface area for biting, easier detectability due to increased CO2 production, a larger heat signature, reduced defensive behavior, and differences in host activity level [7], [26], [27], [28], [29]. Although our study design does not allow us to determine which of these or other mechanisms might explain the pattern we observed in Iquitos, the significance of our result across multiple models for body surface area in a house is consistent with the idea that mosquitoes are following cues (olfactory and/or visual) when selecting a host to feed upon. This effect of body surface area does appear, however, to be modulated somewhat by the amount of time that individuals spend at home. The significant increase of fit in the power model when incorporating both body surface area and time-in-house, and the marginally significant increase in the linear model (p-value = 0.066), are consistent with the hypothesis that people accumulate more bites at a location if they spend more time there. Our results indicate that this effect of total time-in-house is saturating and relatively weak, and other work is suggestive of an even weaker effect whereby frequency of visitation, but not duration, drives exposure to Ae. aegypti bites and infection risk [12]. To clarify what appears to be a nuanced effect of time-in-house on biting risk, we also considered models with more complex representations of time-in-house, but found them to be inconclusive given the available data. In combination, our results suggest that one's risk of being bitten is driven primarily by sensory cues that Ae. aegypti use to detect people. Future work with larger sample sizes and more detailed accounting of time-in-house and movement in and out of the house would help to further resolve the determinants of relative biting risk within a person's home.
As we and others have shown [13], [14], [15], [16], [17], [18], [19], [20], not all hosts have an equal probability of being bitten by mosquito vectors. The assumption of homogeneous biting has historically been used in calculations to determine how difficult an infectious disease is to control [21]. The most common measure of this, the basic reproductive number [37], is predicted to be higher in calculations based on models that allow for heterogeneous biting than in calculations based on models that assume homogeneous biting [22], [23], [24]. This indicates that controlling transmission could be more difficult than predicted by models that assume that all hosts have the same probability of being bitten. If, however, individuals who receive the most bites are identifiable, it may be possible to target interventions and more efficaciously control disease [20].
There were several limitations in our study regarding collection of adequate data for fitting our models. Due to technical issues associated with not being able to fingerprint all of the engorged mosquitoes we collected, we were limited in our ability to test alternative models defining biting risk. This included more complicated relationships between biting and the specific times at which participants were home. Although we had detailed time in house information for human household residents, we did not keep track of non-residents visiting the house or the risk of a resident being bitten at other places they visited during their daily activities [38]. Visitors might have influenced mosquito-biting decisions. The design of our study also precluded us from defining some individual attributes that might independently influence host attractiveness to mosquitoes, such as skin microflora [26].
We were, however, able to isolate important effects that influence how Ae. aegypti bites are distributed among its natural human hosts. Doing so required introducing a new statistical framework for assessing the contributions of different personal factors to one's relative risk of being bitten. Follow-up studies on Ae. aegypti or other household-biting mosquitoes should similarly account for the time people spend in a house and weight each individual's risk relative to other household residents. In particular, our results validate previous studies pointing to adults and/or larger people as the primary recipients of mosquito bites and underscore the importance of the time people spend at a location where mosquitoes bite. Moreover, our analyses reveal that the relationships between such factors can have nonlinear effects on an individual's risk, with time in house having a sub-linear effect and body surface area having a super-linear effect. More detailed understanding of these and other factors that contribute to an improved understanding of biting risk will be an important component of efforts to target interventions, such as vaccines for dengue virus that are currently under development.
|
10.1371/journal.pntd.0003114 | Re-emergence of Rabies in the Guangxi Province of Southern China | Human rabies cases in the Guangxi province of China decreased from 839 in 1982 to 24 in 1995, but subsequently underwent a sharp increase, and has since maintained a high level.
3,040 brain samples from normal dogs and cats were collected from 14 districts of Guangxi and assessed by RT-PCR. The brain samples showed an average rabies virus (RV) positivity rate of 3.26%, but reached 4.71% for the period Apr 2002 to Dec 2003. A total of 30 isolates were obtained from normal dogs and 28 isolates from rabid animals by the mouse inoculation test (MIT). Six representative group I and II RV isolates showed an LD50 of 10−5.35/ml to 10−6.19/ml. The reactivity of monoclonal antibodies (MAbs) to group I and II RV isolates from the Guangxi major epidemic showed that eight anti-G MAbs showed strong reactivity with isolates of group I and II with titers of ≥10,000; however, the MAbs 9-6, 13-3 and 12-14 showed lower reactivity. Phylogenetic analysis based on the G gene demonstrated that the Guangxi RV isolates have similar topologies with strong bootstrap values and are closely bonded. Alignment of deduced amino acids revealed that the mature G protein has four substitutions A96S, L132F, N436S, and A447I specific to group I, and 13 substitutions T90M, Y168C, S204G, T249I, P253S, S289T, V332I, Q382H, V427I, L474P, R463K Q486H, and T487N specific to group II, coinciding with the phylogenetic analysis of the isolates.
Re-emergence of human rabies has mainly occurred in rural areas of Guangxi since 1996. The human rabies incidence rate increased is related with RV positive rate of normal dogs. The Guangxi isolates tested showed a similar pathogenicity and antigenicity. The results of phylogenetic analysis coincide with that of alignment of deduced amino acids.
| Rabies is a worldwide zoonosis disease and is of considerable public health threat and hazard. The Guangxi province of southern China is a severe rabies epidemic region. Human rabies cases decreased from 839 in 1982 to 24 in 1995 in Guangxi as a result of a dog vaccination campaign. However, the number subsequently underwent a sharp increase, and has since maintained a high level. This study reports the systematic surveillance of rabies in Guangxi over the 30-year period from 1982 to 2012. The data revealed that a re-emergence of human rabies has occurred mainly in rural areas of Guangxi since 1996. Human rabies incidence rate increased follows increased instances of RV positive normal dogs. To further understand this re-emergence of rabies, the biological properties of the rabies virus (RV), including the RV-positive rate of normal dogs, pathogenicity, antigenicity and evolution, have been evaluated. The Guangxi isolates all showed similar pathogenicity and antigenicity. These isolates also exhibited similar topologies with strong bootstrap values in the two groups and were closely bonded. Thus these findings will be helpful to understanding the epidemiological situation for rabies in Guangxi.
| Rabies is a fatal enzootic viral infection of the central nervous system. The disease is widespread throughout the world, and is a serious public health problem in developing countries. The WHO reported that human mortality from endemic canine rabies is estimated to be 55000 deaths per year in Asia and Africa, with 56% of these deaths occurring in Asia. The majority (84%) of these deaths occur in rural areas [1]. Dogs are the principal host of the rabies virus and play a primary role in rabies transmission in Asia. More recently, several reports on the molecular epidemiology of rabies have been published from Asian countries, such as Thailand [2], Indonesia [3], South Korea [4], and China [5]–[6].
The rabies virus (RV) is a member of the Lyssavirus genus and is distributed in a wide range of host species. RV has been extensively studied because of its significant impact on public health, especially considering that it is fatal in people. In China, epidemiological surveillance has shown a re-emergence of human rabies since 1995. Using molecular characterization based on the genetic diversity of the RV isolates, two distinct clades of RV were identified in 2004 [7], [8]. Furthermore, investigation of the molecular epidemiology of RV in southern China demonstrated that the long-distance migration, or transprovincial movement of dogs by humans from high-incidence regions may be one of the causes for the re-emergence of the disease [9]. Evolutionary dynamic analysis of RV based on the G gene [10] showed that the RV currently circulating in China is composed of three main groups and that the rabies viruses in China and Southeast Asia share a common ancestor [11].
Guangxi province is a severe epidemic region of rabies. The human cases of death due to rabies in Guangxi since 1997 rank the highest in China. More than 100 people have died of rabies each year since 2000, with a peak of 602 cases of death in 2004. Phylogenetic analysis based on the 3′-terminus of the N gene showed that RV isolates from Guangxi can be divided into four groups [12], although only two (I and II) are major causative factors of lethal rabies in humans and animals. In this study, we summarize the recent trends in the epidemiological characteristics, antigenicity, pathogenicity and phylogeny of street RV isolates that are highly prevalent in Guangxi.
All animals experiments described in this paper were conducted according to the National Guideline on the Humane Treatment of Laboratory Animals Welfare (MOST of People's Republic of China, 2006) and approved by the Animal Welfare and the Animal Experimental Ethical Committee of Guangxi University (No. Xidakezi2000138). All husbandry procedures were conducted in compliance with the Animal Welfare Act and the Guide for the Care and Use of Laboratory Animals.
The sampling and collection protocol were approved by the Veterinary Administration of Guangxi. The brain samples of normal dogs and cats were collected and provided by the Guangxi Centre for Animal Disease Control and Prevention (the Animal CDC). The brain samples were collected into sterile plastic tubes. Mice used for viral isolation by the mouse inoculation test (MIT) were purchased from the Animal Centre of Guangxi Medical University. The mice were observed for 28 days post-injection, and then were euthanized in a container by halothane inhalant.
For the surveillance of rabies in Guangxi, all counties reported the number of rabies cases in humans to the Guangxi Centre for Disease Control (CDC) and the Animal CDC, an Executive Department of the Guangxi Government. Most counties submitted data monthly, and all data were confirmed at the end of the year by the local CDC. The human rabies cases were diagnosed by histories of animal bites, scratches, or exposure to animal body fluids; and by clinical symptoms: hydrophobia or aerophobia, hypersalivation, excitation, and then paralytic signs in limbs, loss of incoordination, and tremors. Finally, most of the rabies cases were verified by indirect fluorescence assay (IFA).
From May 1999 to July 2010, 28 brain samples were obtained from rabid dogs, cattle, and pigs that were clinically suspected to have rabies; and 3040 brain samples, including 3032 normal dogs and 8 cats, were collected from different regions of Guangxi. All samples were provided by the Animal CDC as part of routine laboratory investigations for suspected cases. Samples were subjected to RT-PCR, and the positive samples were further used for RV isolation by MIT [12]. The mice were purchased from the Animal Centre of Guangxi Medical University. To comply with the “Animal Research: Reporting In Vivo Experiments” (ARRIVE) guidelines, all husbandry and experimental procedures were conducted in compliance with the Animal Welfare Act and the Guide for the Care and Use of Laboratory Animals.
Total viral RNA was extracted from brain samples using Trizol (Invitrogen, CA, U.S.) following the manufacturer's instructions. cDNA was synthesized using 2.5 µg total RNA, 1 µl (25 pMol/µL) sense primer, and 100 U MuMLV reverse transcriptase (Promega) in a 25 µl reaction volume. Each viral gene was amplified by PCR using ExTaq DNA polymerase.
The virulence of RV isolates was determined using four-week-old adult mice. The experiments were performed in microbiological safety cabinets in a biological safety laboratory (Group P2+). Dilution series of virus stock were prepared using DMEM in an ice bath. A total of 0.03 ml of each virus dilution was injected cerebrally into each adult mouse, using 4 mice per dilution. The mice were clinically observed for 28 days. Any death occurring within first 5 days was considered non-specific. The specific signs and symptoms of rabies were recorded as humane endpoints: ruffling of hair, loss of coordination, tremors, paralysis of fore or hind limbs, hind limb palsy, body weight loss, and convulsions. The mice with rabies clinical signs were euthanased. The total number of specific deaths for each dilution was calculated as the LD50 by the Reed-Muench method [13].
Neuro-2A (NA) cells (ATCC number: CCL 131) were grown in Dulbecco's Modified Eagle Medium (DMEM) containing 10% fetal calf serum. BHK-21 cells (ATCC number: CCL-10) were maintained in DMEM supplemented with 10% fetal calf serum.
The brains of original animals or adult mice infected with the street RV isolates were used to prepare sections. The sections were fixed with cold 30% acetone-70% methanol for 1 h and then stained with anti-G MAbs (kindly provided by Dr. Minamoto [14] of Gifu University, Japan) by IFA. The results were also verified in NA or BHK cells. FITC-conjugated goat antibody against mouse immunoglobulin was purchased from Sigma-Aldrich Corp (USA).
The fragment of 1815 nucleotides containing complete G gene was amplified using a pair of primers GP1: 5′ ATC CCT CAA AAG ACT CAA GG 3′ (3293∼3314) and GP2: 5′ CCG TTA GTC ACT GAA ACT GC 3′ (5088∼5107) with cycling conditions of 95°C for 5 min; 35 cycles of 95°C for 1 min, 50°C for 1 min, and 72°C for 1 min. PCR products were purified and cloned into pMD18-T vector, and then sequenced by Takara Corp. Three clones were analyzed for each amplicon of each virus. Whenever clonal differences were identified, the other three clones were analyzed repeatedly until a consistent sequence was obtained. Sequence information was aligned and edited using the Vision X program.
The coding regions of the G gene in the genome dataset of the isolates (accession numbers in Table S1) were modeled for phylogenetic tree reconstruction as described previously [12]. Calculation of the homology of nucleotide sequences was carried out using genetic software (Windows version 6.0.1). Alignments of homologous sequences were performed with the Clustal method of the MegAlign program of the DNAStar version 7.1 package (DNASTAR Inc., U.S.). A neighbor-joining (NJ) tree for all DNA sequences was constructed using the Kimura 2-parameter model with MEGA4.0 software [15].
Since the first rabies outbreak during the 1950s and 1960s, Guangxi has undergone two re-emergences of rabies during the past three decades: one occurred in the 1980's, and we are currently in the midst of the second. The epidemiological characteristics of rabies in humans in Guangxi from 1982 to 2012 are shown in Figure 1A. The incidence of human rabies in Guangxi decreased from 839 in 1982 to 24 in 1995, but more than doubled to 50 cases the following year. Seven years later, in 2002, the human rabies cases increased sharply to 203, which is more than 8 times that of the incidence in 1995. In 2003, the incidence more than doubled again, with 519 cases of human rabies. A peak of 602 cases was observed in 2004, and has decreased gradually since then, but remains in the 200 s range. This geometric pattern of increased incidence over the ten years from 1995 to 2004 was observed in the whole of China; although the Guangxi province was the most serious epidemic region (Figure 1B and C).
To investigate the underlying cause of the RV outbreaks, 28 brain specimens were obtained from 20 dogs, six cattle and two pigs suspected of having rabies. The 28 brain specimens were verified to be RV positive by RT-PCR (100%), and 28 corresponding RV isolates were obtained by MIT. In addition, 3040 brain samples from 3032 normal dogs and 8 cats were collected from different areas of Guangxi from 1999 to 2010. 99 of the brain samples from normal dogs were determined to be RV-positive by RT-PCR. The RV positive rate in Chongzuo was highest, account for 5.41%; Nanning and Wuzhou were 4.18% and 4.0%, respectively. All districts with RV positive rate ≧3.0% including Beihai, Yulin, Qinzhou, Liuzhou and Baise locate in Guangxi (Fig. 2A). Of the 3040 samples, the RV positive rate was 3.26%, and the highest RV-positive rate was found during 2002–2005, with 59 of the 1301 brains from normal dogs collected from April 2002 to April 2005 being positive (4.39–4.71%) (Table 1).
A correlation between human rabies cases and detection of RV in animals was found: human rabies incidence rate increased following increased instances of RV positive rate of dog brain samples (Fig. 2B). For example, from 1999 to 2001, dog brains were 1.12–1.19% RV positive, and human rabies cases in Guangxi were 64, 79, and 138 (be equivalent to 0.136, 0.167 and 0.288 per 105 population), respectively. However, from 2002 to 2005, RV positive brain samples increased to 4.71–4.39%, and the human rabies cases reached 203, 519, 602 and 480 (be equivalent to 0.421, 1.069, 1.231 and 0.975 per 105 population), respectively for these years. Since 2006, the percentage of RV positive brain samples have declined to 2.28–2.86%, and the human rabies cases have decreased correspondingly.
To determine the virulence of the isolated RV strains, six representative Guangxi isolates were tested using adult mice. The mice were cerebrally injected with diluted brain sample emulsions, 30 µl for adult mice (4 mice per group), and the clinical signs were scored post inoculation (Table 2).
In adult mice, all isolates displayed a similar average incubation period of 4–6 d. The shortest incubation period was 4 d in mice infected with GXNN2. The course of sickness for the isolates also displayed a tight range from 1.33–4.5 d. The first signs of agitation appeared on 5 dpi, and the first paralytic signs of the hind limbs appeared at 6 dpi. All mice showed typical rabies signs: ruffled fur, lack of coordination, tremors, and paralysis of fore or hind limbs. The mice with typical rabies symptoms were euthanased. The 50% lethal dose (LD50) was calculated as 10−5.35/ml to 10−6.19/ml in adult mice (Table 2). These results suggest that all of the isolates tested caused a similar course of pathogenicity in mice.
To evaluate the antigenicity of the RV isolates from Guangxi, 11 anti-G MAbs were used to determine the reactivity by IFA. Several representative isolates, including those of groups I and II, were selected for antigenicity testing. Among the anti-G MAbs, 8 showed strong reactivity with group I and II isolates, as well as the control strains RC-HL and ERA, with titers of ≥10000 (+++). However, the MAbs 12-14, 9-6, and 13-3 showed lower reactivity (100–1000 fold, +) with all isolates of group I and II, and also showed somewhat reduced reactivity with the control strains (Table 3). These results are suggestive of similar reactivity among the Guangxi isolates.
To investigate the origins of the Guangxi RV isolates, we selected 25 for sequencing and phylogenetic analysis, including representatives from a variety of districts and collection dates spanning the period from 2000 to 2007 (Table S2). The G gene was sequenced and aligned using ClustalW and subjected to phylogenetic tree reconstruction with the neighbor joining method [16].
Group I included 15 isolates with nucleotide homology of 97.6–99.9% and deduced amino acid homology of 98.1–100%. Group II included 10 isolates with nucleotide homology of 98.1–99.9% and deduced amino acid homology of 97.7–100%. However, between groups I and II, the homology for nucleotide and for deduced amino acid were 86.8–87.7% and 93.5–94.9%, respectively (Table S3). These results imply that group I obviously differs from the group II in evolution.
A maximum likelihood phylogenetic tree was constructed for 133 complete G sequences (Figure 3). The Chinese isolates were independently divided into two major clusters, groups I and II, and were generally separate from the isolates of other countries. Isolates from the Guangxi, Hunan, and Guizhou provinces were of both groups I and II. It is noted that isolates from Guangxi exhibited similar topologies with strong bootstrap values in the two groups and were closely bonded. Group I contains isolates from Guangxi, Hunan, Guizhou, Fujian, Ningxia, Zhejiang and Jiangxi provinces, whilst group II contains isolates of from Guangxi, Hunan, Guizhou, Anhui, Jiangsu, Henan, and Yunnan province. The phylogeny indicated that the virus might be introduced from other provinces.
Comparison of the deduced amino acid sequence of the entire G gene of the isolates with that of the ERA strain, a vaccine for which is most commonly applied in Guangxi, showed several substitutions that distinguish the strains. Based on the G protein variations, our analyses revealed two specific substitutions, F-6V/I and V-7A, in the signal peptide for group I and three specific substitutions, L-4S, P-5S, and A-15V, for group II. However, the mature G protein was noted to contain 20 substitutions. Of which, four substitutions (A96S, L132F, N436S, and A447I) specific to group I, 13 substitutions (T90M, Y168C, S204G, T249I, P253S, S289T, V332I, Q382H, V427I, L474P, R463K Q486H, and T487N) specific to group II (Table 4). These results support the classification of the strains based on the nucleotide sequence.
Guangxi is a severe epidemic region for rabies, with the highest rates in China. Over the past decade, 200–600 people in Guangxi have died of rabies each year (Fig. 1A). Domestic dogs are the principal vector, and 95% of human cases are associated with dog transmission. The major problems in controlling rabies are: 1) The understanding of rabies is extremely poor in rural areas; 2) People bitten by dogs often do not report the incidence and obtain treatment; 3) Dog populations in Guangxi are rising and currently estimated at more than 5 millions about 10% of the human population.
An outbreak of rabies in Guangxi occurred in the 1970s and reached a peak of 877 human cases in 1981. After the introduction of rabies vaccination for dogs, and increased efforts to eradicate stray dogs, the incidence of human rabies decreased to 24 cases in 1995. However, in 1996 the number of human rabies cases increased to 50, rose steeply to 203 in 2003, and reached a peak of 602 in 2004. A high rate of rabies incidence in Guangxi persists, although there has been a gradual decrease since 2004 (Fig. 1A). Nevertheless, it is clear that a re-emergence of rabies has occurred in Guangxi from 1996. However, the cases have been mainly limited to rural areas because pets in the cities tend to receive effective vaccination.
The rabies virus does not have carrier status [17]. However, there will be a degree of replication within the brains of animals at a pre-clinical stage of infection prior to the onset of clinical symptoms. These infected animals at a pre-clinical stage of infection probably secret RV via saliva and can transmit it to humans or other animals by biting. Thus, it is very important to understand whether or not the positive rate of the infected animals at a pre-clinical stage of infection relates to human cases of death. We collected 3040 brain samples from 3032 normal dogs and 8 cats from different areas of Guangxi between 1999 and 2010. Of these, 99 samples from normal dogs were determined to be RV-positive by RT-PCR, and found that human rabies cases increased correlated with RV in normal dogs (Fig. 2B). From the RV-positive samples, 30 RV isolates were obtained by MIT. Several isolates from rabid and normal dogs taken at different times showed similar pathogenicity in mice, indicating that the RV from Guangxi has stable virulence. It is noteworthy that the 99 positive samples were from normal dogs with no clinical symptoms, suggesting that normal dogs in Guangxi have a positivity rate of RV of 3.26%.
Over the past two decades, the use of MAbs for lyssavirus identification has significantly expanded the ability to differentiate individual viruses with reproducible results. We used 10 anti-N and 11 anti-G MAbs to assess the antigenicity of six representative RV isolates from Guangxi. The results showed that the isolates of group I/II that are mainly prevalent in Guangxi have similar reactivity patterns for the N protein (Data not shown). However, a small distinction of anti-G MAbs to the RV isolates occurred, in that the MAbs 12-14, 9-6 and 13-3 showed a weaker reactivity to the Guangxi isolates than the control RV strains. The reactivity patterns with anti-N/G MAbs suggests that the antigenicity of the Guangxi isolates is stable, but may contain some unique regional features.
Molecular typing to differentiate lyssavirus strains has been performed primarily on the N gene in order to evaluate concordance with former classifications by serotyping [18]. The N gene of RV is the most common target for genetic and adaptive evolution analysis because the gene is highly convergent [19]–[22]. For the Guangxi isolates, phylogenetic analysis of the N gene showed similar topologies among isolates with strong bootstrap values that were closely bonded. Alignment of the deduced amino acids revealed ten specific amino acid mutations on the N protein, coinciding with the phylogenetic analysis of the isolates (Data not shown).
Phylogenetic analysis and amino acid comparison demonstrated similar overall results for the RV isolates from Guangxi. The phylogenetic analysis confirms the findings of preliminary surveys, but also indicates that the major groupings can be explained by geographical parameters. Variations of the G protein demonstrated patterns similar to those obtained for the N protein (data not shown). Furthermore, RNA variations in the RV isolates from Guangxi are consistent with differences observed from other geographical regions. These finding suggest that there are strong geographical associations among RV isolates that might have contributed to the re-emergence of the disease in Guangxi.
Amino acid comparison demonstrated, those mutated amino acids in G protein do not specific to functional domains such as antigenic sites II (aa 34–42, and aa198–202) [23] and III (330–340) [24], MAbs sites (epitope sites 147, 184, 251, 263 and 264) [25]–[29], snake venom curareminetic neurotoxin (aa189–214) [30], and pathogenic sites (aa 37, 242, 255, 268 and 333) [31]–[33]. These results signified that the key amino acids in the functional domains were conserved.
Overall, a more precise and thorough documentation of confirmed rabies cases in humans and animals would give a better understanding of the epidemiological situation in the area. In view of above-mentioned reasons, people should maintain a high degree of vigilance when the stray dogs and cats approach.
|
10.1371/journal.pbio.1001467 | Integration of Canonical and Noncanonical Wnt Signaling Pathways Patterns the Neuroectoderm Along the Anterior–Posterior Axis of Sea Urchin Embryos | Patterning the neuroectoderm along the anterior–posterior (AP) axis is a critical event in the early development of deuterostome embryos. However, the mechanisms that regulate the specification and patterning of the neuroectoderm are incompletely understood. Remarkably, the anterior neuroectoderm (ANE) of the deuterostome sea urchin embryo expresses many of the same transcription factors and secreted modulators of Wnt signaling, as does the early vertebrate ANE (forebrain/eye field). Moreover, as is the case in vertebrate embryos, confining the ANE to the anterior end of the embryo requires a Wnt/β-catenin-dependent signaling mechanism. Here we use morpholino- or dominant negative-mediated interference to demonstrate that the early sea urchin embryo integrates information not only from Wnt/β-catenin but also from Wnt/Fzl5/8-JNK and Fzl1/2/7-PKC pathways to provide precise spatiotemporal control of neuroectoderm patterning along its AP axis. Together, through the Wnt1 and Wnt8 ligands, they orchestrate a progressive posterior-to-anterior wave of re-specification that restricts the initial, ubiquitous, maternally specified, ANE regulatory state to the most anterior blastomeres. There, the Wnt receptor antagonist, Dkk1, protects this state through a negative feedback mechanism. Because these different Wnt pathways converge on the same cell fate specification process, our data suggest they may function as integrated components of an interactive Wnt signaling network. Our findings provide strong support for the idea that the sea urchin ANE regulatory state and the mechanisms that position and define its borders represent an ancient regulatory patterning system that was present in the common echinoderm/vertebrate ancestor.
| The initial regulatory state of most cells in many deuterostome embryos, including those of vertebrates and sea urchins, supports anterior neural fate specification. It is important to restrict this neurogenic potential to the anterior end of the embryo during early embryogenesis, but the molecular mechanisms by which this re-specification of posterior fate occurs are incompletely understood in any embryo. The sea urchin embryo is ideally suited to study this process because, in contrast to vertebrates, anterior–posterior neuroectoderm patterning occurs independently of dorsal-ventral axis patterning and takes place before the complex cell movements of gastrulation. In this study, we show that a linked, three-step process involving at least three different Wnt signaling pathways provides precise spatiotemporal restriction of the anterior neuroectoderm regulatory state to the anterior end of the sea urchin embryo. Because these three pathways impinge on the same developmental process, they could be functioning as an integrated Wnt signaling network. Moreover, striking parallels among gene expression patterns and functional studies suggest that this mechanism of anterior fate restriction could be highly conserved among deuterostomes.
| Wnt signaling pathways play fundamental roles in many developmental processes. One of the earliest and most crucial of these roles is the activation of gene regulatory programs that specify different cell fates along the embryo's primary anterior–posterior (AP) axis. Recent comparative analyses suggest that Wnt/β-catenin signaling is an ancient AP patterning mechanism that establishes posterior identity in most metazoan embryos [1]–[9]. In invertebrate deuterostome embryos, which include cephalochordates, urochordates, hemichordates, and echinoderms, localized determinants cause stabilization of β-catenin in posterior blastomeres. This stabilized β-catenin enters nuclei in which it activates genes that specify endomesoderm, marking the site of gastrulation at what corresponded to the vegetal pole of the egg and forming the posterior end of the developing embryo [2],[4],[10],[11]. In the sea urchin embryo, in which the molecular mechanisms of endomesoderm specification are best understood [12], the first evidence of Wnt signaling after fertilization is the presence of β-catenin in the nuclei (nβ-catenin) of posterior cells, beginning at the 16- to 32-cell stage. During the next few cleavages, a detectable gradient of nβ-catenin forms in the posterior half of the embryo, with the highest concentration at the posterior pole [4]. This gradient of Wnt/β-catenin signaling is both necessary and sufficient to activate the gene regulatory networks that establish mesoderm and endoderm cell fates in a posterior-to-anterior wave during late cleavage stages [13],[14].
Wnt/β-catenin signaling also transforms the initial regulatory state that specifies anterior neuroectoderm (ANE) development in those deuterostome embryos in which it has been examined [15]–[22]. In the sea urchin embryo, we refer to this neuroectoderm as ANE because it becomes restricted to a region derived from the animal pole of the egg, which is located opposite to the posterior end of the embryo (see [23]). The initial regulatory state of early sea urchin embryos activates ANE specification by the 32-cell stage, when genes encoding the earliest ANE regulatory proteins are expressed broadly throughout the anterior half of the embryo [22],[24]. These early factors include Six3, which is expressed at the anterior end of bilaterian embryos [25] and has been shown by functional studies to be critical for the specification of anterior-most neuroectoderm in diverse embryos including Tribolium castaneum [26], sea urchins [24], zebrafish, and mouse [27]. Beginning around the 60-cell stage, a progressive posterior-to-anterior down-regulation of ANE factor gene expression in most of the anterior half of the embryo occurs by an unknown mechanism that requires posterior Wnt/β-catenin signaling [22]. This process continues during blastula stages until the ANE regulatory state is confined to a disk of cells around the anterior pole of the mesenchyme blastula [24]. Interestingly, an unknown signal from posterior Wnt/β-catenin signaling also appears to be necessary to pattern the anterior ectoderm along the AP axis in Saccoglossus kowalevskii [2], which belongs to the hemichordates, a sister clade to echinoderms.
Remarkably, Six3 activates a large cohort of genes in the sea urchin ANE that are orthologs of genes expressed in the vertebrate ANE (forebrain/eye field) (Figure 1A), raising the possibility that the common ancestor of sea urchins and vertebrates may have shared this ANE regulatory program [24]. Similar to the sea urchin embryo, an initial widespread regulatory state in late blastula/early gastrula stages of vertebrate embryos supports expression of genes encoding early anterior forebrain/eye field factors throughout the presumptive neuroectoderm, including six3 [28],[29]. Simultaneously, secreted antagonists from the organizer block bone morphogenetic protein (BMP) signaling on the dorsal side and a high-to-low, posterior-to-anterior gradient of nβ-catenin forms in the presumptive neuroectoderm. This Wnt/β-catenin signaling gradient is part of a mechanism that activates genes encoding posterior neuroectoderm factors while down-regulating anterior (forebrain/eye field) factors in the posterior neuroectoderm [16],[18],[21]. By these mechanisms, the neural plate is formed and expression of the presumptive forebrain/eye field factors is restricted to cells at its anterior end, where Wnt antagonists protect them from posteriorization [30]–[33]. Multiple Wnts, Fzl receptors, and Wnt antagonists (i.e., Wnt8, Wnt3a, Wnt1, Fzl8, and Dkk1) have been implicated in posteriorization of the neuroectoderm in vertebrate embryos, as well as members of the fibroblast growth factor (FGF), retinoic acid (RA), and transforming growth factor–beta (TGF-β) signaling pathways [28],[30],[34],[35]. However, the exact functions of these pathways in AP neuroectoderm patterning have been difficult to determine because of their earlier functions as well as the complex cell movements of gastrulation during this process [28],[30],[36]. Moreover, the interactions among these various pathways in AP neuroectoderm patterning are not well understood.
Recent studies suggest that early AP neuroectoderm patterning in vertebrate embryos is independent of information from the dorsal organizer. In both Xenopus and zebrafish, the initial widespread regulatory state promotes neuroectoderm specification throughout most of the embryo in the absence of β-catenin, which blocks dorsal organizer formation, as well as BMP2, BMP4, and BMP7. Expression of neuroectoderm markers is radialized around the AP axis in these embryos, but remarkably they retain normal AP neuroectoderm patterning, and the ANE expands posteriorly when both maternal and zygotic Wnt/β-catenin function is blocked [20],[21]. Interestingly, in the sea urchin embryo, the action of Wnt/β-catenin signaling in early patterning of the neuroectoderm along the AP axis [22] is also separate and distinct from the dorsal-ventral patterning mechanism because it occurs before Nodal and BMP signaling are activated and is required for their expression [37]–[40]. The inhibition of Wnt/β-catenin signaling, and consequently the loss of expression of Nodal and BMP, causes a large majority of cells to differentiate into ANE [22],[24],[41]. Thus, the developmental regulatory mechanisms used by vertebrate embryos for ANE development have striking similarities to those used by sea urchin embryos and may therefore represent an ancestral deuterostome mechanism.
Here we show that the Wnt-dependent restriction of neuroectoderm to the anterior pole involves not only Wnt/β-catenin but also a series of linked steps mediated by Wnt/JNK signaling through Wnt1, Wnt8, and Fzl5/8, the homolog of vertebrate Fzl8. Coordinated progression of signaling through these Wnt pathways and activation of the secreted Wnt antagonist Dkk1 in anterior-most blastomeres establish the definitive ANE around the anterior pole. Signaling through a second Wnt receptor, Fzl1/2/7, and its activation of PKC suppress Wnt/β-catenin and Wnt/JNK ANE restriction activities to coordinate the correct temporal progression of ANE restriction. Collectively, signaling through three different Wnt signaling pathways provides precise spatiotemporal control of neuroectoderm AP patterning along the AP axis.
FoxQ2 and Six3 are essential for the specification of the ANE and are the earliest ANE regulatory genes to be expressed. Their transcripts accumulate in the anterior half of the 32-cell embryo but are never detectable in the posterior half (Figure 1B,C) [22],[24]. We reasoned that posterior repression might depend on Wnt/β-catenin signaling because this pathway is activated in posterior blastomeres by the 16-cell stage [4],[14],[42],[43]. To test this possibility, we blocked nβ-catenin by injecting embryos with mRNA encoding either Tcf-Engrailed (Tcf-Eng) [42] or Axin [44] and examined foxq2 expression at the 32-cell stage (Figure 1B,F; Figure S1). In both cases, foxq2 and six3 were expressed in every blastomere during early cleavage stages (32-cell, Figure 1F; 120-cell, Figure 1G) and ubiquitous expression persisted until late mesenchyme blastula stage (24 hpf) (Figures 1G,H and S1Ae–g). As expected, each perturbation resulted in formation of dauer blastulae with a thickened neuroepithelium covering most of the embryo that produced greatly increased numbers of serotonergic neurons throughout (Figure 1I versus E; Figure S1Ah versus Ad). These 4-d embryos phenocopied ΔCadherin mRNA-injected embryos, which previously were shown to lack nβ-catenin in all but the four vegetal-most blastomeres, the small micromeres during cleavage stages [24]. Together, these data indicate that the factors that activate ANE specification operate in all early blastomeres in these Wnt/β-catenin-deficient embryos and likely are part of a ubiquitous maternal regulatory state. Moreover, these observations indicate that the first step in suppressing the ANE in the posterior half of the embryo depends on the repression or rapid down-regulation of ANE regulatory gene transcription by Wnt/β-catenin signaling.
Previous studies have shown that restriction of foxq2 expression to the anterior pole depends on posterior Wnt/β-catenin signaling. However, Wnt/β-catenin signaling has never been detected in the anterior half of the embryo (the presumptive ectoderm, blue in Figure 1Ja), suggesting that an intermediate signal(s) downstream of posterior Wnt/β-catenin signaling must mediate this second phase of ANE restriction (Figure 1Jb versus Jc; the gray region in this and subsequent figures represents the posterior ectoderm and the orange arrows indicate the second phase of restriction). We hypothesized that this intermediate signal (Figure 2C, signal X) might also involve Wnt signaling, and we tested this idea by exploring the functions of the Wnt [Frizzled (Fzl)] receptors in ANE restriction. Two of the four sea urchin receptors, Fzl5/8 and Fzl1/2/7, were expressed during ANE restriction (Figure S2A) and also in the appropriate cells to mediate this process (Figure S2Ba–h), making them excellent candidates for transducing Wnt signals that eliminate the ANE regulatory state from the posterior ectoderm (Figure S2Bi–Bl).
To determine whether Fzl5/8 signaling has a role in neuroectoderm AP patterning, we injected embryos either with morpholinos targeting Fzl5/8 or with mRNA encoding a C-terminal truncated form of the receptor (ΔFzl5/8) that acts as a dominant negative by competing for Wnt ligands [45]. In contrast to embryos injected with Axin or Tcf-Eng mRNA, those expressing ΔFzl5/8 mRNA had normal foxq2 transcript levels and distributions at the 32-cell stage (cf., Figure 2Aa,Af), suggesting that Fzl5/8 signaling is not required for the initial Wnt/β-catenin-dependent down-regulation of foxq2 mRNA in the posterior half of the embryo. Further evidence that Fzl5/8 is not required for early Wnt/β-catenin activity is provided below. However, at mesenchyme blastula stage (24 hpf), ΔFzl5/8-injected embryos expressed foxq2 ectopically throughout the anterior half of the embryo, indicating that the second phase of its restriction to the anterior pole requires Fzl5/8 function (cf., Figure 2Ab,Ag). Expression of foxq2 also was not correctly restricted in two different Fzl5/8 morphants, although the phenotype was less pronounced (cf., Figure 2Ab,Ag versus Figure S3D,E). We used ΔFzl5/8 for further studies because it gave the more penetrant phenotype, likely because it blocked signaling through both maternal and zygotic Fzl5/8. Importantly, eliminating expression of six3, the critical upstream ANE regulator, from the posterior ectoderm also required functional Fzl5/8 signaling (Figure 2Ad,Ai). Furthermore, the transcript levels per embryo for genes in the 24 hpf Six3-dependent ANE regulatory network (Figure 2B) were significantly elevated in ΔFzl5/8-containing mesenchyme blastula embryos. Interestingly, one of these was zygotic fzl5/8 mRNA itself (Figure 2Ac,Ah), indicating that Fzl5/8 function is required to down-regulate fzl5/8 mRNA levels in the posterior ectoderm. Finally, 3-d pluteus larvae injected with ΔFzl5/8 had an expanded thick neuroepithelium with a greatly increased number of serotonergic neurons (Figure 2Aj). In contrast, the thickened neuroepithelium in normal pluteus-stage embryos was restricted to a small region that produced only 4–6 serotonergic neurons (Figure 2Ae). These observations indicate that a Fzl5/8 signaling-dependent process eliminates the ANE regulatory state required for serotonergic neural development from the posterior ectoderm (Figure 2C).
In addition to the ubiquitous maternal and anterior zygotic expression of fzl5/8 at mesenchyme blastula stage (Figure S2e–h), it was also expressed in a ring of nonskeletogenic mesenchyme cells (24 hpf) (Figures 2Ac and S2Bh). Previously, Croce et al. (2006) [46] showed that Fzl5/8 signaling in these posterior cells works through the c-Jun N-terminal kinase (JNK) pathway to initiate primary invagination movements later during gastrulation. This observation raised the possibility that the earlier ANE restriction process mediated by Fzl5/8 in posterior ectoderm may also depend on the JNK pathway. jnk mRNA was present ubiquitously during ANE restriction (Figure S2C), and indeed, foxq2 failed to restrict to the anterior pole in embryos injected with a splice-blocking JNK morpholino (Figure S3A). This JNK morphant phenotype was weaker than the ΔFzl5/8 phenotype (cf., Figure 2G and Figure S3A), probably because some normal JNK transcripts persisted in the embryo (Figure S3J). It is also possible that maternally synthesized JNK protein persisted in these embryos. As an additional test, we treated embryos with the specific JNK inhibitor, (L)-JNKI1 [46],[47], beginning at fertilization, which produced embryos expressing foxq2 throughout the anterior half of the embryo, mimicking exactly the ΔFzl5/8 phenotype (Figure 2, cf. Ag,Al). Moreover, fzl5/8 and six3 expression was not restricted to the anterior pole (Figure 2Am,An), and these embryos also had an expanded, thickened neuroepithelium and an increased number of serotonergic neurons, as seen in ΔFzl5/8-injected embryos (Figure 2Aj). These results indicate that the second phase of ANE restriction that down-regulates the ANE regulatory state in the anterior half (i.e., the posterior ectoderm) depends on Fzl5/8 function. Moreover, they suggest that JNK activity transduces a Wnt signal X through this Wnt receptor, the production of which depends on Wnt/β-catenin activity in the posterior half of the embryo (signal X, Figure 2C).
To identify the link between Wnt/β-catenin signaling and Fzl5/8, we first searched for genes encoding Wnt ligands that are expressed by the 60-cell stage (i.e., the beginning of the second phase of ANE restriction) in posterior blastomeres and that also depend on Wnt/β-catenin activity. We confirmed the previously reported expression profile of wnt8, which is activated by Wnt/β-catenin [48],[49]: At the 60-cell stage, wnt8 was expressed in both the micromeres and the adjacent blastomere tier (veg2) (Figure 3Aa). Similarly, wnt1 expression was first detected midway through the 60-cell stage (9 hpf) in the micromeres, and it also depended on nβ-catenin (Figures 3Aa and S4A,C). As development progressed, wnt8 expression first moved into the next most anterior tier of blastomeres (veg1) and then, during late blastula stages (18 and 24 hpf), into both veg1 and overlying posterior ectoderm cells (Figure 3Ac,Ad). wnt1 expression continued in the micromeres until midblastula stages (15 hpf) after which it, too, progressively moved to more anterior blastomeres until it reached the endoderm/ectoderm boundary during later blastula stages (24 hpf) (Figures 3Ac,Ad and S4A). Thus, genes encoding the secreted ligands Wnt1 and Wnt8 were expressed in posterior cells when the second phase of ANE restriction begins in the posterior regions of the anterior hemisphere. As restriction proceeded, wnt8 continued to be expressed near cells expressing ANE marker genes, whereas wnt1 expression was more posterior. In order to evaluate whether these secreted ligands were required for ANE restriction in posterior ectoderm, we performed knockdown experiments by injecting either of two different morpholinos designed against each. As shown in Figure 3B, embryos injected with either Wnt1 or Wnt8 morpholinos failed to down-regulate foxq2 expression in posterior ectoderm. ANE restriction was more strongly perturbed in Wnt1 morphants, even though the cells producing it were more distant from the site of action than those producing Wnt8. This raised the possibility that Wnt1 is necessary for later Wnt8 expression. However, this was not the case because, at blastula stage (16 hpf), Wnt8 expression was normal in Wnt1 morphants (Figure S4D). The converse was also true: wnt1 expression did not depend on Wnt8 (Figure S4E). We conclude that production of each of these ligands depends on Wnt/β-catenin signaling, but they do not depend on each other but act in parallel in ANE restriction.
These results suggest that Wnt1 and Wnt8 spatiotemporally link posterior Wnt/β-catenin signaling to Fzl5/8-mediated down-regulation of ANE factors in the posterior ectoderm. To test this hypothesis, we first showed that overexpressed Wnt1 or Wnt8 completely eliminated foxq2 expression in the ANE (Figure 3Cb,Cd). We then tested whether this foxq2 down-regulation required active Fzl5/8. Strikingly, ΔFzl5/8 strongly blocked the suppression of foxq2 expression mediated by either Wnt1 or Wnt8 (100% rescue of Wnt1 or Wnt8 misexpression phenotype; n = 63 and 67, respectively) (Figures 3Cc,Ce and S5Bb,Bd). These results strongly support the conclusion from the Wnt1 and Wnt8 loss-of-function analyses that Fzl5/8-mediated ANE restriction in the posterior ectoderm requires these ligands. Furthermore, suppression of foxq2 expression by both Wnt 1 and Wnt8 also required JNK activity, since the JNK inhibitor rescued the loss-of-ANE phenotypes produced by misexpression of Wnt8 (81% of embryos rescued; n = 126)(Figure 3Ca versus Cf) and, to a lesser extent, Wnt1 (55% of embryos had low to normal foxq2 expression; n = 83) (Figure S5A). These data indicate that Wnt1, Wnt8, and Fzl5/8 function in a Wnt/JNK signaling pathway to effect the second phase of ANE restriction.
The expression pattern of the gene encoding the other early Wnt receptor, fzl1/2/7, suggests that Fzl1/2/7 signaling also could affect neuroectoderm restriction (Figure S2Ba–Bd). We tested this possibility by morpholino knockdown. We were surprised to find that neither six3 nor foxq2 was activated at the 32- to 60-cell stage (Figure 4Aa versus Ah and Figure 4C) and neither mRNA was detectable throughout the normal time of ANE restriction (Figure 4Ah–k, Af- versus Am). As expected, zygotic fzl5/8 expression, which depends on Six3 [24], also required Fzl1/2/7 (Figure 4Ae versus Al). As well, the expression of all other known regulatory factors that depend on Six3 at mesenchyme blastula stage (24 hpf) also required Fz1/2/7 function (Figure 4D). Moreover, the ectoderm in 3- to 4-d Fzl1/2/7 morphants lacked a thickened columnar epithelium corresponding to the ANE in normal embryos (Figure S3F). In 4-d pluteus larvae, which normally have well-established neurons in the ANE, the large majority of Fzl1/2/7 morphants had none (37/41 embryos) (Figure 4Ag versus An, green). They also had a severely reduced number of ciliary band neurons, as assayed by the pan-neural marker SynaptotagminB (Figure 4An, 1e11 antibody, magenta). These results indicate that Fzl1/2/7-mediated signaling is essential for establishment and maintenance of the early neuroectoderm regulatory state, which in turn subsequently is required for the specification and differentiation of all neurons (Figure 4Ag).
The Fzl1/2/7 morphant phenotype is opposite to the Axin or Tcf-Eng misexpression phenotypes as well as those produced by ΔFzl5/8 misexpression or treatment with the JNK inhibitor or JNK morpholino. These observations raise the possibility that Fzl1/2/7 transduces a different Wnt signal, possibly through the Ca2+ pathway. Although the architecture of the Ca2+ pathway downstream of Fzl receptors is not yet well established, one important player in other systems is conventional Protein Kinase C (PKC) [50],[51]. In the sea urchin embryo, genes encoding conventional PKC isoforms are expressed maternally and throughout development and at least one is activated by the 60-cell stage (Figure 4B) [52]. To test the hypothesis that pPKC, like Fzl1/2/7, is necessary for maintaining ANE specification, we treated embryos with the specific PKC inhibitor, Bisindolylmaleimide 1, which blocks activation through phosphorylation of most Ca2+-dependent PKC isoforms by competing for the ATP binding site [53]. Treatment with this inhibitor at 1–3 µM strongly reduced the level of pPKC (Figure 4B), but had no detectable deleterious effects on the morphology of embryos during ANE restriction. Importantly, the level of pPKC in Fzl1/2/7 morphants was as low as that produced by the PKC-specific inhibitor (Figure 4B), indicating that Fzl1/2/7 function is required for activation of this kinase. Similar to Fzl1/2/7 morphants, foxq2 expression was never initiated in embryos treated with the inhibitor continuously from fertilization to mesenchyme blastula stage (24 hpf) (Figure 4Ao–r). Moreover, six3 and fzl5/8 were not expressed (Figure 4As,At), and in a large majority of embryos (36/39) serotonergic neurons did not develop (Figure 4Au and Figure S3Ia versus Ic), showing that neural differentiation was severely compromised in treated embryos. While these experiments demonstrate that activation of PKC is required for the ANE regulatory state and that Fzl1/2/7 is required for that activation, they do not conclusively prove that Fzl1/2/7 signals through the Ca2+ pathway because PKC activation can occur by other mechanisms. We conclude that Fzl1/2/7 signaling and PKC activity are each essential for early neuroectoderm specification.
Our findings that a Wnt signaling branch utilizing Fzl1/2/7 and PKC activity is necessary for initiating expression of upstream ANE regulatory factors was entirely unexpected because at early stages, Wnt signaling is thought to antagonize this process. We hypothesized that Wnt signaling through this receptor is necessary either for the expression of regulatory genes that specify the ANE or for antagonizing the ANE restriction mechanism from the very earliest stages. To distinguish between these alternatives, we first asked whether Fzl1/2/7 signaling is part of the maternal mechanism that can drive ubiquitous expression of ANE regulatory genes in the absence of Wnt/β-catenin signaling. Within each of three batches of embryos, we injected one set of fertilized eggs with Axin mRNA, a second set with Fzl1/2/7 morpholino, and a third with both Fzl1/2/7 morpholino and Axin mRNA (Figure 5A). As shown above, foxq2 was expressed throughout the embryo in the absence of nβ-catenin, whereas it was completely undetectable in more than 90% (52/57) of embryos lacking Fzl1/2/7. However, it was expressed at high levels throughout all Fzl1/2/7-deficient embryos (47/47) when Wnt/β-catenin signaling was also blocked. These results indicate that maternal factors are still capable of activating foxq2 in embryos lacking Fzl1/2/7 and that the loss of foxq2/ANE fate in Fzl1/2/7 morphants requires a functional Wnt/β-catenin pathway. Thus, Fzl1/2/7 signaling is not a positive regulator of the initial maternal regulatory state that supports ANE specification, but rather it inhibits the Wnt/β-catenin-dependent ANE restriction mechanism.
To test if Fzl1/2/7 also antagonizes the Fzl5/8-JNK-dependent second phase of ANE restriction, we asked whether blocking Fzl5/8 or JNK function could rescue ANE specification in embryos lacking Fzl1/2/7 signaling (Figure 5B,C). Similar to the above experiments, in three different batches of embryos, we found that blocking the function of either Fzl5/8 or the JNK pathway rescued the expanded expression of foxq2 in 99% (n = 72) or 93% (n = 70), respectively, of embryos also lacking Fzl1/2/7. These results suggest that Fzl1/2/7 antagonizes Fzl5/8-JNK-mediated ANE restriction. In the final set of experiments, we tested whether PKC signaling also antagonizes Fzl5/8-JNK-mediated ANE restriction (Figure 5D). Using the same approach, we injected one set of fertilized eggs with ΔFzl5/8, treated a second with the PKC inhibitor, and a third was treated with PKC inhibitor and injected with ΔFzl5/8. Blocking the function of Fzl5/8 in these embryos rescued the expression of foxq2 in a large majority of embryos (77% rescue; n = 83), demonstrating that, like Fzl1/2/7, PKC antagonizes the ANE restriction mechanism by antagonizing Fzl5/8 signaling. Collectively, these results support the idea that the Fzl1/2/7-dependent suppression of Fzl5/8-mediated ANE restriction works through PKC (Figure 5G).
The data suggest that Fzl1/2/7 signaling antagonizes Fzl5/8-JNK-mediated down regulation of genes necessary for ANE specification. Because Fzl1/2/7 functions as early as the 32-cell stage to maintain expression of ANE markers, it might also antagonize Fz5/8 indirectly by down-regulating Wnt/β-catenin activity. To test this possibility, we measured the level of Wnt/β-catenin signaling in 120-cell embryos (12 hpf) during the early stages of ANE restriction using the TCF-luciferase reporter plasmid, TopFlash [54]. Three different batches of embryos that had been injected with ΔFzl5/8 or treated with PKC inhibitor showed no significant difference in TopFlash activity when compared to controls (Figure 5E), suggesting that neither of these proteins affects early Wnt/β-catenin signaling. In contrast, TopFlash activity increased ∼2.5-fold on average in embryos lacking Fzl1/2/7 compared to controls (Figure 5E), indicating that signaling through Fzl1/2/7 negatively regulates Wnt/β-catenin activity in cleavage-stage embryos. Recently published experiments showed that introduction of mRNA encoding a dominant negative form of Fzl1/2/7 caused a reduction in TopFlash activity in cleavage-stage embryos and a loss of endoderm specification [55]. While this appears to conflict with our results, it is important to realize that interference with Fzl1/2/7 activity by misexpression of ΔFzl1/2/7 can interfere with the function of maternal Fzl1/2/7, whereas Fzl1/2/7 morpholino cannot. In keeping with this, embryos in which zygotic Fzl1/2/7 synthesis was blocked with a morpholino still expressed Wnt1 and Wnt8 (Figure 5F), whereas these are not expressed in embryos injected with ΔFzl1/2/7 [55]. Thus, in Fzl1/2/7 morphants, these Wnt ligands up-regulate Wnt/β-catenin- and Wnt/Fzl5/8-mediated ANE restriction, whereas the absence of these ligands in ΔFzl5/8-containing embryos leads to a reduction in Wnt/β-catenin activity. Collectively, these data suggest that Fzl1/2/7 signaling and PKC activity provide a buffer that limits the rate of ANE down-regulation by both of these Wnt signaling pathways (Figure 5G).
A possible concern in the ΔFz5/8 and ΔFz1/2/7 experiments is that elevating the levels of these proteins might influence the balance of signaling between the Wnt signaling pathways, for example, by competing for common components. To test this possibility, we overexpressed either wild-type Fzl5/8 or Fzl1/2/7 mRNA. In both cases, embryos developed normally and had normal foxq2 expression patterns (Figure S6A). Next we showed that elevating the levels of Fzl1/2/7 mRNA did not change ΔFzl5/8's ability to prevent ANE restriction (Figure S6B) or prevent elimination of the ANE by excess Wnt1 mRNA (Figure S6C). Taken together these data indicate that the levels of endogenous Fzl receptors are not limiting. These data contrast with the Wnt1 and Wnt8 misexpression results, which showed that excess ligand can dramatically up-regulate ANE restriction (Figure 3Cb,d), suggesting that it is the levels of Wnt ligand in time and space and not those of the Wnt receptors that control the ANE restriction mechanism
Around the mesenchyme blastula stage (24 hpf), restriction of the ANE is complete and it constitutes a separate regulatory domain at the anterior end of the embryo with well-defined borders. Expression of fzl5/8 is also restricted to this domain (Figure 2), raising the question of why Fzl5/8-mediated signaling does not continue to down-regulate the ANE regulatory state there. We hypothesized that the secreted Wnt antagonist, Dkk1, might play a role because, in most of the major clades, competition between anterior Wnt antagonism by Dkk1 and posterior Wnt signaling has been shown to regulate cell fates along the primary (AP) axis [8]. Very low-level dkk1 expression was detectable as early as the 120-cell stage by qPCR (Figure 6A), and increased during the time of ANE restriction, reaching maximal levels by the mesenchyme blastula stage (24 hpf). At this time dkk1 expression could be detected by in situ hybridization at the anterior end of the embryo as well as in a ring of cells surrounding the future site of gastrulation (Figure 6A, inset). Thus, dkk1 was expressed at the right time and place to prevent anterior Wnt-mediated ANE down-regulation. Interestingly, expression of dkk1 depended on Fzl5/8 signaling (Figure 6B), raising the possibility that Fzl5/8 signaling limits its own activity in anterior cells by promoting a negative feedback mechanism through this Wnt antagonist.
To test whether Dkk1 protects the ANE regulatory state from Wnt-mediated down-regulation, we monitored the expression of a set of genes encoding ANE regulatory factors in Dkk1 morphants at mesenchyme blastula stage by in situ hybridization. Each of the genes tested was severely down-regulated in these embryos (Figures 6Cg–k and S3H), and no serotonergic neurons developed in 4-d plutei (Figure 6Cl). Furthermore, overexpression of Dkk1 mRNA prevented restriction of foxq2 expression (Figure 6Db) and rescued foxq2 expression in embryos also overexpressing wnt1 mRNA (88% rescue; n = 65) (Figure 6Dc versus d). Together these results indicate that Dkk1 can block the Wnt1/Fzl5/8-JNK-dependent ANE restriction mechanism. Interestingly, overexpression of Dkk1 also rescued foxq2 expression in Fzl1/2/7 morphants (98% rescue; n = 64) (Figure 6Ec,Ed), suggesting that it may interfere with either Wntβ-catenin or Fzl5/8 signaling or both. There is some support for both possibilities. First, Dkk1 likely inhibits Fzl5/8 activity because the morphological phenotype (unpublished data) and foxq2 expression pattern (cf., Figures 1F, 2Ag, and 6Db) of Dkk1 mRNA-injected embryos were more similar to those of embryos lacking functional Fzl5/8 than to those lacking Wnt/β-catenin signaling (cf., Figure 6Db and Figures 1H and 2Ag). Second, misexpressed Dkk1 can also interfere with endomesodermal gene expression, which depends on the Wnt/β-catenin pathway (Figure S4B).
The data presented here show that patterning the neuroectoderm along the AP axis of the early sea urchin embryo depends on an elegant spatiotemporal coordination and integration of the activities of three different Wnt signaling pathways. Throughout this process, a balance is achieved between the initial regulatory mechanisms that can specify the ANE ubiquitously, those that subsequently suppress it in posterior regions, and those that limit ANE suppression. The consequence is that ANE tissue is stably positioned only at the anterior pole of the embryo by the mesenchyme blastula stage. To summarize our current model (Figure 7), the first phase of ANE restriction requires Wnt/β-catenin and occurs very rapidly in posterior blastomeres by the 32- to 60-cell stage. Wnt/β-catenin signaling simultaneously activates expression of Wnt1 and Wnt8; these cells and these ligands initiate the second phase of ANE down-regulation in the posterior ectoderm (non-ANE ectoderm in the anterior hemisphere) by activating the Fzl5/8-JNK pathway, beginning around the 60-cell stage. As development progresses, Wnt1 and Wnt8 mRNAs accumulate in more anterior blastomeres, behind the front of ANE down-regulation. Whether these secreted ligands diffuse to the overlying ectoderm to directly activate Fzl5/8 or whether they act indirectly to stimulate production of other Wnt ligands that signal through this receptor is not known. Regardless, it is clear that Wnt1, Wnt8, Fzl5/8, and JNK are all required for full ANE down-regulation in the posterior ectoderm and suppression of transcription of fzl5/8 itself. Clearly, Fzl5/8 plays a pivotal role in the ANE restriction process because it is necessary not only for the second phase of ANE restriction but also to stop that process in the third phase of ANE patterning when Fzl5/8 signaling leads to the expression of the Wnt receptor antagonist Dkk1 at the anterior pole. Thus, the coordination between the timing of auto-repression of fzl5/8 transcription and activation of Dkk1 by Fzl5/8 ensures that this negative feedback loop reproducibly defines the ANE at the anterior pole of the embryo by mesenchyme blastula stage (24 hpf). The relative timing of Dkk1 production in the anterior ectoderm and ANE restriction in the rest of the embryo is critical and carefully controlled by a third Wnt pathway working through Fzl1/2/7 and PKC activities that limit Wnt/β-catenin and Wnt/JNK functions during the first two phases of ANE clearance.
Because all of these Wnt pathways affect the same developmental process (i.e., the specification of ANE versus non-ANE fates along the primary axis), they may function as components of an interactive Wnt signaling network rather than as separate pathways with different roles. Yet it appears that posterior Wnt/β-catenin and anterior Wnt/JNK signaling define two adjacent early regulatory domains in the sea urchin embryo. While our data suggest that these two signaling pathways activate different downstream regulatory programs in order to down-regulate ANE factors, both pathways are linked spatially and temporally by the activities of at least two common signaling components, Wnt1 and Wnt8 [49],[56]. These results are in keeping with recent evidence that individual Wnt ligands are able to activate distinct Wnt signaling branches, often in the same or adjacent territories [57]–[59]. However, it remains to be determined whether Wnt1 and/or Wnt8 act directly on cells in the anterior hemisphere in the ANE restriction process, although it is interesting that wnt8 expression moves into posterior ectoderm cells as ANE factors move out. Alternatively, Wnt1 and Wnt8 may act indirectly by reinforcing the nβ-catenin gradient in the anterior-most cells of the posterior half of the embryo (i.e., near the equator), activating production of an unidentified intermediate Wnt ligand that is secreted from even more anterior cells and that activates Fzl5/8 signaling.
We found that the cardinal ANE regulatory genes, six3 and foxq2, are not expressed in Fzl1/2/7 knockdowns. This unexpected phenotype is the exact opposite of the ANE expansion produced by interference with Wnt/β-catenin, Fzl5/8, and JNK signaling. The function of Fzl1/2/7 begins as early as the 32-cell stage, around the time that nβ-catenin is first detectable in posterior blastomeres [4] and at least 2 h before the ANE restriction process mediated by Fzl5/8 and JNK is observed. Since Fzl1/2/7 signaling significantly suppresses Wnt/β-catenin signaling during early cleavage stages, we propose that it reduces Wnt/β-catenin-dependent Fzl5/8 and JNK activities. This model suggests that Fzl1/2/7 signaling is essential for controlling the rate of progression of the ANE restriction mechanism along the AP axis, providing a “timing buffer” that prevents premature elimination of the ANE regulatory state during the early cleavage and blastula stages. We propose that one function of this Fzl1/2/7 “timing buffer” is to allow sufficient Fzl5/8-dependent accumulation of Dkk1 in the ANE by later blastula stages to protect it from Wnt signals and define its borders.
Since Fzl1/2/7 does not appear to signal through either the Wnt/JNK or the Wnt/β-catenin pathways during ANE restriction, we propose that it transduces signals through the Wnt/Ca2+ pathway. This mechanism may be similar to the situation in several other systems where Wnt/Ca2+ signaling affects early development either through the intracellular messengers CamKI, Calcineurin, and the transcription factor, NF-AT, or through PKC [60]. Similar to what we report here, the Wnt/Ca2+ pathway has been shown to antagonize Wnt/β-catenin signaling during vertebrate D/V axis specification [61],[62]. Interestingly, we found that blocking PKC activity with either an inhibitor or with Fzl1/2/7 morpholino had exactly the same effect on phosphorylated PKC levels and on the Fzl5/8-JNK-dependent re-specification of ANE to ectoderm fate. However, inhibiting the function of Fzl1/2/7 elevated Wnt/β-catenin activity, whereas loss of PKC activity did not. This result suggests that Fzl1/2/7 signaling activates two branches that affect ANE restriction, one that antagonizes early Wnt/β-catenin activity and another, mediated by pPKC, that blocks Fzl5/8-mediated ANE restriction in the anterior hemisphere. Thus, if Fzl1/2/7 mediates Wnt/Ca2+ signaling in the sea urchin embryo, it could affect several different downstream parallel pathways, any or all of which are necessary to prevent premature and complete elimination of the ANE regulatory state. Moreover, the involvement of Wnt/Ca2+ signaling in AP neuroectoderm patterning would be a first.
These considerations suggest that the function of Fzl1/2/7 in the early embryo is context-dependent, and we propose that the balance of information sent by this receptor through different Wnt signaling pathways is essential for correct specification and patterning. Recent data from several laboratories suggest that the same Fzl receptors can activate different Wnt signaling pathways, even in the same cells [50],[60],[63]. For example, the sea urchin Fzl1/2/7 homologue, Fz7, activates Wnt/β-catenin signaling and D/V axis specification in the early Xenopus embryo [63], but it also later activates Wnt/JNK and possibly the PKC signaling pathways that are required in the same general territory for convergent extension movements during gastrulation [57],[64],[65]. In the sea urchin embryo, our results and those of Lhomond et al. (2012) [55] are consistent with two early roles for Fzl1/2/7 – one stimulating endoderm specification via nβ-catenin through maternal Fzl1/2/7 in posterior blastomeres and another produced by zygotic Fzl1/2/7 that antagonizes early Wnt/β-catenin and subsequent Wnt/JNK signaling through an alternative Wnt pathway (Ca+2) that operates throughout the embryo. The balance between these pathways may favor Wnt/β-catenin signaling in the posterior half of the cleavage stage embryo because of localized Wnt/β-catenin pathway-specific co-factors in that part of the embryo [66],[67].
Striking parallels are emerging in the regulatory mechanisms that sea urchin and vertebrate embryos use to establish neural regulatory states at the anterior pole. Both embryos require Six3 for anterior neural development and share many homologous factors [24],[27]. Moreover, as shown here, in the absence of Wnt/β-catenin, and consequently of Nodal, BMP, and all other known signaling pathways, the regulatory state of all of the cells in the sea urchin embryo supports development of ANE from the very beginning of its specification. These data indicate that an initial ubiquitous maternal regulatory state activates ANE specification and that one of the most important roles of posterior Wnt/β-catenin signaling is to break the symmetry of this neural-promoting state. Similarly, in vertebrate embryos, an initial regulatory state is capable of activating ANE markers throughout the embryo in the absence of Wnt, Nodal, and BMP signaling [15],[20]–[22],[68]. Thus, this initial, broad activation of ANE specification, and its subsequent down-regulation, could be a widely shared property of deuterostome embryos.
The Wnt-dependent mechanism used for AP neuroectoderm patterning is still incompletely understood in vertebrates, in part because complex cell movements during patterning and the involvement of Wnt signaling in earlier specification events obscure the spatial and temporal relationships among the individual players [28],[30],[36]. In vertebrates, the only known Wnt pathway involved in the early restriction of ANE factors to the anterior pole is Wnt/β-catenin signaling [16]–[18]. Here, we show for the first time, to our knowledge, that the anterior Dkk1-posterior Wnt/β-catenin neuroectoderm patterning mechanism observed in vertebrates exists in a nonchordate deuterostome. These data strongly suggest the general Dkk1-Wnt/β-catenin AP patterning mechanism present in extant pre-bilaterian embryos was likely co-opted to pattern the neuroectoderm along the AP axis in the deuterostome ancestor. In addition to a posterior-to-anterior gradient of Wnt/β-catenin signaling, AP neuroectoderm patterning in the sea urchin embryo also requires Wnt/JNK signaling and an additional pathway mediated by Fzl1/2/7 that may function in Wnt/Ca2+ signaling. At present these are completely novel findings, but the fact that orthologs of several Wnt signaling components that function in these additional pathways in sea urchins (Fzl8, Wnt1, Wnt8, Dkk1) (Figure 8A,C) also are involved in posteriorizing the neural plate of vertebrate embryos (Figure 8A) [17],[31],[69] raises the possibility that this entire multistep mechanism was present in the common echinoderm/vertebrate ancestor and still operates to specify anterior neural identity in deuterostome embryos. Supporting this view, recent studies in hemichordates indicate that expression of homologues of sea urchin foxq2, sfrp1/5, and six3 demarcate an anterior-most region of the embryo that is homologous to the vertebrate anterior neural ridge secondary patterning center [19]. Interestingly, these factors are initially broadly expressed and restricted to this region by an unknown mechanism that depends on posterior Wnt/β-catenin signaling and appears to require Fzl5/8 function in the anterior part of the embryo (Figure 8D). Moreover, there are similarities in the expression patterns of ANE genes (dkk1, dkk3, six3, foxq2) and those specifying endomesoderm (wnt1 and wnt8) between the invertebrate chordate amphioxus and the sea urchin embryo (Figure 8B). For example, foxq2 is initially expressed in a broad region and subsequently restricted to the anterior-most region. It can be completely cleared from this region of the embryo by LiCl treatment, which can elevate Wnt/β-catenin signaling [11], raising the possibility that amphioxus also uses the same ANE patterning mechanism. Thus, there is accumulating evidence that the ANE clearance mechanism described here may be used in a wide variety of deuterostomes. However, to date, only the work reported here reveals the intricate, interdependent Wnt signaling mechanisms that are required to confine the ANE regulatory state to the anterior end of the embryo.
Strongylocentrotus purpuratus sea urchins were obtained from Point Loma Marine Invertebrate Lab, Lakeside, CA; The Cultured Abalone, Goleta, CA; or Marinus, Garden Grove, CA. Embryos were cultured in artificial seawater at 15°C. For drug treatments, eggs attached to a protamine sulfate-coated plate were fertilized in the presence of 2 mM 4-Aminobenzoic acid (PABA), and fertilization envelopes were removed by shear force. Treatments with the cell-permeable JNK Inhibitor 1, (L)-form, (EMD/Calbiochem) and the PKC inhibitor, Bisindolylmaleimide 1 (EMD/Calbiochem), were performed by diluting the stock solution to 50 µM or 3 µM, respectively. JNK Inhibitor 1, (L)-form is a specific inhibitor that blocks interactions between JNK and its transcriptional substrates, such as c-Jun and c-Fos, resulting in a knockout phenotype [46],[47]. Bisindolylmaleimide 1 is a selective inhibitor that specifically competes with the ATP binding site of most PKC isoforms [53]. As controls for the PKC inhibitor experiments, DMSO was added alone. These experiments were repeated with at least three different embryo batches, and each produced the same results.
The 24-h blastula total cDNA was used to obtain full-length clones for dkk1, frizzled5/8, frizzled1/2/7, wnt1, and a partial clone of jnk by PCR. The following primers were based on the sea urchin genome sequence: Sp-dkk1 Forward 5′-AGAATGGCGGCTCCTTCTGC-3′, Reverse 5′-TCATAATACAGTTAACTGGC-3′; Sp-frizzled5/8 Forward 5′-AGAATGGCTGCCTTCAGTGGAAC-3′, Reverse 5′-TCACACCTGTACATTTGGTA-3′; Sp-frizzled1/2/7 Forward 5′-AGAATGGGTTGGTTGGTGAGA-3′, Reverse 5′-TCATACATTGGCTGGTGCAC-3′; Sp-wnt1 Forward 5′-AGAATGAAACTTGAGTGGTTTG-3′, Reverse 5′-CTACAAGCATCTGTGCACG-3′; Sp-jnk Forward 5′-GAATGTGACGCATGCCAAGC-3′, Reverse 5′-GATCACCGCCGTCGTCTATTG-3′.
Full-length dkk1 and wnt1 cDNA sequences were inserted into pCS2+ vector for mis-expression studies. ΔFzl5/8-pCS2 and Wnt8-pCS2 were obtained from Jenifer Croce (CNRS/Villefranche sur Mer, France) and Christine Byrum (College of Charleston, Charleston, SC), respectively. pCS2 constructs were linearized with Not1 and mRNA was synthesized with mMessage Machine kit (Ambion), purified by LiCl precipitation and ∼20 pl injected at the following concentrations: Fzl1/2/7 mRNA = 1.0–1.5 µg/µL; Fzl5/8 mRNA = 2.0 µg/µL; ΔFzl5/8 mRNA = 2.0 µg/µL; Wnt1 mRNA = 0.01–0.1 µg/µL; Wnt8 mRNA = 0.5–1.0 µg/µL; Dkk1 mRNA = 3.0 µg/µL; Axin mRNA = 1.0 µg/µL; Tcf-Eng mRNA = 0.5–1.0 µg/µL.
S. purpuratus EST sequences for wnt1, fzl1/2/7, and fzl5/8 as well as sequence information from 5′ RACE on dkk1 were used to generate translation-blocking morpholino oligonucleotides. A splice-blocking morpholino oligonucleotide was designed for the second exon–intron boundary of wnt8, which produces transcripts encoding a protein lacking sequence from the second exon, which was verified by PCR (Figure S2A). The morpholinos were produced by Gene-Tools LLC (Eugene, OR). The sequences and injection concentrations were: Wnt1 MO1: 5′-ACGCTACAAACCACTCAAGTTTCAT-3′ (1.8 mM); Wnt1 MO2: 5′-ATCCTCATCAAAACTAACTCCAAGA-3′ (0.4 mM); Wnt8 splice MO: 5′-GTAAAGTGTTTTTCTTACCTTGGAT-3′ (0.7 mM); Wnt8 MO2: 5′-GTACACTCCAATAAAAGAAATCAAA-3′ (0.6 mM) [49]; Fzl1/2/7 MO1: 5′-CATCTTCTAACCGTATATCTTCTGC-3′ (1.3 mM); Fzl1/2/7 MO2: 5′-ACAGATCTCCTTTAACAAGGGTAGA-3′ (2.2 mM); Fzl5/8 MO1: 5′-GGATTGTTCCACTGAAGGCAGCCAT-3′ (2.25 mM); Fzl5/8 MO2: 5′-ATGTTTATGGTCTGATGGCAATCGC-3′ (0.6 mM); Dkk1 MO1: 5′-GCGTCTAAATCCTAAATTCCTTCCT-3′ (1.5–1.6 mM); Dkk1 MO2: 5′-ATCGTTGGTAGTTGCAGAAATTCGT-3′ (0.7–0.85 mM); and JNK splice MO: 5′-CCTCATCGTTCTAGACTCACCGTTC-3′ (1.0–1.25 mM).
Embryos were injected immediately after fertilization with solutions containing FITC, 20% glycerol, and mRNA and/or morpholino oligonucleotides. All injected embryos were cultured at 15°C. Microinjection experiments were performed using at least three different batches of embryos, and each experiment consisted of 50–150 embryos unless otherwise stated. Experiments were scored only if a change in phenotype or marker expression was seen in at least 85%–90% of the manipulated embryos.
qPCR was performed as described previously [24]. Each experiment was repeated with embryos from at least three different mating pairs, and each PCR reaction was carried out in triplicate. The primer set information can be found in Table S1. For developmental expression analysis, the number of transcripts per embryo was estimated based on the Ct value of the z12 transcript [70].
The probes for each gene analyzed correspond to the full-length cDNA sequence. Alkaline phosphatase and three-color fluorescent in situ hybridization were carried out as previously described [24],[71]. For the three-color in situ hybridization, foxq2 was labeled with fluorescein and detected with Cy5-TSA, wnt1 was labeled with DNP and detected with Cy3-TSA, and wnt8 was labeled with DIG and detected with fluorescein-TSA.
Embryos were fixed in 2%–4% paraformaldehyde in artificial seawater at RT for 20 min and washed 5 times in phosphate-buffered saline containing 0.1% Tween-20. Embryos were incubated with primary antibodies at 4°C overnight at a dilution of 1∶1,000 for serotonin (Sigma, St. Louis, MO) and synaptotagminB/1e11 [72]. Primary antibodies were detected by incubating embryos with Alexa-coupled secondary antibodies for 1 h at RT. Nuclei were stained with DAPI.
Protein extracts were prepared by adding 30 µL of lysis buffer (25 mM Tris-HCL pH 7.4; 150 mM NaCl; 5 mM EDTA; with PhosSTOP phosphatase and Complete Mini protease inhibitor cocktails; Roche, Indianapolis, IN) to a pellet of 300 injected embryos. Embryos were crushed 4–5 times with a pestle, immediately spun at 16,000 RCF for 15 min at 4°C, and the supernatant was stored at −80°C until use. Samples were thawed on ice and 4× NuPage Running Buffer containing 4% SDS and 10% 2-ME was added. Samples were heated at 80°C for 3–5 min and 20 µL of each sample was run on 4%–12% NuPage Bis-Tris gradient gel (Invitrogen, Grand Island, NY), transferred to nitrocellulose. Membranes were probed overnight at 4°C in Phosphate Buffered Saline+0.1% Tween-20 (PBST)+3% BSA with a poly-clonal Phospho-PKC(pan) (β11 Ser660) antibody (Cell Signaling Technology, Danvers, MA) (1∶250) that recognizes a region that includes serine 660 and detects endogenous levels of phosphorylated PKC α, β1, β11, δ, ε, η, and θ. The recognition sequence is conserved in S. purpuratus PKC isoforms. Membranes were washed 3–5 times in room temperature PBST and probed for 1 h at room temperature in PBST+3% Bovine Serum Albumin with an enhanced chemiluminescent anti-rabbit IgG horseradish peroxidase secondary antibody (GE Healthcare, Piscataway, NJ). After 3–5 more washes in PBST, the membranes were developed and imaged.
Promega Dual Luciferase Reporter System (Promega) was used to perform dual luciferase assays. Embryos (350–400) were injected with linearized TopFlash-Firefly Luciferase (REF) and Endo16-Renilla Luciferase plasmids at concentrations of 20 ng/µL and 10 ng/µL, respectively, along with 10 ng/µL of linearized genomic DNA carrier. The Firefly and Renilla luciferase signals were recorded with a plate style luminometer using Promega's suggested protocol. The level of luciferase activity was normalized to the level of Renilla activity for each experiment. All experiments were repeated three times using separate batches of embryos.
|
10.1371/journal.ppat.1005020 | Type VI Secretion System Transports Zn2+ to Combat Multiple Stresses and Host Immunity | Type VI secretion systems (T6SSs) are widespread multi-component machineries that translocate effectors into either eukaryotic or prokaryotic cells, for virulence or for interbacterial competition. Herein, we report that the T6SS-4 from Yersinia pseudotuberculosis displays an unexpected function in the transportation of Zn2+ to combat diverse stresses and host immunity. Environmental insults such as oxidative stress induce the expression of T6SS-4 via OxyR, the transcriptional factor that also regulates many oxidative response genes. Zinc transportation is achieved by T6SS-4-mediated translocation of a novel Zn2+-binding protein substrate YezP (YPK_3549), which has the capacity to rescue the sensitivity to oxidative stress exhibited by T6SS-4 mutants when added to extracellular milieu. Disruption of the classic zinc transporter ZnuABC together with T6SS-4 or yezP results in mutants that almost completely lost virulence against mice, further highlighting the importance of T6SS-4 in resistance to host immunity. These results assigned an unconventional role to T6SSs, which will lay the foundation for studying novel mechanisms of metal ion uptake by bacteria and the role of this process in their resistance to host immunity and survival in harmful environments.
| One unique feature of type VI secretion system is the presence of multiple distinct systems in certain bacterial species. It is well established that some of these systems function to compete for their living niches among diverse bacterial species, whilst the activity of many such transporters remains unknown. Because metal ions are essential components to virtually all forms of life including bacteria, eukaryotic hosts have evolved complicated strategies to sequester metal ions, which constitute a major branch of their nutritional immunity. Therefore the ability to acquire metal ions is critical for bacterial virulence. This study reveals that the T6SS-4 of Yersinia pseudotuberculosis (Yptb) functions to import Zn2+ from the environment to mitigate the detrimental effects such as hydroxyl radicals induced by diverse stresses. Expression of the transporter is activated by multiple regulatory proteins, including OxyR and OmpR that sense diverse environmental cues. Zinc ion acquisition is achieved by translocating a Zn2+-binding substrate YezP, which is co-regulated with T6SS-4 by OxyR. Our results reveal a novel role for type VI secretion system, which is important in the study of the mechanism of metal ion acquisition by bacteria and the role of this process in bacterial pathogenesis and survival in detrimental environments.
| Specialized protein secretion systems are essential for many bacteria to survive in interactions with their hosts or within specific environmental niches. Among these, type VI secretion system (T6SS) is a complex macromolecular apparatus found in more than 25% of sequenced Gram-negative bacteria genomes, ranging from pathogens to environmental species [1]. Structurally related to the contractile phage tail sheath, the T6SS is composed of 13 conserved proteins and a variable array of accessory elements [1,2]. The extracellular components of the T6SS, Hcp (hemolysin-coregulated protein) and VgrG (valine glycine repeat) form a needle-like injection device closely resembling the bacteriophage tail, in which VgrG forms a cell-puncturing tip, and Hcp forms a tail-tube structure through which effector proteins are believed to travel [1,2]. ClpV and IcmF, two conserved components with ATPase activity that powers the T6SS, are crucial for the secretion of Hcp, VgrG and its cognate protein substrates [1,2].
A striking feature of T6SS is that many genomes harbor multiple gene clusters coding for evolutionarily distinct T6SSs, which presumably play different roles in the lifecycle of the bacteria [1,3]. Several T6SSs associated with pathogens are necessary for full virulence towards eukaryotic host cells [4,5]. In Vibrio cholera, T6SS is required for virulence in animal infection or for resistance to predation by amoebae hosts such as Dictyostelium discoideum [4,5]. In Burkholderia mallei, the cluster 1 T6SS expressed by this organism is essential for bacterial survival in a hamster model of glanders and the tssE mutants exhibit growth and actin polymerization defects in RAW 264.7 murine macrophages [6]. On the contrary, some T6SSs appear to be antivirulence factors because mutants lacking such systems are more pathogenic [7,8]. Deletion of the T6SS in Helicobacter hepaticus led to mutants that adhere and enter epithelial cell at high efficiencies than wild-type bacteria [5]. In these scenarios, effectors, the T6SS apparatus or its components may stimulate the host immune response to suppress the virulence of wild-type bacteria.
The best-characterized function of T6SSs is to compete in bacterial communities by delivering bacteriolytic toxins to target cells [2,9]. For example, a T6SS in Pseudomonas aeruginosa delivers at least two families of effectors into target bacterial cells, which function as peptidoglycan hydrolases and phospholipase, respectively [9,10]. These effectors mediate antagonistic bacterial interactions in either inter- or intraspecies context to gain a survival advantage in specific niches. Similarly, Agrobacterium tumefaciens uses T6SS to translocate antibacterial DNases to attack neighboring bacterial cells in plant hosts [11]. Interestingly, in each case, the toxicity of the effectors toward the bacterial cell itself is inhibited by specific immunity proteins, which directly interact with the effectors [9,11]. Roles of T6SSs in biological processes beyond infection and inter-species competition have also been suggested [12–14], but little is known about the underlying mechanisms.
Whereas the genomes of many bacteria harbor one to two T6SS gene clusters [1], the closely related Yersinia pseudotuberculosis (Yptb) and Yersinia pestis contain four and five such clusters, respectively [1]. These systems likely confer distinct functions for specific niches in the lifecycle of the bacterium, thus representing excellent models for the study of the potentially versatile function of T6SSs. Here we found that the T6SS-4 of Yptb functions to acquire zinc ions (Zn2+) into bacterial cells from the environment, which mitigates the hydroxyl radicals induced by oxidative stresses. Our results reveal that diverse environmental insults activate the expression of T6SS-4 via OxyR, the primary regulatory protein for bacterial oxidative stress and that zinc acquisition is achieved by T6SS-4-mediated translocation of a zinc-binding protein into the extracellular milieu. While it is well established that when appropriately deployed, some T6SSs confer the bacterium surviving advantages in niches with multiple bacterial species by delivering bacteriolytic toxins to competing cells, our results uncover a novel function of T6SS in the acquisition of essential nutrients, which enhances bacterial survival under harsh environments and/or during its interactions with hosts.
To determine the function of the T6SS-4 in Yptb, we analyzed its promoter region and identified a DNA element highly similar to the recognition site for OxyR, the primary regulatory protein for bacterial oxidative stress (S1 Fig). We then examined the interaction between OxyR and this putative operator by electrophoresis mobility shift assay (EMSA). Incubation of a probe containing the T6SS-4 promoter with His6-OxyR led to the formation of DNA-protein complexes (Fig 1A). The interactions between His6-OxyR and the T6SS-4 promoter are specific because excessive unlabeled probe abolished the formation of the protein-DNA complex; similarly, mutations in the predicted OxyR binding site disrupted the formation of such complexes (Fig 1A). DNase I footprinting analysis revealed that the putative OxyR binding site was protected from digestion in DNA-OxyR complexes, further indicating the recognition of this DNA element by OxyR (Fig 1B). Thus, OxyR specifically recognizes an operator within the T6SS-4 promoter, most likely to influence its activity.
Next we determined the effects of OxyR on the expression of the T6SS-4 by measuring the transcription of chromosomal PT6SS-4::lacZ fusions. Deletion of oxyR significantly reduced the activity of the promoter, which can be fully restored by a complementation plasmid expressing the regulatory protein (Fig 1C). Consistent with the operon-like organization of the T6SS-4 structural genes, qRT-PCR analyses revealed that the expression of other T6SS-4 components such as clpV4 (ypk_3559), hcp4 (ypk_3563) and vgrG4 (ypk_3558), also required OxyR, in a manner highly similar to katG (ypk_3388), one of the established target genes of this regulatory protein (Fig 1D). Further analysis at protein level indicated that similar regulation was observed for protein production of hcp4, in which deletion of OxyR diminished its cellular level (Fig 1E). In contrast, the expression of T6SS1-3 in Yptb was not detectably affected by the deletion of oxyR, pointing to the specificity of the regulation (Fig 1F). Thus, OxyR specifically activates the expression of T6SS-4, suggesting that its function is relevant to the environmental cues sensed by this regulatory protein.
OxyR is a global oxidative stress regulator that controls the expression of genes such as katG, gorA, grxA, ahpCF and oxyS, all important in protection against oxidative stress [15]. The activation of T6SS-4 by OxyR prompted us to examine whether this transporter plays a role in protection against oxidative stress. We thus determined the viability of Yptb T6SS-4 mutants after H2O2 challenge. As a control, the ΔkatG mutant was expectedly sensitive to H2O2, so were the mutants defective in the Cu/Zn or Fe/Mn superoxide dismutase (SOD) (Fig 2A). The ΔoxyR mutant and mutants lacking essential T6SS-4 structural genes are significantly more sensitive to H2O2 than wild-type bacteria (Fig 2A). For example, about 39.0% wild-type bacteria survived after exposing to H2O2 for 1 hr, but the survival rates for the ΔicmF4 mutant were only 10.7% (Fig 2A). Further, the sensitivity of the ΔclpV4 mutant to oxidative stress can be fully alleviated by expressing wild-type gene but not the E304A/E677A mutant deficient in hydrolyzing ATP (Fig 2B), supporting a role of T6SS-4 in combating oxidative stress. Importantly, the conductance of the T6SS-4 channel is required for such resistance as the VgrG4-GFP fusion known to block T6SS secretion [13] rendered the bacteria sensitive to H2O2; such inhibition did not occur when VgrG4 was fused to 6 tandem histidine residues (VgrG4-His6), a fusion known not to block the function of the transporter [13] (Fig 2C). In agreement with these observations, similar to katG, one of the classical target genes of OxyR, the expression of T6SS-4 was induced by oxidative stress (Fig 2D).
Oxidative stress induces the production of deleterious reactive oxygen species (ROS), including the highly destructive hydroxyl radicals (HRs), which were generated via Fenton chemistry [16,17]. We thus used three fluorescent dyes with ranging specificity, to independently measure intracellular ROS levels in relevant Yptb strains challenged with H2O2. Among the three fluorescent dyes used, 2′,7′-dichlorodihydrofluorescein diacetate (H2DCFDA) detects H2O2 and ROO•, 3′-(p-hydroxyphenyl) fluorescein (HPF) detects HRs, and 5-(and-6)-chloromethyl-2′,7′-dichlorodihydrofluorescein diacetate, acetylester (CM-H2DCFDA) detects H2O2, ROO• and HRs [18–20]. Although the absolute units of the fluorescence signal varied, mutants lacking oxyR or essential components of T6SS-4 contained significantly higher amounts of ROS, especially HRs than wild-type bacteria in assays using each of the three ROS reporter dyes (Fig 3A). The ROS-induced fluorescence signals were specific because no signal was detected in control samples treated with H2O2 but without the dyes or with dyes but not treated with H2O2 (S2 Fig).
Expression of the corresponding genes eliminated the HRs accumulated in the mutants. For example, the level of HRs in the ΔclpV4 mutant was almost completely eliminated by complementation with the wild-type gene but not by the clpV4 mutant (clpV4M) defective in ATPase activity (Fig 3B). Consistently, treatment with chemical HRs mitigation agents 2,2′-dipyridyl or thiourea [21,22] reduced intracellular HRs levels induced by H2O2 in mutant bacteria (Fig 3C), further validating the notion that HRs accumulation contributes to bacterial death. Furthermore, when added into bacterial cultures challenged by oxidative stress, each of these two chemicals was able to increase the survival rates of T6SS-mutants to levels comparable to those of wild-type bacteria (Fig 3D), further validating the notion that HRs accumulation contributes to bacterial death. Together, these results indicate that T6SS-4 is critical in neutralizing HRs accumulated in Yptb under oxidative stress conditions.
Metal ion homeostasis regulates cellular level of HRs [23]. For example, the manganese transporter MntABC and the zinc uptake system ZosA contribute to oxidative stress resistance in bacteria by mitigating HRs [24,25]. To test the hypothesis that T6SS-4 is involved in metal ion homeostasis, we measured the total metal contents in bacteria challenged with H2O2 using inductively coupled plasmon resonance atomic absorption spectrometry (ICP-MS). Our results revealed that deletion of T6SS-4 in the ΔznuCB background significantly lowered intracellular Zn2+ levels and that the expression of znuCB or clpV4 partially restored such defects (Fig 4A). In contrast, the accumulation of Mn2+ was not affected in these mutants (S3 Fig). Such defects clearly did not result from potentially lower live bacterial cells because the 20 min treatment did not detectably affect bacterial viability (S4 Fig). Thus, T6SS-4 likely is involved in Zn2+ uptake. Consistent with this notion, exogenous Zn2+ was able to restore the growth of Yptb in H2O2 but only in the presence of a functional T6SS-4 or ZnuABC transporter (Fig 4B). Bacterial sensitivity to oxidative stress by mutants lacking both the canonical Zn transporter ZnuCB and T6SS-4 cannot be rescued by exogenous Zn2+ (Fig 4B). As expected, mutants lacking SOD(Cu/Zn) or SOD(Fe/Mn) are more sensitive to oxidative stress and such sensitivity can be partially rescued by exogenous Zn ions (Fig 4B). However, the sensitivity of ΔznuCBΔclpV4(Vector) and ΔznuCBΔclpV4(clpV4M) cannot be rescued by exogenous Zn ions (Fig 4B). In contrast, the sensitivity of strains ΔznuCBΔclpV4(znuCB) and ΔznuCBΔclpV4(clpV4) can be rescued by exogenous Zn ions (Fig 4B), further suggesting that T6SS-4 functions similarly to ZnuCB in zinc uptake.
The fact that the T6SS-4 participates in zinc acquisition predicts that the vegetative growth rate of the ΔclpV4ΔznuCB mutant will be affected by Zn2+ starvation under oxidative conditions. This prediction was confirmed by comparing the growth of the wild-type and the ΔclpV4ΔznuCB mutant in the presence of the Zn2+ chelator TPEN (N,N,N′,N′-tetrakis(2-pyridylmethyl) ethylenediamine) under H2O2 stress (S5 Fig). Whereas the growth of all tested strains is nearly identical in YLB medium (S5A Fig), the growth of the ΔclpV4ΔznuCB mutant is severely impaired compared to the wild-type in the presence of TPEN under H2O2 stress. The expression of either znuCB or clpV4 from a complementation plasmid almost completely rescued the sensitivity of the ΔclpV4ΔznuCB mutant to TPEN (S5C Fig). Moreover, the growth defect of the mutant was completely rescued by the addition of excessive Zn2+, further supporting the role for T6SS-4 in Zn2+ acquisition (S5D and S5E Fig).
Because the production of HRs contributes to the cellular toxicity under diverse stress conditions [20,26], we reasoned that if T6SS-4 functions in Zn2+ uptake, it should be required for maximal bacterial survival under these stresses. Indeed, in the ΔznuCB mutant background, T6SS-4 is required for the resistance to diverse stress conditions such as low pH, high temperature, heavy metal, diamide and antibiotic (gentamicin) (Fig 4C). Taken together, these results point to a role of T6SS-4 in importing Zn2+ from the environment under diverse stress conditions.
The observation that TSSS-4 is involved in Zn2+ uptake points to the notion that the expression of this transporter should be responsive to low Zn2+ conditions. Indeed, the addition of exogenous Zn2+ repressed the expression of T6SS-4 in wild-type Yptb (Fig 4D). Chelating Zn2+ from the medium by TPEN led to robust expression from the promoter, and such induction can be repressed by exogenous zinc ions (Fig 4D). Thus, the expression of T6SS-4 is responsive to the levels of Zn2+ in the environment, which is consistent with its role in acquiring this metal ion from the extracellular milieu.
Zinc transport by T6SS-4 can be achieved by direct ion translocation via the secretion channel or by a Zn2+-binding carrier protein translocated by the secretion system. We distinguished between these two possibilities by analyzing predicted proteins adjacent to structural components of T6SS-4 for putative Zn2+-binding motifs with HHpred [27]. Such analyses revealed that Ypk_3549, a 117-residue protein encoded by a gene located at the end of the T6SS-4 gene cluster, contains a putative zinc finger motif (S6 Fig). No putative promoter can be identified upstream of ypk_3549, suggesting that this gene is part of the T6SS-4 operon. Indeed, similar to structural components of T6SS-4, the expression of ypk_3549 is activated by OxyR under oxidative conditions (S7 Fig). BLAST analysis revealed that homologs of this gene are present in the genomes of Yersinia pestis, Serratia marcescens and possibly Burkholderia oklahomensis. Because Ypk_3549 is a putative substrate of T6SS-4 that may bind Zn2+, we designated it YezP (Yersinia extracellular zinc-binding protein).
Analysis with isothermal titration calorimetry [28] revealed that YezP bound Zn2+ with a Kd of 0.53 μM (Fig 5A upper panel). Importantly, mutation of residue histidine-76 predicted to participate in the formation of the zinc finger reduced its affinity to Zn2+ for more than 150-fold (Kd = 152.14 μM) (Fig 5A lower panel). Unexpectedly, the YezPH76A mutant still bound to Zn2+, although at a markedly lower affinity. Similar Zn2+-binding activity of YezP and YezPH76A was detected when the interaction was measured with 4-(2-pyridylazo) resorcinol (PAR) [29] (S8 Fig). However, this protein did not detectably bind to iron ions (S9 Fig), which differs from the zincophore yersiniabactin from Yersinia spp. capable of binding both zinc and iron ions [30]. The residual Zn2+ binding activity of YezPH76A may result from a second noncannonical zinc-binding motif in the protein. Consistent with its zinc-binding activity and OxyR-dependent expression, YezP is required for Zn2+ accumulation in the cells in the ΔznuCB background (S10 Fig). Similarly, in line with the fact that low extracellular Zn2+ concentrations induced its expression, higher levels of YezP were detected in bacterial culture supernatant when Zn2+ was sequestered by the chelator TPEN (S11 Fig).
The above results suggest that T6SS-4 is involved in either the secretion or import of YezP. To distinguish between these two models, we expressed YezP-VSVG in relevant Yptb strains and examined its secretion. Significant amounts of YezP-VSVG can be readily detected in culture supernatant of wild-type bacteria (Fig 5B upper panel). Mutations in T6SS-4 structural genes almost completely abrogated the secretion of YezP; the residual secretion was completely abolished in a mutant lacking all 4 T6SSs in Yptb (Fig 5B upper panel). Furthermore, deletion of yezP did not affect the secretion of Hcp4 (Fig 5B lower panel), indicating that this protein was not involved in substrate secretion by T6SS-4. Interestingly, akin to YezP, the secretion of Hcp4 was completely abolished only in a mutant lacking all 4 T6SS of Yptb (Fig 5B lower panel), indicating the existence of limited substrate cross recognition among these transporters. These results establish that YezP is a substrate secreted by T6SS-4.
That YezP is a zinc-binding substrate of T6SS-4 suggests that it is required for maximal bacterial survival under oxidative challenge. Indeed, the ΔyezP strain exhibited sensitivity to H2O2 at levels similar to those of T6SS-4 mutants and such sensitivity can be fully complemented by wild-type, and partially by the H76A mutant which still retains residual Zn2+ binding activity (Fig 6A). We next determined whether recombinant YezP restored the ability of relevant Yptb mutants to survive oxidative challenge. Inclusion of recombinant YezP in cultures of the ΔyezP strain fully restored its resistance to H2O2 (Fig 6A). More importantly, recombinant YezP protein also protected the T6SS-4 mutant ΔclpV4 from toxicity imposed by H2O2 (Fig 6A), indicating that after T6SS-4-mediated translocation, zinc uptake by YezP occurs independently of the secretion system. Consistent with its partial Zn2+-binding activity, YezPH76A still detectably conferred resistance to oxidative stress in mutants defective in T6SS-4 or its coding gene (Fig 6A). In agreement with its role in Zn2+ acquisition to neutralize HRs, deletion of yezP resulted in accumulation of these harmful agents in bacterial cells to levels similar to those observed in T6SS-4 mutants (Fig 6B). Such accumulation can be eliminated by either expression of yezP from a plasmid or by recombinant YezP protein (Fig 6B).
We further determined the importance of Zn2+ sequestration by recombinant YezP by adding the zinc chelator TPEN to the protein solution used for complementation. The inhibitory effects of TPEN can be neutralized by recombinant YezP in a dose-dependent manner (Fig 6C). Although the addition of 0.05 μΜ YezP protein increased the survival rate of the ΔyezP mutant treated with 2.5 μΜ TPEN, it did not increase the survival rate of the same mutant in the presence of 5 μΜ TPEN when the concentration of Zn2+ was 1.0 μM. Instead, the inhibition by 5 μΜ TPEN in the presence of 0.05 μΜ YezP can be significantly reversed by the addition of exogenous zinc (Fig 6C). Similar to T6SS-4 mutants, the ΔyezP mutant was sensitive to agents such as low pH, heavy metal, high temperature, diamide and antibiotic (gentamicin) that induce HRs production [17] (Fig 6D), further indicating that zinc transport by T6SS-4 functions to combat cellular stress induced by a wide spectrum of environmental cues.
The host immune system imposes significant stress to a pathogen. Yptb is an enteric pathogen with a tropism for lymphoid tissue; it also hijacks macrophages as Trojan horses for dissemination and subsequent systemic infection [31]. Oxidative burst is an important microbial killing mechanism in phagocytes. We thus tested the induction of T6SS-4 after the bacterium being phagocytosed by primary macrophages. Wild-type Yptb was used to infect primary macrophages from C57BL/6 mice for different durations and the expression of yezP, clpV4 and vgrG4 was measured by qRT-PCR. Compared to broth grown bacteria, the expression of clpV4 and vgrG4 was induced for about 13–20 folds 15 min after infection, and 25–60 folds of induction was detected at 30 min post infection (Fig 7A). Although at lower rates, the expression of yezP was also significantly induced in phagocytosed bacteria (Fig 7A).
The induction of T6SS-4 in primary macrophages suggests that this transporter may be important for the virulence of Yptb in animal infection. We thus examined this hypothesis by orogastrically inoculating relevant bacterial strains into C57BL/6 mice. Wild-type bacteria caused more than 50% lethality within two weeks of inoculation. On the other hand, consistent with the stress sensitivity phenotypes, mice infected with mutants lacking T6SS-4 or yezP survived better at similar rates (Fig 7B). Similarly, the mutant lacking the classical zinc transporter znuCB that plays important roles in competition with vertebrate host for Zn2+ is less virulent to mice [30]. Notably, mutations of znuCB together with T6SS-4 or yezP resulted in mutants that almost completely lost the virulence against mice (Fig 7B), further implying the importance of T6SS-4 in the resistance to host nutritional immunity. Thus, zinc transportation by T6SS-4 plays an important role in the interactions of Yptb with mammalian hosts.
The best-studied function of T6SSs is their role in the killing of competing species in specific niches using bacteriolytic effectors [2]. Previous studies have suggested roles of T6SSs in nonbiological challenges such as stress resistance [13,14], but the underlying mechanisms remain largely unknown. In this report, we found that the T6SS-4 from Y. pseudotuberculosis functions to combat multiple adverse stresses and host nutritional immunity by translocating a zinc-binding effector.
Bacterial cells need to deal with insults of distinct origins in different phases of their life cycle or in different environmental niches. The production of detrimental HRs is emerging as a potentially important consequence of diverse environmental challenges [17,20,26]. In addition to their roles as cofactors in many essential enzymes, transition ions such as Zn2+ and Mn2+ are capable of mitigating HRs to reduce their damage [24,32]. Consistent with the notion that T6SS-4 functions to combat oxidative stress induced by diverse cues, its expression is also induced by high osmolality and low pH conditions via the osmotic/acid stress regulator OmpR [13,14], which is consistent with its role in the resistance to a broad range of adverse stresses. Similarly, although the mechanism has not yet been well studied, a T6SS in Vibrio anguillarum is regulated by the general stress response regulator RpoS and is involved in its resistance to hydrogen peroxide, ethanol and low pH [12]. The response of T6SS-4 to distinctly different signals via multiple regulatory proteins [13,14] and the role of Zn2+ in HRs mitigation provide a molecular explanation for “cross-protection”, a phenomenon in which bacterial cells subjected to one stressful condition often became resistant to stress created by distinctly different insults [33].
Metal ions can be transported into bacterial cells by specific ion transporter [34] or by chelators, such as siderophores, which are high-affinity iron-binding molecules [35]. Our discovery of T6SS in bacterial ion acquisition significantly expanded the function of specialized protein secretion system. Due to their multiple components, the expression and assembly of T6SSs presumably will consume more energy. As a result, this system may only be activated when metal ions imported by the more classical transporters are not sufficient. Alternatively, the expression or activity of the classical metal ion transporters may be inhibited under certain environmental conditions. We proposed a model that under normal conditions, the ZnuCB transporter fulfills the need of Zn2+ of the cells, and the T6SS-4 in Yptb contributes to this process as a Zn2+ scavenger when the bacterium encounters stress conditions that lead to the production of HRs and potentially other cell damaging mechanisms by oxidative or other adverse conditions capable of activating OxyR (Fig 8). The establishment of a role of T6SS in dealing with non-biological challenges may provide an explanation to the wide spread of these protein secretion systems. It will be of great interest to determine whether bacteria not normally associated with a eukaryotic host employ T6SS to facilitate their survival in unfavorable environmental conditions.
For pathogenic bacteria, one benefit of the metal ion acquisition activity of T6SS is to compete for Zn2+ within the host, or other essential metal ions to fight against nutritional immunity [36]. Given the essential role of Zn2+ in bacterial physiology, it is not surprising that the ability of T6SS-4 in acquiring Zn2+ might offer Yptb an advantage in pathogenesis because Zn2+ in mammalian hosts is strictly sequestered by a defense mechanism termed nutritional immunity [36]. Indeed, the classical zinc transporter ZnuCB was found to play pivotal roles in competition against vertebrate host for Zn2+ [36–38]. Accordingly, deletion of znuCB attenuated the pathogenicity of important pathogens such as Acinetobacter baumannii [38], Brucella abortus [39] and Campylobacter jejuni [40]. Consistently, we found Yptb T6SS-4 mutants are attenuated in virulence against mice. Notably, mutations of T6SS-4 or yezP together with znuCB resulted in mutants that are almost completely avirulent in a mouse infection model, indicating the importance of T6SS-4 in the resistance to host nutritional immunity. These results may also explain the observation that mutations in znuCB did not affect the virulence of Y. pestis [41]. Interestingly, the siderophore yersiniabactin has recently been shown to participate in Zn2+ acquisition in Y. pestis [42]. Furthermore, mutants lacking both the ZnuABC system and yersiniabactin are defective in virulence [42]. The fact that the ybt locus for yersiniabactin biosynthesis are absent in the Yptb strain YpIII used in our current study (http://www.ncbi.nlm.nih.gov/nuccore/169748796) may explain the strong phenotypes of mutants defective in both T6SS-4 and ZnuCB. It will be interesting to determine whether Y. pestis mutants lacking all three known Zn2+ acquisition systems display further reduction in virulence.
Zn2+ transported by T6SS-4 of Y. pestis may compensate the effects caused by the loss of the canonical transporter. Lethal systemic infection by Yptb has multiple phases, including initial survival in the gastrointestinal track, invasion through M-cells into the Peyer’s patches and the subsequent trafficking to the deep tissue via macrophages [43]. The observed loss of virulence can result from impairment in the competitiveness against the microflora prior to infection or by the inability to compete with Zn2+ sequestration mechanisms in host cells or a combination of both. Hence, this finding provided a new perspective for revealing the mechanisms of T6SS in pathogenesis.
Zinc acquisition by secreted zinc-binding proteins seems to be a mechanism shared by taxonomically diverse microorganisms. Secreted Zn2+-chelating compounds (zincophores) analogous to siderophores have been identified in pathogenic bacteria such as Pseudomonas aeruginosa [44] and Y. pestis [42]. Recently, the Zn2+-binding protein Pra1 important for zinc acquisition from hosts by the fungal pathogen Candida albicans was identified [45]. Similar to Pra1, YezP appears to contain multiple Zn2+-binding motifs (Fig 5). In Mycobacterium tuberculosis, the type VII secretion system ESX-3 is necessary for optimal growth in zinc-limited conditions, implying the involvement of a similar mechanism in zinc acquisition by this pathogen [46].
The affinity of YezP for Zn2+ (Kd = 0.53 μM) is considerably lower than that of canonical Zn importers such as ZnuA (<20 nM) [47] and ZinT (22 nM) [48], but it is comparable to or even higher than other Zn importers like YiiP (1000 nM) [49], the cation diffusion facilitator (CDF) (865 nM) [50] and the membrane zinc transporter (MTP1) (23 μM) [51]. Despite the relatively lower affinity for Zn2+, YezP is crucial for Yptb survival in stress conditions or for successful colonization of a mammalian host, particularly in the absence of the canonical Zn2+ transporter ZnuCB. Under our experimental conditions, even in the presence of strong chelators such as TPEN, recombinant YezP was able to provide Zn2+ sufficient for the cells to survive under oxidative stress (Fig 6). Future elucidation of the mechanism of the transfer of Zn2+ to the cell by YezP may explain how this protein functions in the presence of a chelator with much higher affinity for the ions. Similarly, the exact mechanism and timing of YezP’s contribution to the infection process needs further investigations.
The Zn2+-binding protein substrate YezP represented a novel type of T6SS effectors distinct from those extensively-studied bacteriolytic toxins or eukaryotic cell-targeting effectors. Our study suggests that T6SS-4 secretes a proteinaceous Zn2+ chelator as a strategy to acquire this nutrient, which implies the existence of a mechanism for subsequent acquisition of zinc from this protein by bacterial cells. Clearly, this process occurs independent of T6SS-4 as recombinant YezP was able to rescue the sensitivity of its mutant to oxidative stress. Evidently, the implication of ion transport by T6SS is broader than bacterial interactions with hosts or competing species. The ability to more effectively acquire nutrients is beneficial to the bacteria, pathogenic or environmental species, when they encounter challenges in specific niches. Future study aiming at the identification of the machinery for importing Zn2+ from YezP or the Zn2+/protein complex surely will lead to a better understanding of mechanism not only for the function of these widely distributed secretion systems, but also for metal ion uptake by bacteria.
All mouse experimental procedures were performed in accordance with the Regulations for the Administration of Affairs Concerning Experimental Animals approved by the State Council of People’s Republic of China. The protocol was approved by the Animal Welfare and Research Ethics Committee of Northwest A&F University (protocol number: NWAFU 2014002).
Bacterial strains and plasmids used in this study are listed in S1 Table. Escherichia coli were grown in LB with appropriate antibiotics at 37°C. Y. pseudotuberculosis (Yptb) strains were cultured in Yersinia-Luria-Bertani (YLB) broth (1% tryptone, 0.5% yeast extract, 0.5% NaCl) or M9 medium (Na2HPO4, 6g L-1; KH2PO4, 3g L-1; NaCl, 0.5g L-1; NH4Cl, 1g L-1; MgSO4, 1 mM; CaCl2, 0.1 mM; glucose 0.2%) at 26°C with appropriate antibiotics when necessary. The Y. pseudotuberculosis strain YPIII was the parent of all derivatives used in this study. In-frame deletions were generated by the method described previously [13]. Antibiotics were added at the following concentrations: nalidixic acid, 15 μg ml-1; ampicillin, 100 μg ml-1; kanamycin, 50 μg ml-1; tetracycline, 10 μg ml-1; chloramphenicol, 30 μg ml-1.
Primers used in this study are listed in S2 Table. The lacZ fusion reporter vectors pDM4-T6SS1-4p::lacZ were made in our previous study [52]. To construct the lacZ fusion reporter vector pDM4-katGp::lacZ, primers katGp-SalI-F/KatGp-XbaI-R were used to amplify the 585 bp katG promoter fragment from Yptb genomic DNA. The PCR product was digested with SalI/XbaI and inserted into similarly digested pDM4-T6SS-4p::lacZ to produce pDM4-katGp::lacZ. To construct T6SS-4 promoter with mutations in the OxyR binding site, overlap PCR was performed to replace the consensus binding sites (60 bp) with identical amount of irrelevant base pairs. Briefly, to replace the OxyR binding site, primer pairs T6SS-4p-SalI-F/T6SS-4pM-R and T6SS-4pM-F/T6SS-4p-XbaI-R were used to amplify the up-fragment and down-fragment of T6SS-4 promoter, respectively. Overlap PCR was carried out using the up-fragment and down-fragment as template and T6SS-4p-SalI-F/T6SS-4p-XbaI-R as primer pair to obtain the DNA fragment T6SS-4pM. This fragment was further digested with SalI and XbaI and inserted into similar digested pDM4-T6SS-4p::lacZ to construct pDM4-T6SS-4pM::lacZ.
The plasmid pDM4-ΔoxyR (ypk_4079) was used to construct the ΔoxyR in-frame deletion mutant of Yptb. A 918-bp and a 900-bp fragments flanking oxyR were amplified with primer pair oxyR-1F-BglII/oxyR-1R and oxyR-2F/oxyR-2R-SalI, respectively. The upstream and downstream PCR fragments were ligated by overlap PCR. The resulting PCR products were digested with SalI and BglII, and inserted into the SalI/BglII site of pDM4 to produce pDM4-ΔoxyR. The knock-out plasmid pDM4-ΔznuCB (ypk_2141–2142), pDM4-ΔicmF4 (ypk_3550), pDM4-Δhcp1 (ypk_0385), pDM4-Δhcp2 (ypk_0803), pDM4-Δhcp3 (ypk_1481), pDM4-ΔkatG (ypk_3388), pDM4-Δsod(Fe/Mn) (ypk_1863), pDM4-Δsod(Cu/Zn) (ypk_3445) and pDM4-ΔyezP (ypk_3549) were constructed in similar manners by using primers list in S2 Table.
To complement the oxyR mutant, primers oxyR-F-SphI/oxyR-R-SalI were used to amplify the oxyR gene fragment from Yptb genomic DNA. The PCR product was digested with SphI/SalI and was inserted into similarly digested pKT100 to produce pKT100-oxyR. The complementary plasmid pKT100-znuCB and pKT100-yezP was constructed in similar manners using primers znuCB-F-SphI/znuCB-R-SalI and 3549-F-SphI/3549-R-SalI, respectively. The complementary plasmid pKT100-clpV4 and pKT100-clpV4M were made in our previous study [13].
Site-directed mutagenesis was carried out by overlap PCR to substitute the histidine residue at position 76 of YPK_3549 (YezP) into an alanine residue (YezPH76A). Briefly, DNA of mutant YezPH76A was amplified by two rounds of PCR. Primer pairs 3549-F-SphI/3549-H76A-R and 3549FH76A-F/3549-R-SalI were used to amplify segments 1 and 2 respectively. The second round of PCR was carried out by using 3549-F-SphI/3549-R-SalI as primer pair while fragment 1 and fragment 2 as templates to obtain the YezPH76A fragment. The YezPH76A DNA fragment was digested by SphI/SalI and cloned into similar digested pKT100 to produce pKT100-yezPH76A.
To express His6-tagged OxyR and Fur (ferric uptake regulator; ypk_2991), primers oxyR-F-BamHI/oxyR-R-SalI and fur-F-BamHI/fur-R-SalI were used to amplify oxyR and fur fragments from genomic DNA of Yptb. The PCR products of oxyR and fur were digested with BamHI/SalI and inserted into the BamHI/SalI sites of pET28a resulting in plasmids pET28a-oxyR and pET28a-fur. To express GST-tagged YezP, primers 3549-F-BamHI and 3549-R-SalI were used to amplify the yezP gene from genomic DNA of Yptb. The PCR product of yezP was digested with BamHI/SalI and inserted into the BamHI/SalI sites of pGEX6P-1 resulting in plasmid pGEX6P-1-yezP. To construct the site-directed mutagenesis plasmid pGEX6P-1-yezPH76A, primers 3549-F-BamHI and 3549-R-SalI were used to amplify the yezPH76A fragment from pKT100-yezPH76A plasmid DNA and was subcloned into similarly digested pGEX6P-1.
Plasmids pME6032-yezP-VSVG was constructed for protein secretion assay. Briefly, primers 3549-F-EcoRI and 3549taa-R-VSVG-BglII were used to amplify the yezP gene from Yptb genomic DNA. The PCR product of yezP-VSVG were digested with EcoRI/BglII and inserted into the EcoRI/BglII site of pME6032 to generate pME6032-yezP-VSVG. The plasmid pME6032-hcp4-VSVG was constructed in similar manners by using primers hcp4-F-EcoRI and hcp4taa-R-VSVG-BglII.
For complementation, complementary plasmids pKT100-oxyR, pKT100-znuCB, pKT100-clpV4, pKT100-clpV4M, pKT100-yezP and pKT100-yezPH76A were introduced into respective mutants by electroporation. The integrity of the insert in all constructs was confirmed by DNA sequencing.
To express and purify His6- and GST-tagged recombinant proteins, pET28a and pGEX6p-1 derivatives were transformed into E. coli BL21(DE3) and XL1-Blue competent cells, respectively. For protein production, bacteria were grown at 37°C in LB medium to an OD600 of 0.5, shifted to 22°C and then induced with 0.2–0.4 mM IPTG, and cultivated for an additional 12 h at 22°C. Harvested cells were disrupted by sonification and purified with the His•Bind Ni-NTA resin or GST•Bind Resin (Novagen, Madison, WI) according to manufacturer’s instructions. The GST tag was removed by incubation with PreScission Protease (GE healthcare) for 20 h at 4°C, and tag removed proteins were eluted from the GST•Bind Resin with PBS. The purity of the purified protein was verified as >95% homogeneity with SDS-PAGE analysis. Protein concentrations were determined by the Bradford assay [53].
Secretion assays for YezP (Ypk_3549) and Hcp4 were performed according to described methods [54]. All samples used for secretion assays in this study were taken at mid-exponential phase corresponding to an OD600 = 1.5–2.0. Briefly, strains were inoculated into 100 ml YLB broth and incubated with continuous shaking until OD600 reached 1.5–2.0 at 26°C. 2 ml culture was centrifuged and the cell pellet was resuspended in 100 μl SDS-sample buffer; this whole-cell lysate sample was defined as YezPIN. 90 ml of the culture was centrifuged, then the supernatant was filtered through a 0.22 μm filter (Millipore, MA, USA), and the proteins were extracted by filtration over a nitrocellulose filter (BA85) (Whatman, Germany) three times. The filter was soaked in 100 μl SDS sample buffer for 15 min at 65°C to recover the proteins present, and the sample was defined as YezPOUT. All samples were normalized to the OD600 of the culture and volume used in preparation. Secretion assays for Hcp4 was carried out by a similar procedure.
Western blot analysis was performed as previously described [13]. Samples were resolved by SDS-PAGE and transferred onto PVDF membranes (Millipore). The membrane was blocked in 5% (w/v) nonfat milk for 4 h at room temperature, and incubated with primary antibodies at 4°C overnight: anti-Hcp4, 1:500; anti-VSVG (Santa Cruz biotechnology, USA), 1:500; anti-ICDH, 1:6000; anti-Pgi, 1: 2,000. The membrane was washed three times in TBST buffer (50 mM Tris, 150 mM NaCl, 0.05% Tween 20, pH 7.4), and incubated with 1:5,000 dilution of horseradish peroxidase conjugated secondary antibodies (Shanghai Genomics) for 1 h. Signals were detected using the ECL plus kit (GE Healthcare, Piscataway, NJ) following the manufacturer's specified protocol. The Hcp4, Pgi and ICDH antisera were made in our previous studies [13,55].
The lacZ fusion reporter vectors pDM4-T6SS-1p::lacZ, pDM4-T6SS-2p::lacZ, pDM4-T6SS-3p::lacZ, pDM4-T6SS-4p::lacZ and pDM4-katGp::lacZ were transformed into E. coli S17-1λpir and mated with Yptb strains according to the procedure described previously [13]. The lacZ fusion reporter strains were grown in YLB broth and β-galactosidase activity was assayed with ONPG as the substrate [56]. The β-galactosidase results shown represent the mean of one representative assay performed in triplicate, and error bars represent standard deviation. Statistical analysis was carried out with Student’s t-test.
DNase I footprinting assays were performed according to [57] with minor modifications. The promoter region of T6SS-4 was PCR amplified with primers T6p4 footprinting-F/T6p4 footprinting-R and the fragment was cloned into the pMD-18T vector (TaKaRa), which was further used as the template for preparation of fluorescent FAM labeled probes with primers M13R(FAM-labeled) and M13F(-47). The FAM-labeled probes were purified by the Wizard SV Gel and PCR Clean-Up System (Promega) and quantified with NanoDrop 2000C (Thermo). For the DNase I footprinting assay, 400 ng probes were incubated with different amounts of His6-OxyR in a total volume of 40 μl in the same buffer. After incubation for 30 min at 30°C, 10 μl solution containing about 0.010 unit DNase I (Promega) and 100 nmol freshly prepared CaCl2 was added and further incubate for 1 min at 25°C. The reaction was stopped by adding 140 μl DNase I stop solution (200 mM unbuffered sodium acetate, 30 mM EDTA and 0.15% SDS). Samples were then extracted with phenol/chloroform, precipitated with ethanol and the pellets were dissolved in 35 μl MiniQ water. The preparation of the DNA ladder, electrophoresis and data analysis were the same as described before [57], except that the GeneScan-LIZ500 size standard (Applied Biosystems) was used.
Electrophoretic mobility shift assay was performed using biotin 5′-end labeled promoter probes. Bio-T6SS-4p and Bio-T6SS-4pM, amplified from pDM4-T6SS-4p::lacZ and pDM4-T6SS-4pM::lacZ, respectively, with primers T6p-oxyR-F-5′biotin/T6p-oxyR-R-5′biotin. The unlabeled T6SS-4p competitor DNA was amplified from pDM4-T6SS-4p::lacZ with primers T6p-oxyR-F/T6p-oxyR-R. All PCR fragments were purified by EasyPure Quick Gel Extraction Kit (TransGen Biotech, Beijing, China). Each 20-μl EMSA reaction solutions were prepared by adding the following components according to the manufacturer’s protocol (LightShift Chemiluminescent EMSA kit; Thermo Fisher Scientific): 1×binding buffer, 50 ng poly (dI-dC), 2.5% glycerol, 0.05% NP-40, 5 mM MgCl2, 20 fmol Biotin-DNA, 4 pmol unlabeled DNA as competitor and different concentrations of proteins. Reaction solutions were incubated for 20 min at room temperature. The protein-probes mixture was separated in a 6% polyacrylamide native gel and transferred to a Biodyne B Nylon membrane (Thermo Fisher Scientific). Migration of biotin-labeled probes was detected by streptavidin-horseradish peroxidase conjugates that bind to biotin and chemiluminescent substrate according to the manufacturer’s protocol.
Bacteria were harvested during the mid-exponential phase and RNA was extracted using the RNAprep Pure Cell/Bacteria Kit and treated with RNase-free DNase (TIANGEN, Beijing, China). The purity and concentration of the RNA were determined by gel electrophoresis and spectrophotometer (NanoDrop, Thermo Scientific). First-strand cDNA was reverse transcribed from 1 μg of total RNA with the TransScript First-Strand cDNA Synthesis SuperMix (TransGen Biotech, Beijing, China). Quantitative real-time PCR (qRT-PCR) was performed in CFX96 Real-Time PCR Detection System (Bio-Rad, USA) with TransStart Green qPCR SuperMix (TransGen Biotech, Beijing, China). For all primer sets (S2 Table), the following cycling parameters were used: 95°C for 30 s followed by 40 cycles of 94°C for 15 s, 50°C for 30 s. For standardization of results, the relative abundance of 16S rRNA was used as the internal standard.
Mid-exponential phase Yptb strains grown in YLB medium were collected, washed, and diluted 50-fold into M9 medium or PBS buffer with 0.4% glucose containing H2O2 (1.5 mM), CdCl2 (0.1 mM), Diamide (0.3 mM), or Gentamicin (0.2 μg/ml) and incubated at 26°C for 1 h, or subjected to heat shock (42°C) for 0.5 h. After treatment, the cultures were serially diluted and plated onto YLB agar plates, and colonies were counted after 20 h growth at 26°C. Percentage survival was calculated by dividing number of CFU of stressed cells by number of CFU of cells without stress [13]. All these assays were performed in triplicate at least three times.
To detect intracellular ROS, the fluorescent reporter dye 3′-(p-hydroxyphenyl) fluorescein (HPF, Invitrogen), 5-(and-6)-chloromethyl-2′,7′-dichlorodihydrofluorescein diacetate, acetylester (CM-H2DCFDA, Life Technologies) and 2′,7′-dichlorodihydrofluorescein diacetate (H2DCFDA, Invitrogen) were used, as previously described [20]. Briefly, 1 ml samples were collected after treatment and then resuspended in 1 ml of PBS containing 10 μM HPF, CM-H2DCFDA or H2DCFDA, respectively. Samples were incubated in dark for 20 min. The cells were then pelleted, the supernatant removed, and were resuspended in 1 ml filtere-sterilized PBS. Two hundred microliters of the resultant cell suspension were transferred to a dark 96-well plate. Fluorescence signals were measured using a SpectraMax M2 Plate Reader (Molecular Devices) with excitation/emission wavelengths of 490/515 nm (HPF), 495/520 nm (CM-H2DCFDA and H2DCFDA). The results shown represented the mean of one representative assay performed in triplicate, and error bars represent standard deviation. Statistical analysis was carried out with Student’s t-test.
Intracellular ion content was determined as described previously [58,59]. Briefly, cells were grown in YLB until mid-exponential phase. After 20 ml culture solutions were collected and washed with PBS for two times, the pellets were re-dissolved in 20 ml PBS buffer containing 0.4% glucose, 1.5 mM H2O2 and 1 μM Zn2+, and then incubated further for 20 min. These cultures were centrifuged at 4000 rpm for 10 min. The wet cell pellet weight was measured and bacteria were chemically lysed using Bugbuster (Novagen, Madison, WI) according to the manufacturer’s instructions. Bacteria were resuspended in Bugbuster solution by pipetting and incubation on a rotating mixer at a slow setting for 20 min. Total protein for each sample was measured by using NanoDrop ND-1000 spectrophotometer (NanoDrop Technologies) according to the manufacturer’s instructions. Each sample was diluted 100-fold in 2% molecular grade nitric acid to a total volume of 10 ml. Samples were analyzed by Inductively coupled plasma mass spectrometry (ICP-MS) (Varian 802-MS), and the results were corrected using the appropriate buffers for reference and dilution factors. Triplicate cultures of each strain were analyzed during a single experiment and the experiment was repeated at least three times.
Zn2+ binding was measured using isothermal titration calorimetry (ITC) at 25°C with a NANO-ITC 2G microcalorimeter (TA Instruments, USA). A control experiment in the absence of protein was performed to measure the heat generated due to Zn2+ dilution in the buffer. To obtain apo-protein, samples were dialyzed for 10 h at 4°C against 250 μM EDTA and 5 mM o-phenanthroline in 50 mM HEPES (pH 8.0), 150 mM NaCl, 15% glycerol, followed by three dialysis steps in 50 mM HEPES (pH 8.0), 150 mM NaCl, 15% glycerol to remove EDTA and o-phenanthroline. The dialysis buffer was used to prepare a 0.5 mM ZnSO4 solution used for titration. Protein concentrations in the sample solution were 40 μM. After a stable baseline was achieved the ZnSO4 titration was performed by a total of 25 injections of 5 μl into protein solutions (volume = 1.5 ml) until the protein sample was saturated with zinc. Blank titrations of the ZnSO4 solution into the dialysis buffer were performed to correct for the dilution heat of the zinc solution. Data reduction and analysis were performed with the Nano Analyze software (TA Instruments) fitting them to an independent binding model [60].
Zn2+ binding was also detected by using the Zn2+-binding dye 4-(2-pyridylazo)-resorcinol (PAR) [29]. Free PAR shows a peak absorbance at about 410 nm, shifting to 500 nm when PAR binds Zn2+. To determine if proteins binds Zn2+, increasing concentrations proteins (0–5 μM) were added to solutions containing 10 μM PAR and 5 μM Zn2+ in 50 mM HEPES (pH 8.0). A control experiment in the absence of protein was performed to obtain the spectra for free PAR bound to Zn2+, compared to the spectra gained in the presence of protein. A decrease in the absorbance at 500 nm, accompanied by an increase in the absorbance at 410 nm exhibits binding of Zn2+ by the protein.
Purified recombinant His6-Fur (Ypk_2991, ferric uptake regulator) and YezP proteins in 50 mM Tris-HCl (pH 7.4), 150 mM NaCl were incubated with 1 mM ferrous sulphate for 1 h at room temperature. The products were resolved on a 15% native PAGE. The gel was then stained with potassium ferricyanide solution (100 mM K3[Fe(CN)6] in 50 mM Tris-HCl, pH 7.4, 100 mM NaCl) for 10 min in the dark and destained with 10% trichloroacetic acid/methanol solution [61]. After taking an image of the stained gel, it was subjected to Coomassie blue staining using standard techniques. The iron binding protein Fur was used as a positive control.
Bone marrow derived macrophages (BMDMs) from 6-week old C57BL/6 mice were prepared as previously described [62] and seeded in six-well multiplates at the density of 2x106 cells per well. BMDMs were challenged with the indicated Y. pseudotuberculosis strains, which were cultured in YLB at 28°C for 20 hours before infection, at an MOI of 10. At the indicated time points, infected macrophages were washed 3 times with HBSS (hank’s balanced salt solution) to remove extracellular bacteria before being collected for total mRNA extraction using RNAqueous Total RNA Isolation Kit per manufacturer’s instruction. Indicated mRNA species were quantified using SYBR Green Real-Time PCR Master Mixes system (Life Technology), bacterial cells grown in YLB broth were used as controls. For standardization of results, the relative abundance of 16S rRNA was used as the internal standard.
All mice were maintained and handled in accordance with the animal welfare assurance policy issued by Northwest A&F University. Mid-exponential phase Yptb strains grown in YLB medium at 26°C, washed twice in sterilized PBS and used for orogastric infection of 6–8 weeks old female C57BL/6 mice using a ball-tipped feeding needle. For survival assays 3×109 bacteria of each strain were applied to different groups of mice, and the survival rate of the mice was determined by monitoring the survival everyday for 21 days [63,64].
Statistical analyses of survival assay, intracellular ion content determination, ROS determination and expression data were performed using paired two-tailed Student’s t-test. Survival times were analyzed using Kaplan-Meyer curves and comparisons were performed using the Log-Rank test. Statistical analyses were performed using GraphPad Prism Software (GraphPad Software, San Diego California USA).
Accession numbers for the genes described in this study in NCBI are: yezP, ypk_3549; clpV4, ypk_3559; hcp4, ypk_3563; icmF4, ypk_3550; vgrG4, ypk_3558; hcp1, ypk_0385; hcp2, ypk_0803; hcp3, ypk_1481; oxyR, ypk_4079; znuB, ypk_2142; znuC, ypk_2141; fur, ypk_2991; katG, ypk_3388; sod(Fe/Mn), ypk_1863; sod(Cu/Zn), ypk_3445.
|
10.1371/journal.pntd.0004080 | Household Socioeconomic and Demographic Correlates of Cryptosporidium Seropositivity in the United States | Cryptosporidium are parasitic protozoa that infect humans, domestic animals, and wildlife globally. In the United States, cryptosporidiosis occurs in an estimated 750,000 persons annually, and is primarily caused by either of the Cryptosporidium parvum genotypes 1 and 2, exposure to which occurs through ingestion of food or water contaminated with oocytes shed from infected hosts. Although most cryptosporidiosis cases are caused by genotype 1 and are of human origin, the zoonotic sources of genotype 2, such as livestock, are increasingly recognized as important for understanding human disease patterns. Social inequality could mediate patterns of human exposure and infection by placing individuals in environments where food or water contamination and livestock contact is high or through reducing the availability of educational and sanitary resources required to avoid exposure.
We here analyzed data from the National Health and Nutritional Examination Survey (NHANES) between 1999 and 2000, and related seropositivity to Cryptosporidium parvum to correlates of social inequality at the household and individual scale. After accounting for the complex sampling design of NHANES and confounding by individual demographics and household conditions, we found impaired household food adequacy was associated with greater odds of Cryptosporidium seropositivity. Additionally, we identified individuals of non-white race and ethnicity and those born outside the United States as having significantly greater risk than white, domestic-born counterparts. Furthermore, we provide suggestive evidence for direct effects of family wealth on Cryptosporidium seropositivity, in that persons from low-income households and from families close to the poverty threshold had elevated odds of seropositivity relative to those in high-income families and in households far above the poverty line.
These results refute assertions that cryptosporidiosis in the United States is independent of social marginalization and poverty, and carry implications for targeted public health interventions for Cryptosporidium infection in resource-poor groups. Future longitudinal and multilevel studies are necessary to elucidate the complex interactions between ecological factors, social inequality, and Cryptosporidium dynamics.
| We examined if and how social inequality in the United States influences seropositivity to Cryptosporidium parvum. By using nationwide data on parasite seropositivity, demographics, and household metrics of socioeconomic status provided through the National Health and Nutritional Examination Survey, we quantified how measures of social inequality affect the odds of parasite infection. After adjusting for the complex sampling design of NHANES and potential confounding by individual demographics and household conditions, we found household food inadequacy was associated with greater parasite seropositivity. Additionally, we found that individuals of non-white races and ethnicities and those born outside of the United States have significantly greater odds of seropositivity than white, domestic-born counterparts. Furthermore, our study suggests evidence for direct effects of family wealth on cryptosporidiosis risk, in that persons in low-income households have elevated odds of parasite seropositivity relative to those in high-income families. These results refute the claim that cryptosporidiosis in the United States in independent of poverty and social marginalization and carry implications for targeted public health interventions for this parasitic infection in resource-poor groups.
| Cryptosporidium are parasitic protozoa that infect humans, domestic animals, and wildlife globally [1]. In humans, cryptosporidiosis is a major cause of global diarrheal illness, and in the United States an estimated 750000 cases occur annually [2]. Although direct mortality from Cryptosporidium infection is rare and often limited to immunocompromised individuals, cryptosporidiosis can cause significant morbidity that in turn can result in high healthcare expenses and losses to productivity [3–5]. Human cryptosporidiosis is primarily caused by either of the Cryptosporidium parvum genotypes 1 and 2, also known as Cryptosporidium hominis and Cryptosporidium parvum [6,7]. Human exposure to Cryptosporidium parvum occurs primarily through ingestion of food or water contaminated with oocytes shed from infected hosts [8,9]. While genotype 1 is limited to human transmission cycles [10], transmission of genotype 2 is based in livestock reservoir hosts such as cattle [11]. Although the majority of cryptosporidiosis cases are caused by genotype 1, the zoonotic capacity of genotype 2 is increasingly recognized as important for understanding human disease patterns [12–14]. For example, cryptosporidiosis cases in the United Kingdom are highest in agricultural areas that utilize cattle manure [15], and water-borne outbreaks in Ireland and the United States have been traced back to cattle [16,17]. Furthermore, exposure directly from livestock has also been observed, particularly among persons working closely with cattle such as children and farmworkers [18,19].
Despite potential linkages between livestock reservoir sources and cryptosporidiosis, we know less about how social inequality may mediate patterns of human exposure and infection. This is unfortunate, as Cryptosporidium is now included in the WHO Neglected Disease Initiative [20], and epidemiological evidence suggests low socioeconomic conditions may amplify human risk. A cross-sectional study in Venezuela found individuals residing in poor urban sectors and in thatched roof–style housing had higher Cryptosporidium prevalence [21]. In a similar setting in Brazil, household water and nearby domestic animals tested positive for Cryptosporidium oocysts [22]. Such work suggests living in physically impaired environments can increase exposure to contaminated water or animals harboring infection [23]. An impaired social environment could also influence patterns of human exposure, as individuals within these environments may lack resources necessary for proper sanitation or educational avoidance of transmission routes. For example, women in Kenyan agricultural communities had greater exposure to cattle and contaminated food and water owing to power hierarchies within households [24].
A major limitation of past work on social determinants of cryptosporidiosis is a focus on either the individual or household level at small spatial scales, restricting understanding interactions between scales and broader inference [25]. To simultaneously address individual- and household-scale socioeconomic drivers of cryptosporidiosis risk, we utilized the National Health and Nutrition Examination Survey (NHANES) from the United States to ask how poor physical and social conditions affect the odds of seropositivity to Cryptosporidium parvum. Several reviews of neglected infections in the United States have noted that cryptosporidiosis is without significant links to poverty or social marginalization [26,27], and socioeconomic factors remain absent in syntheses of risk factors for this disease in the United States [28]. Yet a prior analysis of NHANES found that Hispanics, African Americans, and women all have greater odds of Cryptosporidium seropositivity [29]. Incorporation of household-scale socioeconomics may therefore improve our understanding of how an impaired physical or social environment contextualizes these individual-level relationships. Furthermore, reorienting our focus on cryptosporidiosis towards socioeconomic determinants could offer tangible opportunities for public health interventions and environmental management.
Our analysis used cross-sectional data from NHANES, a series of large nationally representative surveys conducted by the National Center for Health Statistics (NCHS) based on a stratified, multistage, probability cluster design. Data are collected through household interviews, standardized physical examinations, and collection of biological samples at mobile examination centers. A nationally representative sample is selected annually, but data are released in two-year cycles to protect confidentiality and increase statistical reliability. All data were obtained from NHANES between 1999 and 2000, the only two years for which Cryptosporidium serological testing occurred.
To ensure adequate sample size, NHANES 1999–2000 oversampled low-income persons, adolescents 12–19 years of age, persons ≥ 60 years of age, non-Hispanic blacks, and Mexican Americans. Data were weighted to represent the total civilian non-institutionalized U.S. household population and to account for oversampling and nonresponse to the household interview and physical examination. The weights were further ratio-adjusted by age, sex, and race and ethnicity to the U.S. population control estimates from the Current Population Survey adjusted for undercounts. Of the 5663 study participants aged 6 to 49 and who underwent physical examination, 4359 individuals had serum samples available for evaluation and had data available for relevant socioeconomic and demographic covariates for this study. Participants with hemophilia or recipients of chemotherapy within four weeks were excluded.
NHANES is reviewed and approved annually by the NCHS institutional review board, and informed written consent was obtained from all participants or their parents or legal guardians. All individual records were anonymized through unique respondent sequence numbers within NHANES.
Infection with Cryptosporidium parvum is accompanied by the production of parasite-specific antibody (Ig) of all major classes [30,31]. NHANES used an experimental enzyme-linked immunosorbent assay (ELISA) to measure IgG antibodies to two surface antigens to Cryptosporidium parvum, 17kDA and 27kDA [32,33]. In experimental human studies, IgG reactivity to these antigens peaks within 4 to 6 weeks [34]. The IgG response to 17kDA declines to near background levels by 4 to 6 months, whereas the same antibody response to 27kDA can remain elevated for at least 6 to 12 months [35,36]. Evidence from animal and human studies suggests that this antibody response requires inoculation with Cryptosporidium parvum oocysts and that seropositivity to both antigens develops after either asymptomatic or symptomatic infection [34,37]. Hence the IgG response to both antigen groups likely reflects recent or current Cryptosporidium parvum infection and not merely exposure [36,38].
To determine seropositivity to Cryptosporidium parvum, serum samples were tested for reactivity to both antigen groups through the ELISA methods detailed by NHANES and [32,33]. Briefly, sample absorbance was measured using a Molecular Devices UVmax kinetic microplate reader, and IgG levels were assigned a unit value based on the eight-point positive control standard curve with a four-parameter curve fit. The 1:50 dilution of the positive control was assigned a value of 6400 units; unknown samples with absorbance values above the standard curve were diluted further and reassayed. Cutoff values to determine seropositivity are not reported within NHANES; however, prior studies have used cutoffs for seropositivity to the 17kDA and 27kDA antigens as a sample absorbance greater than 86 units, exceeding the mean plus three standard deviations of the negative control, or exceeding 10% of the positive control [38–40].
NHANES reports seropositivity separately to the 17kDA and 27kDA antigen groups as binary outcomes. Because of our interest in the social determinants of Cryptosporidium seropositivity, which likely remain constant through the duration of both IgG antibody responses, we here report seropositivity as a positive IgG response to both 17kDA and 27kDA antigen groups. Following experimental studies, a seropositive response to both of these antigens represents a likely recent or current infection with either genotype 1 or genotype 2 of Cryptosporidium parvum [36,38]; however, a seropositive result does not distinguish between the distinct human or zoonotic sources of oocysts nor whether or not an individual is currently infected.
We examined three household-level indicators of social inequality available through NHANES: food adequacy, annual income, and the poverty income ratio (PIR). Food adequacy was defined as households reporting “enough and the kinds of food wanted,” “enough but not always the kinds of food wanted”, and “sometimes/often not enough to eat.” This variable was recoded as “enough,” “some,” and “not enough” to eat and could serve as a proxy for socioeconomic status, in which resource-poor households are unable to access adequate quantities and qualities of food [41,42]. To consider direct effects of financial resources on Cryptosporidium seropositivity, annual income was defined as total combined family income and was divided evenly into categories of less than $25,000, between $25,000 and $45,000, and greater than $45,000. The PIR was calculated within NHANES to provide a relative measure of income relative to poverty thresholds. Annual family income was divided by the poverty guidelines specific to family size and the appropriate year and state. A PIR less than one indicates a household below the poverty threshold, whereas a ratio of one or greater indicates income above that poverty threshold. We reclassified the PIR into even categories of households below the poverty threshold (PIR < 1), households one to three times above the poverty threshold (PIR 1–3), and households with income more than three times above the poverty threshold (PIR 3+).
In addition to these primary household socioeconomic variables, we also considered confounding by the source and treatment status of household water, as Cryptosporidium parvum is predominantly transmitted through water-borne pathways and untreated water may also be symptomatic of low socioeconomic status. The source of household water was defined as a private or public water company, a private or public well, or another source. Water treatment was defined as whether or not the following treatment devices were used within a household: pitcher water filter, ceramic or charcoal filter, water softener, aerator, or reverse osmosis. Our analysis also considered the size of a household, defined as the number of rooms per home, to account for larger households potentially stemming from greater wealth.
We also considered confounding by demographic covariates of individual race and ethnicity, age, gender, country of birth, and education [29]. Age was defined in one-year intervals, and race and ethnicity were defined by self-report and categorized as non-Hispanic white, non-Hispanic black, Mexican American, and other. Country of birth was categorized as the United States, Mexico, or other. Education was defined by the highest grade of school completed, which we reclassified into a binary variable describing whether or not individuals completed high school. Lastly, associations between socioeconomic status and Cryptosporidium parvum could be confounded by individual health status, as immunocompromised persons are more susceptible to infection [1,43]. Lymphocytes play key roles in the immune defense against Cryptosporidium, and in particular T helper lymphocytes (CD4+ cells) are required for parasite clearance [31,44]. We therefore considered the number of thousand lymphocytes per microliter of blood in our analyses, derived from the total count of leukocytes times the differential count of percent lymphocytes. Although NHANES has directly quantified CD4+ counts, these data were only available for human immunodeficiency virus–positive individuals, which represent a very small subset of the sample tested for Cryptosporidium seropositivity (n = 34 records, < 1% of dataset).
All statistical procedures were conducted to account for the design of multistage stratified, cluster-sampled, unequally weighted survey samples such as NHANES using the package survey in R [45,46]. All seroprevalence estimates were weighted to represent the total U.S. population and to account for over-sampling and nonresponse to interviews and physical examinations [47]. We first performed univariate analyses of primary exposures and potential confounders by fitting survey-weighted generalized linear models with the binomial outcome as the serological response to the 17kDA and 27kDA antigens. These logistic regressions were fit through pseudolikelihood and used inverse-probability weighting and design-based standard errors [48,49]. Standard error estimates were calculated using the Taylor series linearization method to account for the complex sampling design [50], and we used a Wald test to test if all coefficients associated with each covariate differed from zero to examine overall variable significance [51,52]. To adjust for the NHANES survey design, the degrees of freedom for Wald tests were calculated as the number of primary sampling units (n = 27) minus the number of strata (n = 13). This method is recommended for retaining power when considering individual-level covariates within a survey-adjusted analysis [49,53]. Due to this limited degrees of freedom imposed by the sampling design (df = 14), we only included covariates with p < 0.20 from a Satterthwaite-adjusted F statistic as potential confounders to avoid overfitting final multivariable models (shown in Table 1).
We next constructed three survey-weighted multiple logistic regression models. These models incorporated household socioeconomic variables separately owing to strong associations between each (S1 Table). For each household socioeconomic variable (food adequacy, annual income, PIR), we included confounder variables within each model to then test associations between impaired environments and Cryptosporidium seropositivity. We again used Wald tests as the omnibus test of covariate significance within each model, and those variables with p ≤ 0.05 from a Satterthwaite-adjusted F statistic were considered significant. We also tested for differences between groups within each exposure variable after adjusting for the potentially inflated false-discovery rate associated with multiple comparisons using the Benjamini and Hochberg correction and multcomp package in R [54,55]. Crude and adjusted odds ratios and 95% confidence intervals were reported for all variables in the final survey-weighted models. Odds ratios and confidence intervals from the survey-weighted logistic regression models were also calculated using the adjusted degrees of freedom described above. Owing to the cross-sectional nature of NHANES, neither cause–effect relationships nor distinction between recent or current Cryptosporidium infection can be established, and hence odds ratios should be interpreted accordingly.
Of the 4359 persons tested for seropositivity to the Cryptosporidium parvum 17kDA and 27kDA antigen groups in our NHANES sample, 925 persons were IgG positive. This corresponds to a weighted seroprevalence for individuals aged 6 and 49 years of 21.2% (95% CI = 18.5–23.9%). Seroprevalence estimates specific to household food adequacy, family income, PIR, water source and treatment, household size, age, gender, race and ethnicity, country of birth, education, and immunocompetence (lymphocyte count) are shown in Table 1. Seropositivity was influenced by all household socioeconomic variables in our univariate analyses (all F2,14 > 2.7, all p ≤ 0.10), with living in low-income households, close to the poverty line, and with food inadequacy associated with increased odds of seropositivity. Demographics were also strong correlates of seropositivity, with odds increasing with age (F1,14 = 188.96, p < 0.001); with seroprevalence higher among non-Hispanic blacks, Mexican Americans, and other racial and ethnic groups than non-Hispanic whites (F3,14 = 81.29, p < 0.001); and with seroprevalence higher among persons born outside the United States compared to those born within the country (F2,14 = 52.53, p < 0.001). Odds of seropositivity differed little between men and women (F1,14 = 2.2, p = 0.162) and were not influenced by individual lymphocyte count (F1,14 = 0.86, p = 0.371). Lastly, we found significant associations between serostatus and home water treatment (F1,14 = 4.91, p = 0.045) but not with the source of household water (F2,14 = 1.42, p = 0.28); household size was also a weak negative predictor of seroprevalence (F1,14 = 2.58, p = 0.132).
For multivariable models of household socioeconomic conditions and Cryptosporidium seropositivity, we therefore included individual age, reported race and ethnicity, country of birth, education, gender, household size, and household water treatment status as confounders. While univariate tests found no association between individual immunocompetence and Cryptosporidium seropositivity, we also included the total lymphocyte count in our multivariable models to adjust for immunosuppressed individuals being more susceptible to the parasite [31,44]. Additionally, although the effect of household socioeconomic status on Cryptosporidium seropositivity could depend on individual age and warrant age-stratified analyses, we found no support for an interaction between age and all three socioeconomic conditions (S2 Table, S1 Fig) and thus retained age as a separate fixed effect in all models.
Our survey-weighted models showed that household socioeconomic conditions were significant correlates of Cryptosporidium seropositivity after adjusting for individual age, race and ethnicity, country of birth, education, gender, immunocompetence, and household size and water treatment status (Table 2). We first found that household food inadequacy was associated with elevated odds of Cryptosporidium seropositivity (F2,14 = 4.06, p = 0.04; Fig 1A). After adjustment, the odds of Cryptosporidium seropositivity were 1.4 times higher for persons in households with some food inadequacy (OR = 1.4, p = 0.04) compared to those in food-adequate households. This trend appeared to be as strong for persons living in households with high food inadequacy, but this elevated odds of seropositivity compared to persons living in food-adequate households was not significant (OR = 1.31, p = 0.65). Likewise, we observed a negative trend between household annual income and seropositivity, although the overall association was not significant (F2,14 = 3.22, p = 0.07; Fig 1B). Compared to individuals in homes with an annual income of <$25,000 per annum, those in households with an income between $25,000 and $45,000 appeared to have lower risk, although this effect was not significant (OR = 0.78, p = 0.21). Yet relative to the lowest income bracket, individuals in households earning greater than $45,000 had 39% lower odds of Cryptosporidium seropositivity (OR = 0.61, p = 0.03). After adjustment for confounders, the PIR was the strongest household socioeconomic correlate of Cryptosporidium seropositivity and showed a non-linear relationship with seroprevalence (F2,14 = 8.46, p < 0.01). Relative to individuals in families living below the poverty threshold, those in households with an income one-to-three times above the poverty line had slightly elevated odds of seropositivity (OR = 1.26, p = 0.13), yet individuals in families with an income more than three times above than the poverty line had a suggested 25% reduction in risk (OR = 0.75, p = 0.14); however, neither of these differences was statistically significant. When comparing individuals in families with an annual income one-to-three times above the poverty threshold to those in families earning greater than three times the poverty line, however, this increase in income was protective against Cryptosporidium seropositivity (OR = 0.60, p < 0.001; Fig 1C).
Our analyses also found individual demographics to be strong correlates of Cryptosporidium seropositivity. In our most conservative survey-weighted logistic model containing annual family income, we found that after adjustment for other covariates, risk of seropositivity increased by 6% per year increments in age (OR = 1.06, p < 0.001) and was at least 1.76 times higher for non-Hispanic blacks and Hispanics compared to non-Hispanic whites (F3,14 = 11.89, p < 0.001; Table 2). The odds of Cryptosporidium seropositivity were also at least 2.27 times higher for persons born outside the United States (F2,14 = 16.25, p < 0.001; Fig 1D). After adjusting for other covariates in this conservative model, we found no significant associations between Cryptosporidium seropositivity and household size, water treatment status, individual education, gender, or immunocompetence (Table 2).
Our analysis of NHANES 1999–2000 sought to examine if and how social inequality, at both household and individual scales, influenced the odds of Cryptosporidium parvum seropositivity. In low- and middle-income countries, Cryptosporidium infection associates explicitly with poverty and social marginalization [56,57]. However, regions of high socioeconomic status report greater risk in some high-income countries [15]. Such findings have perhaps prompted the claim that cryptosporidiosis is without significant links to social inequality in the United States and resulted in socioeconomic status being absent in discussions of risk factors for this infection [26–28]. Our results suggest that, contrary to these assertions, cryptosporidiosis risk in the United States is highest for individuals living in households with poor food adequacy; in families where income is low and close to the poverty threshold; and for older, non-white, and foreign-born persons.
First, we found food inadequacy was a strong household predictor of serological status, with persons in homes reporting some food inadequacy (enough to eat but not always of the foods desired) showing elevated odds of Cryptosporidium seropositivity. This result was robust to adjustment for household size, water treatment, age, race and ethnicity, education, gender, country of origin, and immunocompetence, yet after adjustment there was also no greater risk of seropositivity in homes with poor food adequacy compared to food-adequate households. As the direction of the food inadequacy effect suggested dose response, however, our finding of no difference in risk could be driven by the small sample size of food-inadequate households. More broadly, the significant association between food adequacy and cryptosporidiosis could suggest several pathways through which social inequality influences Cryptosporidium infection. Food inadequacy could first function as a proxy for household socioeconomic status, as food scarcity is typically driven by a broader lack of financial resources [41,42]. An effect of household food inadequacy through this poverty pathway could suggest reduced access to educational or sanitation resources that allow individuals to avoid Cryptosporidium transmission pathways such as contaminated water or contact with livestock reservoir hosts. For example, low levels of education and access to the media were both associated with poor hand washing practices and hence greater parasite exposure risk in Kenya [58].
Our multivariable models containing either annual family income or the household PIR both are suggestive of this poverty pathway, as a high annual income and living far above the poverty threshold were both protective against Cryptosporidium seropositivity. Yet a significant overall association between household financial resources and seroprevalence was only observed for the PIR, and for both variables we did not find a significant dose-response relationship after adjustment. Specifically, our analysis showed no difference in the odds of Cryptosporidium seropositivity between individuals in households living below the poverty line and those in households at any degree above the poverty threshold. Instead, we only found a difference in the odds of seropositivity between individuals in households close to the poverty threshold (1–3 times) and those in households with income greater than three times above the poverty line. These mixed findings for a direct effect of poverty on cryptosporidiosis could therefore suggest more immediate links between household food inadequacy and risk. Food inadequacy could first increase susceptibility to parasite infection through reductions to host nutritional status and therefore immunocompetence [59,60]. Food inadequacy could also push households towards accessing food products from avenues where food safety and sanitation may not be regulated, such as at open street markets [61].
This latter exposure-based pathway between household food inadequacy and cryptosporidiosis risk seems particularly plausible, as small vegetable markets have been identified as a source of Cryptosporidium in low-income regions [62]. Additionally, as the lymphocyte-based measure of immunocompetence had little effect on predicting seropositivity in univariate analyses and was incorporated into all multivariable models, the observed association in our multivariable model between household food inadequacy and cryptosporidiosis could be driven more by food-borne exposure rather than health status. Because household water quality was also included in our multivariable model in the form of water treatment, this potential exposure-based effect of household food inadequacy on Cryptosporidium seropositivity could also be driven more by consumption of contaminated produce rather than contaminated water.
Along with this household socioeconomic correlate of Cryptosporidium seropositivity, our analysis also identified older individuals, those of non-white ethnicities and races, and individuals born outside the United States as at greater risk of cryptosporidiosis. Increased odds of seropositivity with age are to be expected, given that the risk of ever being infected by Cryptosporidium would accumulate over time. For the other individual risk factors identified, another analysis of NHANES similarly found non-white Hispanics and blacks to have higher seroprevalence to the Cryptosporidium antigen groups used here [29]. Our finding of higher seropositivity in immigrants across the United States is also consistent with smaller-scale findings that children living along the Texas–Mexico border have higher seroprevalence than non-border children and that immigrants from Mexico have higher cryptosporidiosis prevalence than American-born counterparts in Los Angeles [63,64].
These results again highlight the potential links between social marginalization and Cryptosporidium seropositivity, as non-white, immigrant populations are more prone to experience unemployment, live in economically poor neighborhoods, and have reduced access to resources [65,66]. Such groups may therefore be more likely to lack access to sanitation or educational resources for avoiding parasite exposure and be prone to live in physically impaired or remote environments where access to clean food is limited or where contact with domestic animals is frequent. For example, many low-income immigrants in the United States find their employment in agriculture [67,68], which likely amplifies exposure to Cryptosporidium oocysts through contact with soil and water contained with livestock excrement [69]. The recreational use of open natural water sources in such regions may also elevate the odds of parasite exposure. These results in turn suggest that public health interventions for cryptosporidiosis in the United States could focus on improving awareness of Cryptosporidium exposure routes in such marginalized and resource-poor groups. Further research could also monitor the potential for water and food contamination in regions where high-risk groups reside and test if structural aspects of the physical environment amplify Cryptosporidium seropositivity [21].
Future research on the social epidemiology of cryptosporidiosis in the United States could also utilize multilevel analyses to tease apart the relative contribution of individual and household socioeconomic determinants of seropositivity while accounting for potential neighborhood effects [70,71]. Specifically, we did not find clear evidence of an overall dose–response relationship between individual seropositivity to Cryptosporidium and household financial resources (annual income or PIR), despite suggestive trends. Although this could be due to a small sample size for select groups of the study population, the structuring of NHANES could also have limited identifying a strong income effect. Associated geographic data (e.g., zip codes, census block) for NHANES are supplied as restricted access, and thus we only included household and individual correlates in these analyses. Yet a stronger effect of income could manifest spatially at the neighborhood scale, where low income level could cluster food-inadequate households and the demographic groups found to be at risk in our multivariable models [72,73].
An additional needed area of work on the social epidemiology of cryptosporidiosis in the United States is the pursuit of longitudinal rather than cross-sectional approaches. Owing to the cross-sectional design of NHANES within the two-year study period tested for Cryptosporidium seropositivity, we were unable to distinguish between IgG-positive individuals with Cryptosporidium infection and those that had only been recently infected and recovered. Longitudinal sampling across a socioeconomic gradient could help tease apart in which cases seropositivity is due to current or recent infections and how this varies by poverty and social marginalization, particularly if investigators measure IgM antibodies alongside IgG antibodies or changes in titers over time [74]. This could be particularly useful to account the positive association between age and seropositivity observed in this study and other analyses of NHANES [29], as IgG titers can increase with age owing to repeated Cryptosporidium infections. Together such studies would allow researchers to assess new Cryptosporidium infections in relation to acute exposures and relate direct infection to impaired social and physical conditions.
Likewise, our analysis of NHANES was limited by serological testing with the 17kDA and 27kDA antigen groups, which can identify recent or current infection with Cryptosporidium parvum but cannot distinguish between genotype 1 and genotype 2 of the parasite (C. hominis and C. parvum, respectively; [6]. Rather, genotyping methods on human fecal samples could elucidate whether observed cryptosporidiosis or Cryptosporidium seropositivity is due to infection with the human-origin genotype 1 or zoonotic genotype 2 [7,75]. Differentiation of human versus animal sources of infection in combination with analyses of socioeconomic risk factors could further improve our understanding of how impaired physical and social conditions interact with Cryptosporidium transmission. For example, a location-based study in the UK found that Cryptosporidium hominis cases were more frequent in urban areas of high socioeconomic status, whereas Cryptosporidium parvum cases (zoonotic genotype 2) were more common in rural areas where more oocysts were detected in agricultural soil, presumably from cattle [15].
Within the United States, genotyping methods could be particularly useful to elucidate how the individual- and household-scale correlates of poverty and social marginalization identified in our analyses interact with the abundance of livestock reservoir hosts and hence human infection with zoonotic Cryptosporidium parvum. Specifically, the density of livestock reservoir hosts such as cattle could be related to regional socioeconomic status [76,77], in turn driving greater exposure of marginalized groups to Cryptosporidium parvum genotype 2 in the United States. One analysis of sporadic cryptosporidiosis cases in Scotland accordingly found Cryptosporidium prevalence to be highest in rural regions with high livestock density [78], suggesting a neighborhood poverty influence on seropositivity. Hence ecological and multilevel analyses could test for an interactive influence of livestock density and social marginalization variables identified here (e.g., food adequacy, immigration status, race and ethnicity) on cryptosporidiosis while accounting for neighborhood income.
Our analyses here demonstrate clear associations between social marginalization, poverty, and cryptosporidiosis in the United States, thereby carrying important implications for targeted public health interventions for this infection in resource-poor groups. Alongside direct effects of Cryptosporidium infection on mortality in immunocompromised individuals, morbidity from cryptosporidiosis can range from subtle to severe effects quality of life that can impose serious restrictions on economic wellbeing [3,4]. As cryptosporidiosis is estimated to occur in 750000 persons across the United States annually, these effects can scale up to over $100 million per year in healthcare costs and losses to productivity [2,5]. Therefore, understanding interactions between socioeconomic and environmental conditions in combination with longitudinal and genotyping approaches will be key to guiding prevention and intervention strategies to cryptosporidiosis within the United States. Analyses in this spirit will more broadly help address the complex relationships between ecological factors, social inequality, and infectious disease risk [79,80].
|
10.1371/journal.pgen.1002987 | A Mimicking-of-DNA-Methylation-Patterns Pipeline for Overcoming the Restriction Barrier of Bacteria | Genetic transformation of bacteria harboring multiple Restriction-Modification (R-M) systems is often difficult using conventional methods. Here, we describe a mimicking-of-DNA-methylation-patterns (MoDMP) pipeline to address this problem in three difficult-to-transform bacterial strains. Twenty-four putative DNA methyltransferases (MTases) from these difficult-to-transform strains were cloned and expressed in an Escherichia coli strain lacking all of the known R-M systems and orphan MTases. Thirteen of these MTases exhibited DNA modification activity in Southwestern dot blot or Liquid Chromatography–Mass Spectrometry (LC–MS) assays. The active MTase genes were assembled into three operons using the Saccharomyces cerevisiae DNA assembler and were co-expressed in the E. coli strain lacking known R-M systems and orphan MTases. Thereafter, results from the dot blot and restriction enzyme digestion assays indicated that the DNA methylation patterns of the difficult-to-transform strains are mimicked in these E. coli hosts. The transformation of the Gram-positive Bacillus amyloliquefaciens TA208 and B. cereus ATCC 10987 strains with the shuttle plasmids prepared from MoDMP hosts showed increased efficiencies (up to four orders of magnitude) compared to those using the plasmids prepared from the E. coli strain lacking known R-M systems and orphan MTases or its parental strain. Additionally, the gene coding for uracil phosphoribosyltransferase (upp) was directly inactivated using non-replicative plasmids prepared from the MoDMP host in B. amyloliquefaciens TA208. Moreover, the Gram-negative chemoautotrophic Nitrobacter hamburgensis strain X14 was transformed and expressed Green Fluorescent Protein (GFP). Finally, the sequence specificities of active MTases were identified by restriction enzyme digestion, making the MoDMP system potentially useful for other strains. The effectiveness of the MoDMP pipeline in different bacterial groups suggests a universal potential. This pipeline could facilitate the functional genomics of the strains that are difficult to transform.
| Approximately 95% of the genome-sequenced bacteria harbor Restriction-Modification (R-M) systems. R-M systems usually occur in pairs, i.e., DNA methyltransferases (MTases) and restriction endonucleases (REases). REases can degrade invading DNA to protect the cell from infection by phages. This protecting machinery has also become the barrier for experimental genetic manipulation, because the newly introduced DNA would be degraded by the REases of the transformed bacteria. In this study we have developed a pipeline to protect DNA by methylation from cleavage by host REases. Multiple DNA MTases were cloned from three difficult-to-transform bacterial strains and co-expressed in an E. coli strain lacking all of the known endogenous R-M systems and orphan MTases. Thus, the DNA methylation patterns of these strains have become similar to that of the difficult-to-transform strains. Ultimately, the DNA prepared from these E. coli strains can overcome the R-M barrier of the bacterial strains that are difficult to transform and achieve genetic manipulation. The effectiveness of this pipeline in different bacterial groups suggests a universal potential. This pipeline could facilitate functional genomics of bacterial strains that are difficult to transform.
| Experimental genetic manipulation has been an essential tool for gaining insight into the significance of bacterial metabolism, physiology and pathogenesis [1], [2] and has been instrumental in developing microbial biotechnology [3]. To date, only a limited proportion of the laboratory culturable bacterial species are amenable to genetic manipulation. Among these manipulation-friendly species, many strains are refractory to transformation by exogenous DNA. The currently available laboratory model species satisfied the research need for genetic uniformity, but the handicap in genetic manipulation is a challenge when exploring the unique traits of these non-model species/strains [4].
Restriction-Modification (R-M) systems are composed of restriction enzymes (REases) and DNA methyltransferases (MTases). These systems are widespread in both bacteria and archaea. Approximately 95% of the genome-sequenced bacteria harbor R-M systems, and 33% carry more than four REases [5]. R-M systems have been classified into four groups depending on their subunit composition, cleavage sites, sequence specificity and cofactor requirement [6]. Type I, II and III REases cleave unmethylated DNA at specific sites, and Type IV cut methylated DNA with foreign patterns [6]. R-M systems are believed to act as defenses to protect the prokaryotic cells against invading DNA; exogenous DNA with foreign methylation patterns are recognized and rapidly degraded [7]. Inevitably, this defensive machinery hinders the experimental genetic manipulation of many bacteria species. Moreover, genetic modification becomes even more difficult when the targeted bacteria carry multiple R-M systems.
The Nitrobacter hamburgensis X14 strain oxidizes nitrite to conserve energy and is commonly used in nitrification research [8]. Although the strain was isolated more than 100 years ago [9], limited research on this strain has been published due to the lack of genetic manipulation tools. Genomic sequencing has revealed eleven sets of R-M genes in N. hamburgensis X14 [10]. Bacillus cereus ATCC 10987 is a non-lethal strain in the same genetic subgroup as B. anthracis [11]. Although genetic manipulation has been routine in many other B. cereus strains [12], limited research has been performed in B. cereus ATCC 10987 due to its resistance to genetic manipulation. Transformation of B. cereus ATCC 10987 has been performed with DNA prepared from Bacillus subtilis with low efficiency [13], [14], and four REases have recently been characterized [15]. Bacillus amyloliquefaciens TA208 is an industrial guanosine-producing strain [16] and has been reported to be transformed at low efficiencies with plasmids prepared from Escherichia coli [17].
Here, we describe a mimicking-of-DNA-methylation-patterns (MoDMP) pipeline. An E. coli strain lacking all of the known six characterized R-M systems and orphan MTases was generated to prevent unintentional modification of propagated plasmids or cleavage of DNA with foreign methylation patterns. After expressing multiple active MTases from the target bacteria in the E. coli strain lacking known R-M systems and orphan MTases, the DNA methylation patterns of E. coli were altered to reflect the patterns of the target bacteria. Plasmids prepared from these hosts escaped the host REases, and genetic manipulation could be readily achieved. The pipeline was shown to be effective in all of the three aforementioned strains which are difficult to transform using conventional methods. We report the first genetic transformation of Nitrobacter, the improvement of transformation efficiency by exogenous plasmids in B. cereus ATCC 10987 and B. amyloliquefaciens TA208 using the MoDMP pipeline, and direct mutagenesis using non-replicative plasmids in B. amyloliquefaciens TA208. The MoDMP pipeline may be readily adapted to bacteria carrying multiple R-M systems.
To avoid the unintentional activation of the Type IV R-M systems in the target bacteria, plasmids that are to be used for genetic transformation should be prepared from an E. coli host that does not methylate DNA (dam- dcm- hsdRMS-). Moreover, the expression of MTases in E. coli would induce foreign patterns of modification on the E. coli chromosomal DNA; therefore, the REases that restrict methylated DNA (Mrr, McrA and MrcBC) should be inactivated in the MoDMP host. To date, three E. coli strains that do not methylate DNA or restrict DNA with foreign methylation patterns have been described (E. coli DB24 [18], E. coli HST04 from Clontech and E. coli JTU007 [19]). In this study, an E. coli mutant lacking all of the six characterized R-M systems and orphan MTases genes, namely strain EC135, was generated in the E. coli TOP10 background by deleting the dam and dcm genes. A wild-type recA allele was introduced into the strain prior to dam inactivation to counteract the inviability of the dam recA double mutant strain [20]. The construction of the E. coli EC135 strain is explained in detail in the Supporting Information Methods (Text S1), and validation of the strain is described in the Supporting Information Results (Text S1 and Figure S1).
MTases modify DNA by adding a methyl group to the individual bases, thereby preventing DNA cleavage by the corresponding REases. In the MoDMP procedure, MTases from the difficult-to-transform bacterial strains were used to protect DNA from being degraded by the REases. The genes of 24 putative MTases including two belonging to Type I R-M systems, 19 belonging to Type II R-M systems, two belonging to Type III R-M systems, and one orphan MTase were cloned from the genomes of the three difficult-to-transform strains into the pBAD43 vector (Table 1).
To date, three types of methyl-transferring activity have been described for bacterial DNA MTases, namely N6-methyladenine (m6A), N4-methylcytosine (m4C) and 5-methylcytosine (m5C) modifications. Dot blot assays were conducted to detect the modified bases in the total genomic DNA of the E. coli EC135 strains expressing individual MTase using antibodies against m6A, m4C and m5C. In total, 13 of the putative MTase genes exhibited methyl transfer activity to DNA (Figure 1), and the bases they modified are summarized in Table 1. The spots of the dot blots were also scanned and quantified, and the relative intensity of each spot is shown in Figure S2.
The hybridization signals of BCE_0392, Nham_0582, Nham_0803 and Nham_3225 were weak in the dot blot experiments. To confirm their activity, the total DNA of the E. coli EC135 strains expressing these four MTases individually were digested to deoxynucleosides, and Liquid Chromatography-Mass Spectrometry (LC-MS) assays were performed to detect N6-methyl-2′-deoxyadenosine (m6dA) in the DNA. In High Performance Liquid Chromatography-Quadrupole Time-of-Flight/Mass Spectrometry (HPLC-QTOF/MS) analysis, m6dA (m/z 266.12) was readily detected in the digested DNA of the BCE_0392-, Nham_0582- and Nham_3225-expressing strains in the MS spectrum at the corresponding retention time of standard m6dA, validating that BCE_0392, Nham_0582, and Nham_3225 displayed DNA m6A modification activity in E. coli (Figure 2). Xu et al. has reported that MTase activity was not detected for in vivo translated BCE_0392 protein using [H3]AdoMet and phage λ DNA or pXbaI plasmid DNA as substrate [15]. This might either be caused by the mis-folding of in vivo translated BCE_0392 protein or by the absence of BCE_0392 recognition sites from the substrate DNA they used.
Although m6dA was not detected for Nham_0803 in the HPLC-QTOF/MS analysis (Figure 2), the MTase activity of Nham_0803 could not be ruled out, since the DNA of the Nham_0803-expressing strain displayed slight but noticeable signal increase compared with the E. coli EC135 strain harboring empty vector (Figure S2). Other more sensitive and targeted MS approaches could be useful in detecting the possible modified nucleoside conferred by Nham_0803 [21].
It is noteworthy that the E. coli strain EC135 expressing the Nham_0569 MTase grows much slower than the control strain or strains expressing other MTases (Nham_0803 and Nham_3225), and the final biomass of strain EC135 carrying Nham_0569 was about 60% of that for the control strains (Figure S3). This growth retardation in E. coli may be attributed to the toxicity of the Nham_0569 MTase. The E. coli EC135 strain lacks methylation-dependent REases activity, which will cleave its own DNA when foreign methylation patterns are detected, leading to cell death; however, the modification of m6A by the Nham_0569 MTase may occur on sequences overlapping with Dam sites in E. coli, which participates in DNA mismatch repair and replication initiation. Consequently, the premature and untimely methylation of DNA may interfere with strain proliferation [22]. Subsequent sequence specificity analysis has revealed that Nham_0569 modified GATC sequences (see below).
To mimic the DNA methylation patterns of the strains that are difficult to transform, we co-expressed the active MTases from each strain. By taking advantage of the high rates of recombination in Saccharomyces cerevisiae, MTase genes, with optimized ribosome binding site (RBS) for expression in E. coli, were inserted into the pWYE724 backbone to form three operons. The diagrams of the pMoDMP plasmids are shown in Figure 3. The insertion of the MTase genes was verified by multiple methods, including PCR analysis of plasmids, restriction digestion (Figure S4) and DNA sequencing. The protocol for DNA assembly in S. cerevisiae is very powerful, and up to eight MTase genes from Nitrosococcus oceani ATCC 19707 could be readily assembled in our lab (Zhang et al., unpublished).
DNA from the E. coli EC135 strain expressing multiple MTases was also tested by dot blot assay, with the DNA of the parent strains as the positive controls (Figure 4). DNA from the co-expression strains exhibited multiple methylation signals, indicating the alteration of the DNA methylation patterns in E. coli. It is worth noting that the m4C and m5C signals in B. amyloliquefaciens strain TA208, the m5C signal in B. cereus strain ATCC 10987 and the m4C signal in N. hamburgensis strain X14 were much weaker when compared with their corresponding MTase over-expressing E. coli strain. This signal weakness could be attributed to the different number of MTase target sequences between the genomic sequences of E. coli and the parent strains or to the fact that the B. amyloliquefaciens TA208 strain is an adenine auxotroph, which limits the availability of S-adenosylmethionine (AdoMet). However, pMK4 plasmid DNA prepared from the TA208 strain is resistant to BamHI digestion, which is a homolog to the restriction subunit of the BAMTA208_16650-BAMTA208_16660 systems (see below). Thus, regulational expression of the R-M systems could also explain the weak blot signals in the parent strains; Hegna et al. has reported that the R-M system is activated when B. cereus is grown in the presence of exogenous DNA [23].
To determine the efficacy of the MoDMP pipeline, various shuttle plasmids carrying divergent replicons and conferring different antibiotic resistance were used to transform B. amyloliquefaciens TA208 and B. cereus ATCC 10987. Prior to transforming Bacillus, the shuttle plasmids were methylated in vivo when transformed into the E. coli EC135 strain harboring the pMoDMP plasmids. B. amyloliquefaciens TA208 and B. cereus ATCC 10987 were transformed by these plasmids, and the transformation efficiencies were calculated.
The B. amyloliquefaciens TA208 strain could not be transformed with plasmids prepared from E. coli TOP10 cells but could be transformed with plasmids from the E. coli EC135 strain with low efficiency; this result indicates that a methylation-dependent Type IV R-M system may exist in B. amyloliquefaciens TA208, although its coding gene was not found during annotation of the genome sequence [16]. Hence it may also be that the plasmids methylated at the Dam and Dcm sites would not be inherited in B. amyloliquefaciens TA208, e.g., methylated replication origin would not be bound by the replication protein. The MoDMP protocol increased the transformation efficiencies of all the plasmids tested in B. amyloliquefaciens TA208. The pMK4 plasmid from MoDMP hosts showed the highest transformation efficiency (3×106 CFU/µg DNA), representing a 104-fold increase compared to that of the plasmids from the E. coli EC135 strain. The MoDMP procedure also enabled two previously untransformable plasmids, pAD123 and pDG148StuI, to be transformed at an efficiency of 1×105 CFU/µg DNA (Figure 5A).
For the B. cereus ATCC 10987 strain, the MoDMP pipeline increased the transformation efficiency of the pMK4 plasmid to 2×107 CFU/µg DNA and increased the transformation efficiency of pMK3 by 103 fold compared to those from strains E. coli TOP10 or EC135 (Figure 5B). The plasmids prepared from E. coli TOP10 and EC135 strains showed similar transformation efficiencies, indicating that the putative Type IV R-M systems (BCE_1016 and BCE_2317) in B. cereus ATCC 10987 may be inactive. These results were the same as those obtained in the B. cereus ATCC 14579 strain, which could be transformed by methylated DNA (DNA from non-dam dcm mutant strains) [24], though some researchers prefer to use unmethylated DNA [25].
The high transformation efficiency achieved with the MoDMP method in both Bacillus strains would allow for the direct inactivation of genes using non-replicative integration plasmids.
To further validate the efficacy of the MoDMP procedure, the gene coding for uracil phosphoribosyltransferase (upp) in B. amyloliquefaciens TA208 was selected for inactivation using non-replicative integration plasmids. The B. amyloliquefaciens TA208 strain was transformed with pWYE748 plasmids that had been through the MoDMP host. The pWYE748 plasmid recombines with chromosome of B. amyloliquefaciens TA208 at the upp locus with a low rate (10−6) because it lacks a replication origin for Bacillus (Figure 6A). BS043 was obtained and PCR and sequencing analyses revealed the successful replacement of the upp gene with the chloramphenicol resistance gene in this strain (Figure 6B). Uracil phosphoribosyltransferase converts 5-fluorouracil (5-FU) to 5-fluoro-UMP, which is ultimately metabolized to the toxic compound 5-fluoro-dUMP capable of inhibiting the activity of thymidylate synthetase. The upp/5-FU module has been widely used in many bacterial species for deletion of genes without introducing antibiotic resistance markers [26]. In contrast with the B. amyloliquefaciens TA208 strain, the BS043 strain could grow on minimal medium (MM) supplemented with 5-FU (Figure 6C). These findings suggest that the MoDMP system elevated transformation efficiencies of exogenous plasmid to enable direct gene inactivation, and the upp gene could be used as a counter-selection marker for the in-frame deletion of genes in B. amyloliquefaciens.
N. hamburgensis X14 harbors 11 putative R-M systems, and successful genetic transformation of this strain has not been reported [27]. In this study, the N. hamburgensis X14 strain was transformed with plasmids carrying the Green Fluorescent Protein (GFP) encoding gene gfpmut3a using the MoDMP procedure. Total genomic DNA was extracted from 10 mL of the transformed bacteria cells. The plasmid was rescued to E. coli TOP10 cells, and subsequent plasmid preparation (Figure S5A) and restriction digestion with SalI and PstI (Figure S5B) verified the existence of pWYE561 in the transformed bacterial cell lines. During the subculture process, the bacterial cell lines were monitored for contamination by microscopy and culturing on LB plates at 30°C, and no contamination was observed. Green fluorescent signals were observed in the cytoplasm of Nitrobacter, thereby revealing the successful transformation of Nitrobacter (Figure 7A).
The culture may contain multiclonal cell lines because the transformants were enriched twice through successive sub-inoculation of the transformation cell mixture in liquid culture (see Materials and Methods for details). Using flow cytometry, the ratio of fluorescent cells was determined to be 50.37% (Figure 7B), demonstrating that 50.37% of the cells were positive transformants. Clonal cell lines could be obtained by streaking the transformant-enriched culture on nylon membranes placed on solid medium and periodically transferred to fresh plates, as described by Sayavedra-Soto et al. in Nitrosomonas europaea [28].
To make the MTase expression vectors more useful, the modification sequences of MTases were determined when expressed individually or co-expressed. As shown in Figure S6A, BAMTA208_6525 protected plasmid from cleavage by BamHI (GGATCC), BglII (AGATCT), and partially from BclI (TGATCA), indicating that BAMTA208_6525 modifies RGATCY and partial TGATCA sequences. BAMTA208_6715 protected pMK4 from cleavage by HaeIII (GGCC), Fnu4HI (GCNGC) and Bsp1286I (GDGCHC), and BAMTA208_19835 and BAMTA208_16660 protect pMK4 from TseI (GCWGC) and BamHI (GGATCC) cleavage, respectively. When co-expressed, the four active MTases from B. amyloliquefaciens TA208 could protect the plasmids from cleavage by all of the REases tested in the individual expression experiments. However, DNA from the B. amyloliquefaciens TA208 strain was only resistant to BamHI cleavage and partially resistant to Fnu4HI and TseI cleavage (Figure S6B), indicating that the expression of BAMTA208_16660 in the native strain was complete, whereas those of BAMTA208_6715 and BAMTA208_19835 were incomplete, and BAMTA208_6525 was not expressed.
For the B. cereus ATCC 10987 strain, BCE_0393 could protect plasmid from cleavage by at least 12 REases, i.e., Fnu4HI (partial), TseI, BbvI (GCAGC), HaeIII, EaeI (YGGCCR), HpaII (CCGG), MspI (CCGG, partial), NlaIV (GGNNCC), BssHII (GCGCGC), HhaI (GCGC, partial), AvaII (GGWCC) and PspGI (CCWGG, partial), and the modification sequences of BCE_0393 were concluded as GCWGC, GGCC, CCGG, GGNNCC, GCGCGC, GGWCC and CCWGG (partial). BCE_0365 protected DNA from cleavage by TseI and BbvI, indicating that it modifies GCWGC sequence, BCE_4605 protect DNA from cleavage by AvaII via modification of GGWCC sequence, and BCE_5606 and BCE_5607 both protect DNA from cleavage by BceAI [ACGGC(N)12/14] (Figure S7A). These results are consistent with the reports of Xu et al. [15], except for that “GGWCC” was added to the modification sequences of BCE_0393 in this study. The multi-specificity nature of the prophage MTase BCE_0393 and its sequence overlapping with other MTases from B. cereus ATCC 10987 indicated that it plays a major role in the MoDMP pipeline of this strain.
The pMK4 plasmids prepared from the E. coli strain expressing BCE_0392 was challenged with various REases which might be sensitive to m6A modification, including AvaII, BamHI, BbvI, BceAI, BglII, BsiEI, Bsp1286I, BspDI, BstNI, BspHI, DpnII, EaeI, EcoRI, Fnu4HI, HincII, HindIII, HpaII, HinfI, NlaIV, PstI, PshAI, PspGI, SalI, ScrFI, SwaI, SpeI, TaqI and TseI, but resistance to cleavage was not observed. Therefore BCE_0392 might modify sequences that are not recognized by these REase, and new techniques like single-molecule DNA sequencing other than restriction analysis using commercialized REases should be useful in identifying the sequence specificity of BCE_0392 [29]. DNA nicking-associated concatenation activity was also detected for BCE_0392 in vivo [15], suggesting that this ParB-Methyltransferase might participate in phage DNA replication or phage packaging, since BCE_0392 was located in a prophage region in the chromosome of the B. cereus ATCC 10987 strain [11].
When co-expressed, BCE_0393, BCE_0365, BCE_4605, BCE_5606 and BCE_5607 protected all of the pMK4 plasmid from AvaII and BceAI digestion, protected most of the pMK4 plasmids from Fnu4HI, TseI, BbvI, HaeIII, EaeI, HpaII and NlaIV cleavage, provided pMK4 partial protection from HhaI digestion and provided pHCMC05 full protection from BssHII cleavage (Figure S7B). However, pMK4 plasmid prepared from the B. cereus ATCC 10987 strain was only resistant to TseI, BbvI, AvaII and BceAI digestion, and partially resistant to Fnu4HI digestion (Figure S7B), indicating that BCE_0393 is not completely expressed in its native host.
For N. hamburgensis X14, Nham_0569 could protect DNA from cleavage by at least 10 REases sensitive to m6A modification, i.e., DpnI (GAmTC), DpnII (GATC), PvuII (CAGCTG), SspI (AATATT), SpeI (ACTAGT), MfeI (CAATTG), NlaIII (CATG), AseI (ATTAAT), HinfI (GANTC) and TfiI (GAWTC), but full protection was not achieved (Figure S8A). Therefore, Nham_0569 might be a multi-specific enzyme harboring at least eight modification sites, i.e., GATC, CAGCTG, AATATT, ACTAGT, CAATTG, CATG, ATTAAT and GANTC, or a new member of the recently characterized non-specific DNA adenine MTase [30]. Nham_3225 protected DNA from HinfI and TfiI cleavage by modifying GANTC sequence (Figure S8A). The pMK4 plasmids prepared from the E. coli EC135 strains expressing Nham_0582 and Nham_0803 were not resistant to the cleavage by DpnII, EcoRI, DraI, PvuII, SspI, HinfI, HindIII, BspHI, BamHI, SpeI, KpnI, SacI or ApaLI, and the specificity of Nham_0582 and Nham_0803 was not identified.
The four active MTases from N. hamburgensis X14 provided DNA partial protection from cleavage by DpnI, DpnII, PvuII, SspI, SpeI, MfeI, NlaIII, AseI, HinfI and TfiI when co-expressed, and the genomic DNA of the N. hamburgensis X14 strain was sensitive to DpnII digestion, partially resistant to DpnI digestion and resistant to SpeI, AseI, HinfI and TfiI digestion (Figure S8B). These results indicated that Nham_0569 was only partially expressed in its native host.
The modification sequences of MTases are summarized to Table 1. The MoDMP hosts and difficult-to-transform bacteria showed similar DNA methylation patterns based on the REase digestion analysis, but the DNA from MoDMP hosts have more modification sites than corresponding difficult-to-transform bacteria. And this was mainly caused by the limited expression of some MTases in their native hosts, especially some prophage-derived MTases, i.e., BAMTA208_6525, BCE_0393 and Nham_0569, which are multi-specific MTases.
Genetic transformation of bacteria harboring multiple R-M systems has been problematic using conventional methods. It has been long recognized that the exogenous MTases over-expressed in E. coli could modify DNA in vivo and protect them from digestion by their cognate REases [31]. Strategies based on this fact have been developed to overcome the restriction barrier of bacteria, including in vitro or in vivo plasmid modification prior to transformation [32], [33], heat inactivation of the REases [34] or gene knock-outs [35]. However, it has been reported that the inactivation of the SauI Type I R-M system is insufficient for Staphylococcus aureus to efficiently accept foreign DNA [36]. In this study, a strategy has been developed to mimic the DNA methylation patterns of the difficult-to-transform bacteria in a modified E. coli strain. To achieve this goal, active MTases from the difficult-to-transform bacteria were co-expressed in an E. coli host lacking all of the characterized R-M systems and orphan MTases.
The protocol for genetic transformation of difficult-to-transform bacteria using a plasmid prepared in a different E. coli host is diagramed in Figure S9. As indicated in strain B. amyloliquefaciens TA208, DNA from the E. coli hosts with Dam and Dcm contains methylated bases in GAmTC and CCmWGG sequences but is not methylated at the recognition sequences of the host Type I–III REases; this DNA would then be recognized by Type I–IV REases in the target bacteria (upper left panel in Figure S9). Plasmids prepared from dam dcm EcoKI mutant E. coli would make the strains transformable at a low efficiency due to the plasmids being able to avoid restriction by the Type IV REases in the target bacterium (lower left panel in Figure S9). Many bacterial species have been reported to restrict DNA containing Dam and Dcm methylation; for example, B. anthracis could be transformed by DNA from an E. coli dam dcm mutant strain but not by DNA from E. coli host strains with the wild type alleles [37], [38]. Additionally, DNA prepared from the E. coli SCS110 strain was more accessible to Corynebacterium glutamicum than DNA from E. coli hosts with Dam and Dcm [39]. However, not all difficult-to-transform bacteria behave like this. Bacteria lacking functional Type IV REases could be transformed by DNA prepared from E. coli hosts with Dam, Dcm or EcoKI, albeit at a low efficiency. Currently, it has been shown that the B. cereus ATCC 10987 strain does not restrict DNA with Dam and Dcm methylation. It has also been reported that the plasmids methylated in E. coli TOP10 cells using the MTases of the target bacteria can allow for the genetic manipulation of Bifidobacterium breve [40]. Ryan et al. showed that the bbe02 and bbq67 loci limited the transformation of Borrelia burgdorferi by shuttle vector DNA prepared from E. coli, irrespective of its Dam, Dcm or EcoKI methylation status [41]. The N. hamburgensis X14 strain used in this study may restrict DNA with Dcm methylation; the plasmid-borne putative Type IV R-M system Nham_4502-Nham_4503 has been annotated in the REBASE database [5], though its activity and specificity remain unclear.
As shown in the upper right panel in Figure S9, expression of exogenous MTases in E. coli would result in methylation of chromosomal DNA, and Mrr, McrA and McrBC would recognize and cleave the DNA with foreign patterns, making the strain inviable or resulting in poor MTase expression [42], [43]. Therefore, an E. coli strain lacking all of the known R-M systems and orphan MTases was generated with MTases expressed. The plasmids prepared from this host could escape the REases that recognize unmethylated DNA or DNA methylated in foreign patterns (lower right panel in Figure S9). The MoDMP concept could greatly improve genetic transformation efficiency.
Recently, four REases from the B. cereus ATCC 10987 strain have been cloned and characterized, namely BceSI, BceSII, BceSIII and BceSIV [15]. Only faint and non-specific hybridization blots were observed using the antibodies against m6A and m5C for the MTase of BceSI (BCE_1018) in this study; these faint blots may be caused by the vagaries of dot blot approaches. It might also be that BCE_1018 modifies the DNA in a way other than methylation, such as hydroxymethylation or glucosyl-hydroxymethylation. BCE_1018 was not included in the downstream MoDMP application. Nevertheless, the highest transformation efficiency was achieved using the plasmids modified by six other MTases and was within the acceptable range for gene knock-out experiments (107 CFU/µg DNA). This efficiency may be caused by the low abundance and the weak REase activity of BceSI in strain B. cereus ATCC 10987, as described by Hegna et al. [23]. BceSI was induced only when the strain B. cereus ATCC 10987 was grown in the presence of exogenous DNA.
Nham_3845 (NhaXI) has been reported to be a fused enzyme harboring both restriction and modification subunits, but the m6A or m4C modification activity was not detected [44]; in this study the MTase activity was not detected either. It might be that Nham_3845 modified DNA in ways other than methylation, which could not be detected using immunoblot assays.
The contribution of individual MTases to genetic transformation was not evaluated in this study because a shuttle plasmid containing all of the MTase recognition sequences cannot be defined. A MTase that does not modify one particular plasmid might be useful when other plasmids are to be used. Therefore, to make a universal system for all plasmids, all of the identified active MTases were employed for MoDMP. The orphan MTase BAMTA208_06715, which lacks a counterpart REase, was also used in the MoDMP pipeline of B. amyloliquefaciens TA208. Orphan MTases, such as CcrM, may participate in methylation-directed DNA mismatch repair [45]. Methylated DNA could potentially escape inspection from the host mismatch repair machinery and eventually exhibit an elevated transformation performance, hence the use of BAMTA208_06715 in the MoDMP pipeline.
The use of DNA mimic protein Ocr (overcome classical restriction) alongside the plasmid (TypeOne Restriction Inhibitor, Epicentre; [46]) which specifically inhibits Type I REase activity, also enhances transformation efficiency in bacterial species [47]. Combination of this method with the MoDMP pipeline could further elevate transformation performance in strains which are difficult to transform. Recently, a novel R-M system has been shown to phosphorothioate DNA, preventing the degradation of the DNA by its REase counterparts [48]. The MoDMP concept may also be adapted to those bacteria restricting unphosphorothioated DNA.
In conclusion, we devised a system in E. coli that mimics the DNA methylation patterns of bacterial strains harboring multiple R-M systems. Eventually, the R-M barrier of three represented bacterial strains were overcome, including Gram positive, Gram negative, chemoheterotrophs and chemoautotrophs. The adaptability of this pipeline to different bacterial groups suggests a universal potential. This protocol is very fast; a MoDMP plasmid can be generated in less than one week using the S. cerevisiae assembler, if the MTase activity assay step is omitted and the putative MTases are cloned and expressed directly. We expect that the pipeline will be applicable to other strains of known genome sequence that are resistant to genetic transformation.
The strains E. coli TOP10 and EC135 were used for the cloning and expression of the MTases. S. cerevisiae DAY414 was used for in vivo assembly of the MTase genes. The plasmid pBAD43 was used for the cloning and expression of individual MTases, and pWYE724 was used for co-expression of multiple MTases. Several E. coli-Bacillus shuttle plasmids were used for MoDMP procedure evaluation purposes in the B. amyloliquefaciens TA208 and B. cereus ATCC 10987 strains. Inactivation of upp in B. amyloliquefaciens TA208 was performed with pWYE748. Expression of the GFP variant gfpmut3a in N. hamburgensis was carried out using pBBR1-MCS5. The strains and plasmids used in this study are listed in Table S1.
Putative MTase encoding genes were retrieved from the REBASE database [5]. Genes were PCR amplified and ligated into pBAD43. Individual genes that encode the methylation and specificity subunits of BCE_0839–BCE_0842 system were joined to operons using Splicing by Overlapping Extension (SOE) PCR. All recombinant plasmids were verified by sequencing before use. The E. coli EC135 strain was transformed with pBAD43 plasmids encoding MTase genes. Single colonies were used to inoculate LB medium and cultured until an OD600 reading of 0.2 was reached, and then arabinose was added to a final concentration of 0.2% to induce MTase expression. Expression was induced overnight at 30°C.
The DNA methylation activity of the putative MTases was analyzed using a southwestern dot blot assay as described previously [18]. Total genomic DNA from the E. coli EC135 strains expressing individual or multiple MTases was prepared using a DNeasy Blood and Tissue Kit (Qiagen). DNA concentrations were determined using a Nanodrop 2000C spectrophotometer (Thermo Scientific). The DNA was then denatured at 100°C for 3 min and immediately cold shocked in an ice-water bath. Samples were spotted onto Protran BA85 nitrocellulose membrane (Whatman) and fixed by UV cross-linking. The membrane was blocked in 5% non-fat milk and incubated with rabbit antisera against DNA containing m6A at a dilution of 1∶10,000 (New England Biolabs), rabbit antisera against m4C at a dilution of 1∶10,000 (New England Biolabs), or a mouse monoclonal antibody against m5C diluted 1∶20,000 (Zymo Research). After washing, the membrane was incubated with secondary goat anti-rabbit or anti-mouse antibodies conjugated with horseradish peroxidase (HRP) (Jackson ImmunoResearch) at a dilution of 1∶10,000. The blots were visualized using the ECL prime Western blotting detection reagent (GE Healthcare), and DNA methylation signals were exposed to Kodak X-Ray film.
For quantification of the hybridization signals, the films were scanned and the gray scale of the spots was quantified using Quantity One (Bio-Rad). After normalization, the values were plotted as bar charts.
To obtain the nucleoside samples of genomic DNA for LC-MS analysis, 30 µg of DNA prepared from the E. coli EC135 strain or strains expressing MTases were digested to deoxynucleosides with 50 U of DNA Degradase Plus (Zymo Research); the digestion was carried out in 100 µL volume at 37°C for 18 h. The m6dA standard was purchased from Santa Cruz Biotechnology.
The characterization of m6dA was performed on an Agilent 6520 Accurate-Mass QTOF LC/MS system (Agilent Technologies) equipped with an electrospray ionization (ESI) source. 30 µL of the samples were injected to the Agilent 1200 HPLC using an Agilent Zorbax Extend-C18 1.8 µm 2.1×50 mm column with the column temperature kept at 35°C. Water with 0.1% formic acid and methanol were used as mobile phases A and B, respectively, with a flow rate of 0.2 mL/min. The following gradient was used: 0% B for 3.0 min, increase to 60% B in 4.5 min, 60–95% B over 2.5 min, 95% B for 5 min, and then decreased to 0% B over 0.5 min prior to re-stabilization of 14.5 min before the next injection.
The MS data were collected in positive ionization mode with nitrogen supplied as the nebulizing and drying gas. The temperature of the drying gas was set at 300°C. The flow rate of the drying gas and the pressure of the nebulizer were 600 L/h and 25 psi, respectively. The fragmentor and capillary voltages were kept at 90 and 3,500 V, respectively. Full-scan spectra were acquired over a scan range of m/z 80–1000 at 1.03 spectra/s.
Multiple MTase genes were rapidly assembled by taking advantage of the high DNA recombination activity in S. cerevisiae [49]. The CEN6 replicon was added to pBAD43 followed by TRP1 allele from pDDB78 at the ClaI site to yield pWYE724; the addition of these elements enables replication and screening in S. cerevisiae. The active MTase genes were amplified using PCR primers that contained 50 bp of overlapping sequence to the adjacent gene from their corresponding pBAD43 plasmids. S. cerevisiae DAY414 was transformed with the DNA fragments encoding the active MTases from the individual bacterial strains and the pWYE724 plasmid linearized at the EcoRI and SalI loci. S. cerevisiae DAY414 was then selected for tryptophan autotrophy on synthetic complete (SC) medium lacking tryptophan. S. cerevisiae transformation was performed using the lithium acetate method [50]. Plasmids were rescued into E. coli TOP10 cells as described by Robzyk et al [51]. All recombinant plasmids were verified by restriction digestion and DNA sequencing before subsequent use. The plasmids carrying multiple MTase genes from B. amyloliquefaciens TA208, B. cereus ATCC 10987 and N. hamburgensis X14 were named pM.Bam, pM.Bce and pM.Nham, respectively.
The homologous DNA sequences flanking the upp gene of B. amyloliquefaciens TA208 (641 bp upstream and 669 bp downstream) and the chloramphenicol resistance gene of pMK4 were amplified and joined using SOE-PCR. This cassette was ligated into the pMD19-T vector (Takara) and verified by DNA sequencing. The resulting plasmid was named pWYE748.
A 216 bp promoter region of the Nham_3450 gene was PCR amplified from the genome of the N. hamburgensis X14 strain and joined to gfpmut3a by SOE-PCR. The resulting GFP expression cassette was ligated into pBBR1-MCS5 at the SalI and PstI sites to yield pWYE561.
Various shuttle and integrative plasmids were transformed into the E. coli EC135 strains carrying MTase encoding genes. MTase expression was then induced by incubation with 0.2% arabinose at 30°C to allow the in vivo methylation of these plasmids.
Transformation of B. cereus ATCC 10987 was carried out as described previously with the following modifications [24]. The B. cereus ATCC 10987 strain was cultured in LB medium until the culture reached an OD600 of 0.2 and was then incubated on ice for 10 min. Cells were harvested by centrifugation at 8,000 g at 4°C for 10 min. After washing four times with ice-cold transformation buffer (10% sucrose, 15% glycerol, 1 mM Tris-HCl, pH 8.0), the electro-competent cells were resuspended in 1/125 volume of the original culture. The cells (90 µL) were mixed with 100 ng of the column-purified plasmids and loaded into a pre-chilled 1 mm gap cuvette. After a brief incubation on ice, the cells were shocked with a 2.1 kV pulse generated by a BTX ECM399 electroporator (Harvard Apparatus). The cells were immediately diluted with 1 mL NCMLB medium (17.4 g/L K2HPO4, 11.6 g/L NaCl, 5 g/L glucose, 10 g/L tryptone (Oxoid), 5 g/L yeast extract (Oxoid), 0.3 g/L trisodium citrate, 0.05 g/L MgSO4·7H2O, 69.2 g/L mannitol and 91.1 g/L sorbitol, pH 7.2) and incubated at 37°C for 3 h to allow the expression of the antibiotic resistance genes. Aliquots of the recovery mix were spread onto LB plates supplemented with 5 µg/mL chloramphenicol or 10 µg/mL kanamycin and cultured overnight at 37°C.
Electroporation of B. amyloliquefaciens TA208 was performed using the combined cell-wall weakening and cell-membrane fluidity disturbing procedure described previously [17].
The N. hamburgensis X14 strain was grown in DSMZ 756a medium (1.5 g/L yeast extract, 1.5 g/L peptone (BD Biosciences), 2 g/L NaNO2, 0.55 g/L sodium pyruvate, 1 mL/L trace element solution (33.8 mg/L MnSO4⋅H2O, 49.4 mg/L H3BO3, 43.1 mg/L ZnSO4⋅7H2O, 37.1 mg/L (NH4)6Mo7O24, 97.3 mg/L FeSO4⋅7H2O and 25 mg/L CuSO4⋅5H2O) and 100 mL/L stock solution (0.07 g/L CaCO3, 5 g/L NaCl, 0.5 g/L MgSO4⋅7H2O, 1.5 g/L KH2PO4), pH 7.4) at 28°C in the dark until reaching an OD600 of 0.1. The cells were then harvested by centrifugation at 8000 g at 4°C for 10 min and washed four times with ice-cold 10% glycerol. The cells were resuspended in 10% glycerol at a 1,000-fold greater concentration compared to that of the original culture volume. The cell suspension (90 µL) was mixed with 150 ng of the pWYE561 plasmid and electroporated with an ECM399 electroporator at 1.2 kV. The cells were washed into 100 mL 756a medium and recovered at 28°C with gentle shaking for one day. The bacteria were then grown in the presence of 20 µg/mL gentamycin for one day. The bacterial culture was used at a ratio of 1∶100 to inoculate fresh 756a medium containing antibiotics and was shaken at 180 rpm at 28°C. After about three weeks, the culture became turbid. The bacterial culture was subcultured once more to enrich for transformed cells and took one week to reach an OD600 of 0.1. The culture was tested for contamination microscopically and by streaking the culture onto LB plates. Successful transformation of strain X14 was verified by plasmid preparation using the Plasmid Mini Kit (OMEGA Bio-tek), PCR amplification of gfpmut3a and plasmid rescue. Expression of GFP was observed using a Leica TCS SP2 confocal laser scanning microscope (Leica Microsystems), and the ratio of fluorescent cells was determined using a BD FACS Calibur flow cytometer (BD Biosciences).
The pWYE748 plasmid was transformed into the E. coli EC135 strain harboring pM.Bam. After induction of MTase expression, 1 µg of the pWYE748 plasmid was transferred to B. amyloliquefaciens TA208, and the cells were selected for chloramphenicol resistance. Positive clones were verified by PCR and sequencing using primers (WB605 and WB606) specific to the flanking sequences of the homologous arms. The upp knock-out strain B. amyloliquefaciens BS043 was validated by growth on MM plates supplemented with 10 µM 5-FU [26] and 100 mg/L adenosine.
All of the PCR primers used in this study are listed in Table S2.
The modified plasmid DNA was challenged by the cognate REases to determine the modification sequences of the cloned MTases. To facilitate the identification, the high-copy plasmid pMK4 was transformed to E. coli EC135 harboring individual or multiple MTase genes, and in vivo methylated pMK4 plasmids were prepared and challenged by the cognate REases after linearization by REases that have sole cutting sites in pMK4 (NcoI, EcoRI, SpeI or BamHI). The pMK4 plasmids prepared from E. coli EC135, and the plasmids from B. amyloliquefaciens TA208 or B. cereus ATCC 10987 was used as the negative and positive controls in the experiments of individual MTase and multiple MTases, respectively. The plasmids pWYE690 and pHCMC02 were tested for their resistance to BclI cleavage conferred by BAMTA208_6525 when it was expressed individually and co-expressed due to the lack of BclI site in pMK4. For the same reason, pWYE699 and pHCMC05 were used in testing the protection conferred by BCE_0393 from BssHII cleavage.
Since the broad-host-range plasmid derivative pWYE561 showed a low copy number in E. coli, pMK4 was also used in identification of the modification sites of the MTases from strain N. hamburgensis X14. Genomic DNA of the strain was used as a control for co-expressed MTases.
|
10.1371/journal.pntd.0006429 | Microdeletion on chromosome 8p23.1 in a familial form of severe Buruli ulcer | Buruli ulcer (BU), the third most frequent mycobacteriosis worldwide, is a neglected tropical disease caused by Mycobacterium ulcerans. We report the clinical description and extensive genetic analysis of a consanguineous family from Benin comprising two cases of unusually severe non-ulcerative BU. The index case was the most severe of over 2,000 BU cases treated at the Centre de Dépistage et de Traitement de la Lèpre et de l’Ulcère de Buruli, Pobe, Benin, since its opening in 2003. The infection spread to all limbs with PCR-confirmed skin, bone and joint infections. Genome-wide linkage analysis of seven family members was performed and whole-exome sequencing of both patients was obtained. A 37 kilobases homozygous deletion confirmed by targeted resequencing and located within a linkage region on chromosome 8 was identified in both patients but was absent from unaffected siblings. We further assessed the presence of this deletion on genotyping data from 803 independent local individuals (402 BU cases and 401 BU-free controls). Two BU cases were predicted to be homozygous carriers while none was identified in the control group. The deleted region is located close to a cluster of beta-defensin coding genes and contains a long non-coding (linc) RNA gene previously shown to display highest expression values in the skin. This first report of a microdeletion co-segregating with severe BU in a large family supports the view of a key role of human genetics in the natural history of the disease.
| Buruli ulcer (BU) is a tropical infectious disease caused by Mycobacterium ulcerans. Although being the third most common mycobacterial disease in the world after tuberculosis and leprosy, BU remains a neglected tropical disease and an emerging health emergency in several developing countries. It causes profound skin ulcerations and eventually bone infections. Life-long functional sequelae are observed in more than 20% of patients, most of whom are children. Several observations, in particular the large variability in the clinical severity of the disease after infection, suggested the role of human genetic factors in the development of BU. We report the case of a 5-year old girl from Benin, born of consanguineous parents, who suffered from extensive dissemination of the mycobacterium in the skin, bones and joints. One of her siblings was also affected. The deep genetic exploration of this family led to the identification of a small deletion on chromosome 8 in both patients but absent from unaffected siblings. Interestingly, the deletion is located within a region containing genes encoding for beta-defensins, a family of antimicrobial peptides involved in both innate immunity and healing process of skin wounds. This first report of a microdeletion associated with severe BU in a large family supports the view of a key role of human genetics in the natural history of the disease.
| Buruli ulcer (BU), caused by Mycobacterium ulcerans, is the third most frequent mycobacteriosis worldwide, after tuberculosis and leprosy [1]. It mostly affects rural areas of tropical countries. No reliable estimate of global incidence is currently available but West Africa, annually reporting several thousand of cases, is considered as the principal endemic zone [2]. However, the incidence of BU is currently declining in several African countries including Benin. In 2016, 1,676 BU patients were reported to the World Health Organization (WHO) by African countries as opposed to 5,029 in 2009 (http://www.who.int/gho/neglected_diseases/buruli_ulcer/en/). The reasons for this decline are presently unknown although a role of the introduction of control strategies has been suggested [3].
BU is a devastating necrotizing skin infection classically characterized by pre-ulcerative lesions (nodules, plaques, edematous infiltration), eventually developing into deep ulcers with undermined edges. BU causes life-long functional sequelae in more than 20% of patients, most of whom are children [4]. The occurrence of sequelae in BU patients is not evenly distributed, and severe, sequelae-prone, BU forms have been defined as presentation with edema, osteomyelitis, or large (≥15 cm in diameter) or multifocal lesions [4, 5]. The unexplained variability in the clinical presentation of BU [4–6] together with the indication of familial clustering of cases [7, 8] suggest a role of host genetic factors in the natural history of BU in humans.
This hypothesis is consistent with the discovery of Mendelian predisposition to other mycobacterial infections in the context of the syndrome of Mendelian susceptibility to mycobacterial diseases (MSMD) and severe tuberculosis of childhood [9–12]. In addition, recent studies have reported association between BU and variants located in genes already implicated in MSMD, TB or leprosy [13, 14]. The role of host genetics in BU is further supported by two studies in Ghana [15] and in 11 African endemic countries including Benin [16] reporting highly restricted genetic variation of the microbe after the sequencing of more than 150 isolates of M. ulcerans therefore ruling out a major role of specific M. ulcerans strains in BU clinical outcomes. In the present work, we report an extensive genome-wide study aiming to decipher the genetic basis of a remarkably severe form of BU segregating in a large consanguineous multiplex family from Benin and suggestive of autosomal recessive inheritance.
The family was enrolled at the Centre de Dépistage et de Traitement de la Lèpre et de l’Ulcère de Buruli (CDTLUB) in Pobe, Benin, due to the unusual severity of the clinical course of the disease in patient P1 (details given in the case report section). The family consisted of two unaffected parents, two affected children and seven unaffected siblings living in the village of Tatonnonkon. This village is located in Adja-Ouèrè, a district of the Plateau department at the borders of the Zou and Ouémé departments, with BU prevalence ranging from 10 to 18 per 1,000 [17]. Blood was obtained from the parents, the two affected and three unaffected siblings. DNA was extracted from whole blood according to the Nucleon BACC2 Genomic DNA extraction protocol (GE Healthcare), assayed with the QuantIt Picogreen dsDNA kit (Life Technologies) and processed for the genotyping of >900,000 single nucleotide polymorphisms (SNPs) used for both linkage and copy number variant (CNV) analysis and >900,000 monomorphic nucleotides used for CNV analysis by the Affymetrix Genome-Wide 6.0 array.
The genetic research on susceptibility to BU was approved by the institutional review board of the CDTLUB and the national Beninese BU control authorities (IRB00006860), as well as the ethics committee of the university hospital of Angers, France (Comité d’Ethique du CHU d’Angers). All participants (parents and five children) provided written informed consent or had their parents provide written informed consent on their behalf. Parents of patients P1 and P2 have given written informed consent to publish anonymized case details including pictures and X-rays.
Stringent quality control (QC) procedures were applied. Individual QC consisted of checking the individual call rate (>95%), the match between genetic and declared sex and the match between genetic and declared degrees of familial relatedness. For linkage analysis, family-based SNP QC was performed and only SNPs with a within-family SNP call rate of 100%, a non-zero minor allele frequency (MAF) and no Mendelian errors were retained. As several QC measures (e.g. Hardy-Weinberg equilibrium filters) cannot be applied to a single family because of its intrinsically limited size, we further filtered SNPs on a population basis, using the Yoruba population from the Hapmap project (Affymetrix 6.0 genotyping, [18]). We retained SNPs with a population call rate ≥95% and a non-zero MAF that were in Hardy-Weinberg equilibrium at the 0.01 level.
Linkage analysis was performed by homozygosity mapping, a powerful statistical approach for detecting genetic linkage in the presence of familial consanguinity. We used MERLIN version 1.2 and its clustering option to take linkage disequilibrium into account and specified a recessive model with complete penetrance [19, 20]. Linkage regions were further screened for causal point mutations and/or structural variations, through whole-exome sequencing and CNV analysis. We also estimated homozygosity rates in children using runs of homozygosity as previously described [19, 20].
Whole-exome sequencing was performed on both affected patients. Genomic DNA was sheared with a Covaris S2 Ultrasonicator (Covaris). An adaptor-ligated library was prepared with the Paired-End Sample Prep kit V1 (Illumina). Exome capture was performed with the SureSelect Human All Exon v2 kit (Agilent Technologies), covering 38 Megabases (Mbs) of the genome. Single-end sequencing was performed on an Illumina Genome Analyzer IIx. The sequences were aligned with the human genome reference sequence (hg19/GRCh38 build), with BWA-MEM aligner [21]. Downstream processing was carried out with the Genome Analysis Toolkit (GATK) [22], SAMtools [23], and Picard Tools (http://broadinstitute.github.io/picard/). Substitution and indel calls were both made with GATK HaplotypeCaller v3.3. All calls with a Phred-scaled quality ≤30 were filtered out. Variant annotation was based on the Human genome assembly GRCh38 as implemented in the Ensembl browser (release 88) as previously described [24–26].
Genome-wide CNV analysis of the familial data was performed on Affymetrix 6.0 data with the joint calling algorithm of PennCNV, which takes familial information into account, to improve CNV calls within families [27]. In addition, we took advantage of an ongoing genetic study including 401 healthy local controls (Median age at the time of enrolment = 40 years old; Male:female sex ratio = 0.72) and 402 local laboratory-confirmed BU cases (Median age at the diagnosis = 11 years old; Male:female sex ratio = 0.81) genotyped with the Illumina Omni2.5 chip (which includes over 2.3 million SNPs) to further investigate any CNV-related findings made in the familial study. CNV analysis among these 803 individuals was performed as in the familial data by means of PennCNV. These 803 individuals of Yoruba ethnicity were also enrolled through the CDTLUB in Pobe, Benin, and lived in villages distributed over the Ouémé and the Plateau departments in an endemic area with BU prevalence estimated around 8 to 20 per 1,000 [17, 28].
Refined targeted resequencing by Next-Generation Sequencing (NGS) to detect CNVs was performed in four individuals of the family (the two affected sisters predicted to be homozygous for the deletion by PennCNV, one predicted heterozygote brother and one predicted homozygous wild-type brother) and six independent controls (2 men and 4 women all predicted to be homozygous wild-type) that were part of the 401 local controls described above and aged above 30 years old to increase the likelihood of exposure to M. ulcerans. Targeted resequencing was done by means of ‘capture by hybridization’ approach. Illumina compatible bar-coded genomic DNA libraries were constructed according to the manufacturer’s sample preparation protocol (Ovation Ultralow, Nugen Technologies). Briefly, 1 to 3 μg of each patient’s genomic DNA was mechanically fragmented to a median size of 200 base pairs (bps) using a Covaris. 100 ng of double strand fragmented DNA was end-repaired and adaptors containing a specific eight bases bar-code were ligated to the repaired ends (one specific bar-code per patient). DNA fragments were PCR amplified to get the final precapture bar-coded libraries that were pooled at equimolar concentrations (a pool of 15 libraries was prepared). The biotinylated single strand DNA probes were designed and prepared to cover a 157 kilobases (kbs) chromosomal region on chromosome 8. The limits of the targeted chromosomal region are Chr8:12,532,612–12,690,151 according to the GRCh38 assembly of the human reference genome. During the capture process, bar-coded libraries molecules complementary to the biotinylated beads were retained by streptavidin coated magnetic beads on a magnet and PCR amplified to generate a final pool of post capture libraries covering the targeted chromosomal region on chromosome 8. In total a pool of 15 libraries covering the 157 kbs of interest on chromosome 8 was sequenced on an Illumina HiSeq2500 (Paired-End sequencing 130x130 bases, High Throughput Mode, 15 samples per lane). Finally, sequence reads were aligned to the human hg19/GRCh38 reference genome using the Burrows-Wheeler Alignment version 0.6.2.13 [21].
The index case, P1, a girl born in 2000, was identified as the most severe case of BU ever diagnosed among more than 2,000 patients seen at the CDTLUB in Pobe, Benin since its opening in 2003 (Fig 1 and S1 Fig). The parents were self-declared second-degree cousins. At the age of five years, P1 presented with cachexia, fever and an edematous lower right limb (Fig 1A, left pannel). Mycobacterium ulcerans was identified in the synovial fluid of the right knee by Ziehl-Neelsen staining and IS2404 PCR amplification. P1 also presented with PCR-positive edema of the left foot, leading to the X-ray confirmed diagnosis of osteomyelitis of the left cuboid and osteoarthritis of the right tibia and knee (Fig 1B, left panel). Despite early and prolonged (13 weeks vs. 8 weeks for the standard regimen) antibiotic treatment with rifampicin and streptomycin and several surgical procedures, the infection spread further. Two months after diagnosis, edema of the right arm revealed M. ulcerans osteoarthritis of the right elbow, further confirmed by the aspiration of caseiform matter, Ziehl-Neelsen staining and PCR amplification (Fig 1C, left panel). Sustained fever did not recede before the amputation of the patient’s right leg after three months (Fig 1B, right panel). After six months, osteomyelitis of the left radius was diagnosed (Fig 1C, right panel). One year after diagnosis, left fibula involvement was detected (Fig 1B, right panel).
P1 thus suffered from unusually severe edematous BU with unprecedented dissemination to ten bones and two joints in all four limbs (Fig 1 and S1 Fig). She underwent a number of surgical procedures on all four limbs in the first 18 months after diagnosis. The disease relapsed two, three and five years after the initial diagnosis, with the re-emergence of infection at new sites: the right humerus, left calcaneal tendon and left tibia, respectively (S1 Fig). Mycobacterium ulcerans was repeatedly identified by Ziehl-Neelsen staining and IS2404 PCR on the various lesions of all four limbs, including surgical bone samples, at different time points (S1 Fig). The medical history of P1 also included uncomplicated acute HBV hepatitis and several episodes of classical malaria up to 2016. However, as of December 2017, no additional BU-related clinical events have been reported.
P1’s sister, P2, was born in 1995. At 13 years of age, she was also diagnosed with a severe form of BU, presenting as rapidly spreading edema of the whole right arm, right forearm and right hand and a large plaque of the right elbow. PCR and culture were positive for M. ulcerans. She was given 8 weeks of antibiotic treatment combining rifampicin and streptomycin. Over the three months following the diagnosis, she underwent several excision and curettage surgeries and a skin graft (S1 Fig). Although diagnosed very rapidly after clinical expression of the disease due to the strict clinical supervision of P1 and her family, P2 still presented with a severe form including large edema and plaque. Of particular interest is the observation that the two patients presented a severe non-ulcerative form of BU, a rare occurrence in the clinical landscape of the disease [29]. As easily inferable, P1 and P2 suffer from permanent functional limitations significantly impacting their daily life. They were HIV-negative and had been vaccinated with BCG at birth. Their complete blood formulas were normal. On the contrary, both parents and seven other siblings were unaffected at the time of the study and to the best of our knowledge still were as of December 2017.
After QC, 310,993 SNPs of the Affymetrix 6.0 array remained and were grouped into 126,704 independent clusters for model-based linkage analysis by homozygosity mapping, reaching a genome-wide mean information content of 0.99, based on genotyped individuals. Identity-by-state analysis confirmed that P2’s twin was a fraternal, not identical, twin. Homozygosity was estimated at ~2% in the five children of the family for whom DNA was available, consistent with the level of self-declared consanguinity, i.e. second degree cousin for the parents [30]. Homozygosity mapping aimed at identifying regions of shared runs of homozygosity inherited identical-by-descent by affected individuals, but not by unaffected individuals. We computed the theoretical maximum LOD score at a completely informative marker, given the familial configuration and the genetic model. In our scenario (two affected siblings, three unaffected siblings, consanguinity loop involving second-degree cousins and a recessive genetic model with complete penetrance), this theoretical maximum LOD score was equal to 2.8 (corresponding to a p-value of 3x10-4).
Eight regions spanning a total of 5.7 Mbs and mapping to chromosomes 2 (linkage regions 2.1 and 2.2), 5, 7 (7.1 and 7.2) and 8 (8.1, 8.2 and 8.3) reached or closely approached the theoretical maximum LOD score ((Fig 2). The second region on chromosome 2 (2.2), and both regions on chromosome 7 (7.1 and 7.2) had LOD score of 2.49, 2.68 and 2.35, respectively. All the other linked regions had a LOD score above 2.75. The complete list of genes in each of these regions, retrieved from the Vega database as implemented in the Ensembl browser (vega.archive.ensembl.org), is available in the S1 Table. Interestingly, two of the three linkage regions on chromosome 8, i.e. 8.1 and 8.2, contain clusters of genes encoding beta-defensins (S2 Fig).
We screened the linkage regions for potential causal mutations by means of whole-exome sequencing of the two patients. We searched for homozygous variants predicted to have potential functional effect, i.e. missense, nonsense and splice site mutations, in-frame and out-of-frame small insertions and deletions, common to both patients and located in the linkage regions. We filtered out population variants with a MAF above 1% from public databases such as the Exome Aggregation Consortium (ExAC; exac.broadinstitute.org). No candidate variants fitting these criteria were detected. However, the two linkage regions harboring clusters of beta-defensin genes (i.e. 8.1 and 8.2) displayed low mapping quality with mean values of 2 and 11, respectively, versus values >55 for all other linkage regions, and a mean mapping quality value of 46 for the whole-exome.
Next, we screened the linkage regions for structural variations by means of PennCNV analysis of the seven members of the family [27]. A single homozygous deletion common to both patients was found around position 12,616,035 on chromosome 8 (GRCh38 assembly). The deletion was predicted to be heterozygous in both parents, heterozygous in one unaffected sibling and absent from the other two unaffected siblings studied therefore co-segregating perfectly with the phenotype (Fig 3). This deletion was located close to a cluster of beta-defensin genes in linkage region 8.2 (S2 Fig) and was the only CNV, anywhere in the genome, to display this perfect pattern of familial segregation. The genetic screening of 401 local controls by means of PennCNV (see Materials and Methods section) identified two individuals heterozygous for the deletion but did not detect any homozygous carriers. Based on these results, the frequency of the deletion was predicted to be 2.5x10-3. Of note, CNV prediction algorithms based on genotyping data may not have an optimal resolution to detect heterozygous carriers of small deletion, i.e. true frequency for the deletion may be somewhat higher. The same screening of 402 local BU cases identified two homozygous carriers of the deletion: one female diagnosed at the age of 25 years old with a severe form of BU, i.e. 20 centimeters edema located on the left lower limb therefore classified as 3 in the 3-class WHO severity scale, and another female diagnosed at the age of 15 years old with a less severe form, i.e. a 12 centimeters edema located on the lower limb therefore classified as 2 according to WHO scale. Remarkably, similar to P1 and P2 these two patients also developed a non-ulcerative form of the disease.
We performed targeted resequencing of the CNV region by original NGS experiments, i.e. capture by hybridization approach, the most accurate approach to date to detect CNV in the context of our study. This resequencing was performed in four individuals of the family (the two affected sisters predicted to be homozygous for the deletion by PennCNV, one predicted heterozygous brother and one predicted homozygous wild-type brother) and six independent controls (all predicted to be homozygous wild-type) that were part of the 401 local controls and aged above 30 years old (Fig 4). The distribution of the mean number of reads (X) in the targeted region allowed us to validate unambiguously the deletion at the molecular level, and to refine its breakpoints from chr8:12,609,841 to chr8:12,647,341 (GRCh38 assembly). Over this 37 kbs region the mean (interquartile range) coverage was 0.47 X (0–0) in the two affected sisters homozygous for the deletion, 198.5 X (94–274) in the unaffected heterozygote brother and 342.8 X (196–515) in the seven wild-type homozygous controls (including one unaffected brother with a mean coverage of 330.1 X).
A brief description of this region–derived from the Vega database as implemented in the Ensembl genome browser (vega.archive.ensembl.org)–based on GRCh38 Assembly is given in Fig 4. Three genes of different biotypes have been identified so far including two processed pseudogenes AC068587.5 (Ensembl id: ENSG00000255253; spanning 174bp from 12,638,428 to 12,638,602) and AC068587.2 (ENSG00000244289; 780bps from 12,628,476 to 12,629,256), and one long non-coding RNA (lincRNA) gene AC68587.6 (ENSG00000283674; ~128kbs from12,537,079 to 12,665,588). Remarkably, this lincRNA gene was previously shown to display highest expression values in the skin (https://www.ncbi.nlm.nih.gov/gene/?term=ensg00000283674) [31].
Overall, the observations that 1) exome sequencing found no evidence for a variant with potential functional effect in the genes of the linked regions, 2) the confirmed deletion co-segregates with the phenotype in our family with P1 and P2 being homozygous carriers for the deletion, 3) no homozygous carriers were found in a sample of 402 local controls, 4) two homozygous carriers were found in a sample of 401 local BU cases and 5) the deletion includes a lincRNA with highest expression values in the skin and located close to a cluster of beta-defensin genes, altogether suggest this deletion as a critical trigger of severe non-ulcerative forms of BU.
We report the clinical phenotype and extensive genetic analysis of a multiplex consanguineous family with BU of unusual severity displaying Mendelian inheritance. The index patient P1 suffered from aggressive multifocal edematous BU and M. ulcerans osteoarthritis with several relapses over a period of five years despite intensive medical and surgical care. The disease spread to all four limbs and resulted in severe handicap, including amputation of the right leg. P1’s sister also suffered from severe edematous BU of the right arm, forearm and hand, leading to permanent functional sequelae. The subsequent genetic investigations were performed assuming P1 and P2 should share the same genetic defect because they share similar phenotypes. This assumption may sound somewhat speculative as the clinical picture of P1 appeared more severe than the one of P2. However, two important aspects must be considered. First, because of P1 dramatic clinical course, the whole family was under tight clinical monitoring, and, although diagnosed at a significantly earlier stage than P1, P2 still presented a severe form of BU. Any delay in the diagnosis may have resulted in a more severe clinical outcome. Second, P1 and P2 developed a very severe non-ulcerative form of the disease, a rare clinical picture first emphasized in 2000 [29].
We searched for homozygous mutations located within these linkage signals, by performing whole-exome sequencing in both patients. We did not identify any homozygous mutations common to the two affected sisters within the linkage regions. We could not exclude the possibility of such mutations being present in the two beta-defensin-containing linkage regions 8.1 and 8.2, as the most recent alignment algorithms fail to generate a unique alignment of the sequencing reads in the beta-defensin clusters [32]. Extremely high levels of sequence repetition resulted in very low mapping quality values in these regions, precluding reliable genotype calling. We checked both public and in-house whole-genome sequencing data generated from other individuals and found that this sequencing method, which generally performs better overall, also suffered from this limitation in the defensin region [33]. The ongoing development of novel sequencing technologies based on the generation of reads as long as several Mbs [34] should provide an efficient mean to overcome this issue. Nevertheless, our analysis of whole-exome sequencing data for both patients in regions that were well-covered with excellent mapping quality (i.e. the large majority of exome target sequences) was useful to rule out the presence of homozygous candidate mutations in this family.
Because whole-exome sequencing did not provide any evidence for exonic point mutations explaining the linkage signals, we tested the hypothesis of structural variations as potential causes of these signals by means of in silico CNV analysis. We found a homozygous deletion spanning ~ 10 kbs located within the linkage region 8.2 common to P1 and P2 but absent from unaffected siblings and 401 local controls. Further screening of 402 unrelated BU local cases identified two additional homozygous carriers. Remarkably, the two homozygous carriers of the deletion also developed a non-ulcerative form of the disease. This is suggestive of severe non-ulcerative BU being a very specific clinical form of the disease. CNV detection algorithms have a modest resolution for determining CNV breakpoints and are known to underestimate CNV size, implying that the deletion identified here may be larger [35]. Indeed, targeted resequencing in 10 individuals (including four siblings and six local controls) confirmed the presence of a homozygous 37 kbs deletion in the two patients, heterozygous in one unaffected sib and absent in the remaining individuals. Of note, all in silico predictions were unambiguously confirmed by the resequencing assay further supporting the in-silico results observed in the local sample of 401 controls and 402 cases.
The deletion extends from chr8:12,609,841 to chr8:12,647,341 and comprises the lincRNA gene AC068587.6 located in close vicinity to beta-defensin clusters. LincRNAs are thought to regulate the expression of neighboring genes with very strong tissue specificity [36]. Remarkably, AC068587.6 has been shown to display its highest expression values in the skin ((https://www.ncbi.nlm.nih.gov/gene/?term=ensg00000283674). In addition, it is located close to a cluster of beta-defensins encoding genes (S1 Table). Beta-defensins are a family of antimicrobial peptides involved in innate immunity; they are widely secreted, throughout epithelial tissues, in response to infectious agents [37–40]. Interestingly, beta-defensins have also been implicated in the healing process of aseptic skin wounds [37, 39–41]. The antibacterial and wound repair functions of beta-defensins are consistent with a role for these molecules in the human response to M. ulcerans. They have been shown, in vitro and in vivo, to be upregulated in response to several mycobacteria [42–47], including M. ulcerans [48], and have been specifically implicated in the response to bone infection in both mice and humans [49–51]. These biological functions of beta-defensins together with the highest expression of AC068587.6 in skin, likely the most relevant tissue in BU pathophysiology, strongly support the hypothesis of this lincRNA playing a role in the pathophysiology of BU.
Most of the exposed individuals in foci of highly endemic infection do not develop BU lesions, but some, such as P2, rapidly develop extensive skin lesions, and others, such as P1, rapidly develop multifocal skin and bone lesions despite aggressive medical and surgical treatment. This inter-individual variability in the human response to M. ulcerans may have a genetic origin, a hypothesis supported by the identification of a homozygous deletion within a linkage region containing beta-defensin genes in this large multiplex family from Benin. A number of monogenic predispositions to common infections, such as tuberculosis, have been described and genetically dissected over the last decade [9–12, 52–57] but this is the first report of monogenic inheritance in severe BU. More refined functional investigations, such as silencing RNA followed by defensins dosage, are needed to obtain additional sources of evidence supporting causality between the deletion identified, the lincRNA, beta-defensins and the clinical phenotype [58]. We will need to face up to the non-trivial difficulties posed by the extreme complexity of this repetitive and dynamic region of the human genome, an exciting and promising challenge for future research.
|
10.1371/journal.pntd.0006074 | Csseverin inhibits apoptosis through mitochondria-mediated pathways triggered by Ca2 + dyshomeostasis in hepatocarcinoma PLC cells | Numerous experimental and epidemiological studies have demonstrated a link between Clonorchis sinensis (C. sinensis) infestation and cholangiocarcinoma (CCA) as well as hepatocellular carcinoma (HCC). The underlying molecular mechanism involved in the malignancy of CCA and HCC has not yet been addressed. Csseverin, a component of the excretory/secretory products of C. sinensis (CsESPs), was confirmed to cause obvious apoptotic inhibition in the human HCC cell line PLC. However, the antiapoptotic mechanism is unclear. In the present study, we investigated the cellular features of the antiapoptotic mechanism upon transfection of the Csseverin gene.
In the present study, we evaluated the effects of Csseverin gene overexpression on the apoptosis of PLC cells using an Annexin PE/7-AAD assay. Western blotting was applied to quantify the activation of caspase-3 and caspase-9, the mitochondrial translocation of Bax and the release of Cyt c upon Csseverin overexpression in PLC cells. Laser scanning confocal microscopy was used to analyze the changes of intracellular calcium. Fluorescence assay and immunofluorescence assays were performed to observe the changes of the mitochondrial permeability transition pore (MPTP).
The overexpression of Csseverin in PLC cells showed apoptosis resistance after the induction of apoptosis. Additionally, the activation of caspase-3 and caspase-9 was specifically weakened in Csseverin overexpression PLC cells. The overexpression of Csseverin reduced the increase in intracellular free Ca2+, thereby inhibiting MPTP opening in PLC cells. Moreover, Bax mitochondrial translocation and the subsequent release of Cyt c were downregulated in apoptotic Csseverin overexpression PLC cells.
The present findings suggest that Csseverin, a component of CsESPs, confers protection from human HCC cell apoptosis via the inactivation of membranous Ca2+ channels. Csseverin might be involved in the process of HCC through C. sinensis infestation in affected patients.
| Multiple studies have contributed to the association between Clonorchis sinensis (C. sinensis) infestation and cholangiocarcinoma (CCA) as well as hepatocellular carcinoma (HCC) in past years. However, studies on the underlying pathogenic mechanisms of C. sinensis lag behind those of other parasitic diseases. The excretory/secretory products of C. sinensis (CsESPs) are pathogenic, as these products promote cell proliferation, suppress cell apoptosis and stimulate inflammation. Csseverin, a component of CsESPs, inhibited the apoptosis of the human HCC cell line PLC in our previous study. The present study illustrated that Csseverin conferred human HCC cells protection from apoptosis via an intrinsic pathway (mitochondrial-mediated) triggered by the inactivation of membranous Ca2+ channels.
| Clonorchis sinensis (C. sinensis) causes clonorchiasis, which is widely distributed in East Asia with heavily endemic zones in China, Taiwan, Vietnam, Russia, and Korea[1]. C. sinensis was reclassified as a group-I biocarcinogen for cholangiocarcinoma (CCA) by the International Agency for Research on Cancer (IARC) in 2009[2]. In endemic areas of China, 16.44% of hepatocellular carcinoma (HCC) patients were infected with C. sinensis, while 2.40% of non-tumor patients were infected [3]. This biocarcinogen has been included in control programs of neglected tropical diseases by the WHO[4]. The illumination of the precise mechanism linking C. sinensis with the development of HCC and CCA will help to prevent or postpone disease progression. Excretory-secretory proteins from C. sinensis (CsESPs) play important roles in the interactions between the worm and host, including the pathogenesis of inflammation, immune responses and carcinogenesis induced by the infection.
Escaping from apoptosis is an important aspect of cancer pathogenesis and has been widely recognized as a trait of most types of cancer [5]. In a previous study, we observed that Csseverin (a component of CsESPs), a homologous protein of the gelsolin family, caused obvious apoptotic inhibition in the human HCC cell line PLC. By promoting apoptosis suppression, Csseverin might accelerate the progress of HCC patients combined with C. sinensis infection [6]. It is worth studying the exact molecular mechanisms involved in the anti-apoptotic effects induced by Csseverin.
The gelsolin family had been implicated in the regulation of cell motility, apoptosis and phagocytosis [7]. The expression of gelsolin family proteins is reduced in many cancers, associated with poor prognosis and therapy resistance [8–9]. There is now increasing evidence that gelsolin family proteins are multifunctional regulators of cell apoptosis and cell metabolism, which involves multiple mechanisms[10–14].
Apoptosis can be executed in two distinct signaling cascades: the extrinsic pathway and the intrinsic pathway[15–16]. In the extrinsic pathway, apoptosis is triggered by death receptors, such as FAS-associated death domain protein (FADD), activating caspases 8 and 10 (the initiator caspases), which in turn activate executioner caspases 3, 6 and 7[17]. In the intrinsic pathway, the mitochondrial permeability transition pore (MPTP) plays a pivotal role in regulating the release of pro-apoptotic proteins, such as cytochrome c (Cytc). The released Cytc from mitochondria initiated the assembly of apoptosomes, activating factor 1 (Apaf-1) and caspase 9, an initiator caspase that cleaves and activates caspase 3 and 7[18]. MPTP is regulated by Bcl-2 proteins that induce the oligomerization of BAX (Bcl-2-associated protein) or BAK (Bcl-2 antagonist)[19].
The results of a previous study indicated that Csseverin binds to calcium ions in solution and actin filaments inside cells. We also demonstrated that the co-incubation of PLC cells with Csseverin in vitro led to apoptosis suppression based on the detection of the apoptosis-associated changes of mitochondrial membrane potential[6]. To further understand the anti-apoptotic role of Csseverin, we constructed stable Csseverin-overexpressing PLC cells (pEZ-LV203-Csseverin PLC) to avoid interference from endotoxin through the use of recombinant Csseverin. We detected a suppression effect of Csseverin on the early wave apoptosis of PLC cells. Furthermore, to investigate the mechanisms involved in Csseverin induced apoptosis suppression, we explored the effects of Csseverin on the activation of the caspase cascade, leading to the suppression of the permeability transition pore (MPTP), the mobilization of calcium, and the translocation of Cyt c and Bax.
The Ethics Committee of Sun Yat-Sen University reviewed and approved the protocols and experiments used in this study. The methods were carried out in accordance with the approved protocols. The data were collected and analyzed anonymously.
The human HCC cell line PLC were a gift from Dr. Wang Shutong (the First Affiliated Hospital of Sun Yat-Sen University) and routinely cultured in high glucose DMEM medium (Gibco, USA) supplemented with 10% fetal bovine serum (Gibco, USA) and penicillin-streptomycin (100 units/ml) in 5% CO2 at 37°C. The human 293T cells were kindly provided by GeneCopoeia (Rockville, MD, USA) and maintained in high glucose DMEM supplemented with 10% fetal bovine serum in 5% CO2 at 37°C.
Cox-IV, Caspase 3, Caspase 9, Bax and Cytochrome c were purchased from Cell Signaling Technology (Danvers, MA, USA). β-actin was obtained from Proteintech (USA). Anti-Csseverin serum was prepared as previously described [6].
The pEZ-LV203 lentiviral vector harboring the eGFP reporter gene was purchased from GeneCopoeia (Rockville, MD, USA). The pEZ-LV203 vector and Csseverin gene fragments were digested with EcoRI and Apa I, respectively, and subsequently ligated using T4 DNA ligase. The recombinant plasmid pEZ-LV203-Csseverin was identified by enzyme digestion and sequencing.
To generate the lentivirus, the pEZ-LV203-Csseverin plasmid or PEZ-LV203 control plasmid was cotransfected into 293T cells using the Lenti-Pac HIV Expression Packaging Kit (GeneCopoeia, USA) according to the manufacturer’s instructions. Supernatant containing the recombinant lentiviral particles was collected at 48 h post-transfection, filtered by a Millipore filter and subjected to ultracentrifugation. The lentiviral particles were re-suspended in cold phosphate-buffered saline (PBS) and used to infect PLC cells. The PLC cells were divided into three groups, pEZ-LV203-Csseverin PLC (transfected with pEZ-LV203-Csseverin plasmid), pEZ-LV203 PLC (PEZ-LV203 control plasmid) and PLC (no transfection). After 48 h, the cells were incubated in selection medium containing puromycin (3 mg/ml) for 7 days to select stably Csseverin-overexpressing PLC cells (pEZ-LV203-Csseverin PLC) and control PLC cells (pEZ-LV203 PLC). The transfection efficiency of pEZ-LV203-Csseverin PLC was evaluated by the expression of eGFP, and the Csseverin protein expression levels of pEZ-LV203-Csseverin PLC were measured by Western blot analysis.
http://dx.doi.org/10.17504/protocols.io.kcdcss6[PROTOCOL DOI]
Apoptotic cells were assessed by Annexin PE/7-AAD detection as previously described [6]. Briefly, Control groups (pEZ-LV203 PLC and PLC) or pEZ-LV203-Csseverin PLC cells were plated at a density of 105 cells per well in 6-well plates, and apoptosis was spontaneously induced after serum starvation for 48 h. The cells were collected by centrifugation, washed with cold PBS, and subsequently resuspended in 500 μl of 1× Binding Buffer prior to incubation with 5 μl of Annexin PE and 5 μl of 7-AAD (Keygentec, Nanjing, China). The cell samples were incubated at room temperature for 20 min and subsequently detected by a flow cytometer (Beckman Coulter Gallios, USA) to determine the apoptotic cell fractions.
http://dx.doi.org/10.17504/protocols.io.kcecste[PROTOCOL DOI]
The pEZ-LV203-Csseverin PLC cells were pretreated by serum starvation for 48 h, and pEZ-LV203 PLC and PLC cells were used as controls. A total of 5 × 106 cells were collected and treated with 300 μl of RIPA buffer (150 mM NaCl, 50 mM Tris, pH 7.4, 1% NP40, 0.1% SDS, and 0.5% sodium deoxycholate) supplemented with protease and phosphatase inhibitors (Keygentec, Nanjing, China).
To monitor the shift in Cytc from the mitochondria and Bax from the cytosol, we fractionated the cytosolic and mitochondrial fractions using a Cell Mitochondria Isolation Kit according to the manufacturer's instructions (Beyotime Institute of Biotechnology, China). The pEZ-LV203-Csseverin PLC cells were pretreated by serum starvation for 48 h, and pEZ-LV203 PLC and PLC cells were used as controls. A total of 5 × 106 cells were collected after brief trypsinization, followed by two more washes with PBS, and the cell pellet was resuspended in 200 μl of mitochondria extraction buffer containing 0.02 mM phenylmethanesulfonyl fluoride (PMSF) and proteinase inhibitors (Keygentec, Nanjing, China). After incubating on ice for 20 min, the cells were homogenized using a glass Dounce and pestle. The homogenates were centrifuged at 600 g for 15 min at 4°C, and the resulting supernatant was collected and centrifuged at 11,000 g for 15 min at 4°C to separate the mitochondria (pellet) and cytoplasmic proteins (supernatant). The mitochondria pellet was lysed in mitochondria extraction buffer (KeyGen Biotech, Nanjing, China).
http://dx.doi.org/10.17504/protocols.io.kcecstedx.doi.org/10.17504/protocols.io.kcfcstn [PROTOCOL DOI]
Western blotting analysis to determine the levels of apoptosis-related proteins was performed using standard techniques. The concentration of protein was determined by the BCA protein assay kit (Beyotime Institute of Biotechnology, China). Equal amounts of protein were subjected to Western blotting analysis. The proteins (40 μg) were separated according to molecular weight on a 12% SDS-PAGE gel and transferred onto a polyvinylidene difluoride (PVDF) membrane. The membranes were blocked with 1% bovine serum albumin in Tris-Buffered Saline Tween-20 (TBST, pH 7.4) at room temperature for 2 h, and probed overnight at 4°C with specific primary antibodies at the following dilutions: β-actin and Cox-IV, 1:2000; anti-Csseverin sera, 1:100; caspase 3 and caspase 9, 1:1000; and Bax and Cyt c, 1:500. After washing with TBST, the membranes were incubated with goat-anti-mouse or goat-anti-rabbit horseradish peroxidase-conjugated secondary antibody (1:5000) for 1 h at room temperature. Immunoreactive bands were visualized by the enhanced chemiluminescence detection kit (KeyGen Biotech, Nanjing, China) and quantified using the Gel-pro 4.5 Analyzer (Media Cybernetics, USA).
The intracellular Ca2+ concentration was estimated by co-incubating the cells with a cell-permeant Ca2 + fluorophore, Rhod-2 AM (2 μM). PEZ-LV203-Csseverin PLC cells were seeded at a density of 102 cells onto a confocal culture dish and treated by serum starvation for 48 h, and pEZ-LV203 PLC and PLC cells were used as controls. The cells were washed with cold PBS and incubated in a 5% CO2 humidified incubator at 37°C for 20 min after adding 20 μl of Rhod-2 AM working solution (AAT Bioquest, USA). Next, the cells were washed twice with PBS and the changes of intracellular calcium were evaluated by a laser scanning confocal microscope (Zeiss LSM 710, Germany). The Rhod-2 AM fluorescence was observed at 525 nm excitation (Ex)/590 nm emission (Em).
http://dx.doi.org/10.17504/protocols.io.kcecstedx.doi.org/10.17504/protocols.io.kcgcstw[PROTOCOL DOI]
The mitochondrial permeability transition pore (MPTP) was detected by tetramethyl rhodamine methyl ester (TMRM) in the Cell MPTP assay kit (Genmed Scientific Inc., Arlington, TX, USA). TMRM is a membrane-permeable fluorophore. In live cells, the hydrolysis of TMRM by intracellular esterases produces strongly red fluorescent tetramethyl rhodamine, a lipophilic compound well retained in cell mitochondria. The cytoplasm was stained with the methyl ester derivative of TMRM quenching of mitochondria rhodamine fluorescence. This nature of TMRM enables the assessment of MPTP opening[20].
Briefly, the control groups (pEZ-LV203 PLC and PLC) or pEZ-LV203-Csseverin PLC cells were seeded at a density of 103 cells per well onto 6-well plates, and spontaneous apoptosis was induced through serum starvation for 48 h. The cells were rinsed with GENMED cleaning solution and incubated with 1 ml GENMED staining solution for 20 min at 37°C in the dark. The supernatant was subsequently discarded, and the cells were washed twice with GENMED cleaning solution. Subsequently, the changes of MPTP were monitored using an inverted fluorescence microscope (Leica DMI4000B, Germany).
Quantitative changes of MPTP during cell apoptosis were measured by flow cytometry with the TMRM probe. After induced spontaneous apoptosis by serum starvation for 48 h, 105 cells were harvested and resuspended with GENMED cleaning solution. Subsequently, the cell suspensions were incubated with 0.5 ml of TMRM working solution for 20 min at 37°C in the dark. The staining solution was removed by centrifugation. The cells were washed twice with GENMED cleaning solution, subsequently resuspended in 200 μl of buffer solution and detected using a flow cytometer (Beckman Coulter Gallios, USA).
http://dx.doi.org/10.17504/protocols.io.kcecstedx.doi.org/10.17504/protocols.io.kchcst6[PROTOCOL DOI]
The data were analyzed for statistical significance using SPSS 13.0 software (SPSS, Chicago, IL, USA). The results are expressed as the means±SD from at least 3 independent experiments performed in duplicate. Statistical comparisons of the results were performed using one-way analysis of variance (ANOVA). A P value < 0.05 was considered statistically significant.
As shown in Fig 1A, green fluorescence was observed in pEZ-LV203-Csseverin PLC cells containing the pEZ-LV203-Csseverin vector fused with the eGFP reporter gene (Fig 1A). Compared to control PLC cells (pEZ-LV203 PLC), Csseverin expression was significantly increased in pEZ-LV203-Csseverin PLC cells, as shown by Western blotting (Fig 1B).
We conducted an Annexin PE/7-AAD binding assay using flow cytometry and detected the total ratio of Annexin PE+/7-AAD- and Annexin PE+/7-AAD+ cells. The apoptotic ratio of pEZ-LV203-Csseverin PLC cells was 9.85%, obviously lower than that of the control cells (pEZ-LV203 PLC and PLC), which showed 32.1% and 34.51%, respectively (Fig 2).
To determine the apoptotic pathways involved in the Csseverin-suppressed early wave of apoptosis, we further explored changes in the activities of initiator caspase (caspase 9) and effector caspase (caspase 3) by Western blot analysis. The results showed the accumulation of cleaved caspase 9 and cleaved caspase 3 in the control groups (pEZ-LV203 PLC and PLC), while expression levels of cleaved caspase 9 and cleaved caspase 3 were decreased in pEZ-LV203-Csseverin PLC cells (Fig 3, P < 0.05).
The opening of the MPTP marks the irreversible point of cell apoptosis [21]; therefore, we examined whether the MPTP participates in the anti-apoptotic mechanism induced by Csseverin. TMRM revealed significantly enhanced red fluorescence intensity in Csseverin pEZ-LV203-Csseverin PLC cells compared with the control group (pEZ-LV203 PLC and PLC) (Fig 4A). The geometric mean, indicating the average red fluorescent intensity of pEZ-LV203-Csseverin PLC、pEZ-LV203 PLC or PLC cells emitting red fluorescence, was 4.82、3.58 and 3.42 (Fig 4B), respectively, suggesting decrease in MPTP opening.
In a previous study, we showed that Csseverin binds to Ca2+ in vitro. Since Ca2+ has been demonstrated as a key substrate associated with apoptosis in different cell types [22], and MPTP has been recognized as a major target of Ca2+[21], we further confirmed whether Csseverin-inhibited apoptosis was associated with Ca2+ imbalance in PLC cells. The cells were stained with the fluorescent probe dihydrorhod-2 AM (Rhod-2 AM) for the analysis of intracellular free calcium. The concentration of intracellular free Ca2+ obviously increased in the control groups, pEZ-LV203 PLC and PLC, while intracellular free Ca2+ was predominantly reduced in pEZ-LV203-Csseverin PLC cells (Fig 5).
The loss of mitochondrial membrane potential induces mitochondrial permeability by opening the MPTP, which primarily initiates the translocation of the apoptogenic protein Cyt c from mitochondria into the cytoplasm[23]. Subcellular fractionation was performed to examine Cyt c levels in both cytosolic and mitochondrial compartments. Compared to those of the control groups, the significant downregulation of cytoplasmic Cyt c expression and the upregulation of mitochondrial Cyt c expression in pEZ-LV203-Csseverin PLC cells were observed, indicating an inhibitory effect on the release of Cyt c from mitochondria into the cytoplasm (Fig 6A, P < 0.05).
The mitochondrial translocation of Bax is a key step that prompts the release of Cyt c from the mitochondria[24]. Western blot analysis showed that compared with control cells (pEZ-LV203 PLC and PLC), Csseverin overexpression PLC cells (pEZ-LV203-Csseverin PLC) showed a drastic reduction in the translocation of Bax to mitochondria (Fig 6B, P < 0.05).
Previous studies have shown that Csseverin could induce apoptotic inhibition in spontaneously apoptotic human HCC PLC cells. In the present study, we confirmed the anti-apoptotic role of Csseverin and explored the involved mechanisms. We generated stably Csseverin-overexpressing PLC cells (PEZ-LV203-Cssevein PLC) and control cells (PEZ-LV203 PLC) in the present study. The results demonstrated significant suppression during the early period of apoptosis in pEZ-LV203-Csseverin PLC cells compared with pEZ-LV203-PLC and PLC cells.
Apoptosis occurs via two different pathways: the extrinsic pathway (death receptors) and the intrinsic pathway (mitochondria and endoplasmic reticulum)[15–16]. In a previous study, we observed that Csseverin led to the recovery of mitochondrial membrane potential (MMP) in PLC cells and speculated that the mitochondrial signal pathway may be involved in Csseverin-mediated protection from apoptosis.
Mitochondria are sensitive to the external environment, responding with MMP alterations that lead to the release of apoptosis-related factors and cell apoptosis[25]. There are several specific proteins in the mitochondrial-mediated pathway. Caspase 3 and caspase 9 are the key factors associated with the mitochondrial-mediated pathway. Caspase 9 activity is primarily dependent on the intrinsic pathway (mitochondrial-mediated) regulated by members of the Bcl-2 family[26]. In the present study, compared with control cells (PLC and pEZ-LV203 PLC), we observed a decrease in caspase 9 activity in spontaneously apoptotic pEZ-LV203-Csseverin PLC cells. The reduced activation of caspase 9 subsequently suppressed downstream caspase 3, which was activated through the mitochondrial-mediated pathway. Therefore, these results suggested that via intrinsic (mitochondrial-mediated), extrinsic (death receptors) or other intrinsic (endoplasmic reticulum) pathways, Csseverin might confer protection from the early wave of apoptosis in PLC cells.
Bax, a proapoptotic member of the Bcl-2 family proteins, is an initiator in the mitochondrial-mediated pathway[27]. In healthy living cells, Bax is predominantly located in the cytosol and migrates to the mitochondrial membrane during early apoptosis[28]. This translocation induced Cyt c release from mitochondria to the cytoplasm [29]. Cyt C can combine with procaspase 9 and Apaf-1 to form an apoptosome to activate caspase-9 and other caspases that induce the downstream caspase cascade. We detected the mitochondrial translocation of Bax and the release of Cyt c. The present study showed that the overexpression of Csseverin significantly suppressed the mitochondrial translocation of Bax, followed by the decreased release of Cyt c from mitochondria.
The gelsolin family (include Csseverin) plays a leading role in controlling actin filament reorganization/remodeling[12]. In several models of cell apoptosis, gelsolin has demonstrated an anti-apoptotic property associated with its effects on the dynamic actin cytoskeleton by preventing the loss of mitochondrial membrane potential and activation of caspase 3[30–31]. The organization/remodeling of actin filaments can also release Ca2+ from the F-actin store and open the influx pathway for the external release of Ca2+ into the cell[32]. Intracellular Ca2+ is used as a second messenger to regulate most crucial biological processes, such as cell survival, proliferation and gene transcription[33]. In some experimental systems, the elevation of intracellular Ca2+ levels is regarded as a pivotal element of apoptosis[34–35]. Thus, the Rhod-2 AM Ca2+ fluorophore, which emits red fluorescence, was used to evaluate changes of intracellular Ca2+. Previous studies have shown that Csseverin binds to Ca2+ and cytoskeletal actin filaments [6]. The results of the present study showed a significant decrease of intracellular calcium in Csseverin overexpression PLC cells, associated with the effect of Csseverin on apoptosis suppression.
The intracellular Ca2+ level is affected by mitochondrial Ca2+ sequestration, which might eventually stimulate the prolonged opening of the MPTP. MPTP is a multi-protein complex formed between mitochondrial membranes, and persistent MPTP opening results in the osmotic dysregulation of the mitochondrial membrane[36]. Once the MPTP is opened, various apoptosis-related proteins, such as Bax, could enter mitochondria and lead to a decrease of the mitochondrial membrane potential, the release of Cyt c, and the induction of early apoptosis[37]. We also measured the changes in MPTP using a TMRM probe. Compared with PLC and pEZ-LV203 PLC cells (negative control), enhanced fluorescence intensity was observed in pEZ-LV203-Csseverin PLC cells after induced spontaneous apoptosis by serum-starvation for 48 h, indicating the inhibition of MPTP opening.
Collectively, these data indicated that Csseverin can reduce calcium-mediated MPTP opening, which may be mediated through binding to actin and Ca2+. The inhibition of MPTP opening subsequently suppressed the translocation of Bax to mitochondria and the release of Cyt c from mitochondria, which in turn downregulates caspase 9 activities and caspase 3 protein expression, inducing obvious apoptotic suppression (Fig 7).
Taken together, these findings will be helpful to further illuminate the mechanism involved in tumorigenesis induced by C. sinensis infestation. Whether interventions according to this pathway are effective for the control of the disease progression is worthy of further exploration.
|
10.1371/journal.ppat.1002969 | Rhinovirus Attenuates Non-typeable Hemophilus influenzae-stimulated IL-8 Responses via TLR2-dependent Degradation of IRAK-1 | Bacterial infections following rhinovirus (RV), a common cold virus, are well documented, but pathogenic mechanisms are poorly understood. We developed animal and cell culture models to examine the effects of RV on subsequent infection with non-typeable Hemophilus influenzae (NTHi). We focused on NTHI-induced neutrophil chemoattractants expression that is essential for bacterial clearance. Mice infected with RV1B were superinfected with NTHi and lung bacterial density, chemokines and neutrophil counts determined. Human bronchial epithelial cells (BEAS-2B) or mouse alveolar macrophages (MH-S) were infected with RV and challenged with NHTi, TLR2 or TLR5 agonists. Chemokine levels were measured by ELISA and expression of IRAK-1, a component of MyD88-dependent TLR signaling, assessed by immunoblotting. While sham-infected mice cleared all NTHi from the lungs, RV-infected mice showed bacteria up to 72 h post-infection. However, animals in RV/NTHi cleared bacteria by day 7. Delayed bacterial clearance in RV/NTHi animals was associated with suppressed chemokine levels and neutrophil recruitment. RV-infected BEAS-2B and MH-S cells showed attenuated chemokine production after challenge with either NTHi or TLR agonists. Attenuated chemokine responses were associated with IRAK-1 protein degradation. Inhibition of RV-induced IRAK-1 degradation restored NTHi-stimulated IL-8 expression. Knockdown of TLR2, but not other MyD88-dependent TLRs, also restored IRAK-1, suggesting that TLR2 is required for RV-induced IRAK-1 degradation.
In conclusion, we demonstrate for the first time that RV infection delays bacterial clearance in vivo and suppresses NTHi-stimulated chemokine responses via degradation of IRAK-1. Based on these observations, we speculate that modulation of TLR-dependent innate immune responses by RV may predispose the host to secondary bacterial infection, particularly in patients with underlying chronic respiratory disorders.
| Rhinovirus (RV) is responsible for the majority of common colds. RV infection is also associated with hospitalizations for lower respiratory tract illness, a significant proportion of which are accompanied by bacterial infections including acute otitis media, sinusitis and pneumonia. However, the mechanisms by which RV increases susceptibility to secondary bacterial infections are not understood. In this report, we demonstrate for the first time that RV infection promotes bacterial persistence of non-typeable Hemophilus influenzae (NTHi) in vivo, which was associated with reduced expression of neutrophil-attracting chemokines and neutrophil infiltration into the lungs. Further, RV infection attenuated NTHi or TLR2 or −5 agonist-stimulated chemokine responses in cultured bronchial epithelial cells and alveolar macrophages, suggesting that RV interferes with TLR-related innate immune responses. Next, we found that RV infection caused rapid degradation of IRAK-1, a key adaptor protein in the MyD88-dependent signaling. Inhibition of IRAK-1 degradation restored NTHi-stimulated chemokine responses in RV-infected bronchial epithelial cells. Finally, reductions in IRAK-1 were dependent on TLR2. Together, our results suggest that RV may increase the risk of acquiring secondary bacterial infection by attenuating TLR-dependent innate immune responses.
| Respiratory infection by one pathogen can alter the innate immunity to unrelated pathogens long after resolution of the first infection. This can affect the pathogen clearance and increase disease severity [1]. Severe illness or death due to bacterial infection following viral influenza is one of the most well-documented instances of this phenomenon [2]. Secondary bacterial infections (otitis media, sinusitis, pneumonia) have also been demonstrated following infection with respiratory syncytial virus, enterovirus and rhinovirus (RV) [3], [4], [5]. In addition, RV triggers exacerbations of chronic obstructive pulmonary disease and cystic fibrosis, conditions in which the airways are chronically colonized with bacteria [6], [7].
Most studies examining this problem to date have been focused on understanding the molecular mechanisms by which influenza virus predisposes host to secondary bacterial infection. Influenza is a lytic virus which causes extensive damage to the airway epithelium, thereby increasing exposure of bacterial receptors and inducing apoptotic cell death of macrophages and neutrophils [8], [9], [10]. IFN-γ production during influenza infection decreases the phagocytic and anti-bacterial capacity of alveolar macrophages [11]. Immunosuppressive cytokines such as IL-10 and TGF-β produced after influenza virus infection have been proposed to modify the initial chemokine response to subsequent bacterial infections [12], [13]. Glucocorticoids induced by influenza virus suppress pro-inflammatory cytokine responses to secondary bacterial challenge [14]. In addition to immunosuppressive molecules, antiviral proteins (such as IFNs) produced during viral infections attenuate initial KC/CXCL-1 and MIP2/CXCL2 responses to secondary pneumococcal challenge, resulting in increased persistence of bacteria and death in mouse models of infection [15]. Influenza virus also desensitizes TLR receptors in vivo and decreases pro-inflammatory cytokine responses to bacterial ligands long after the viral infection resolves [16].
Unlike influenza virus, RV does not cause excessive cell damage. Yet, RV infection has been shown to precede otitis media and acute lower respiratory tract infections requiring hospitalization, each of which are associated with bacterial infection [3], [4], [5]. A handful of studies have demonstrated that RV infection enhances bacterial adherence by increasing the expression of host molecules that serves as receptors for bacteria, such as ICAM-1, platelet-activating factor receptor and carcinoembryonic antigen-related cell adhesion molecule [17], [18]. RV infection was also shown to promote internalization of S. aureus into non-fully permissive lung epithelial cells [19]. In addition, RV infection disrupts barrier function and promotes transmigration of bacteria across the polarized airway epithelium [20], [21]. RV was recently shown to attenuate cytokine responses to subsequent challenges with two bacterial products, LPS and lipoteichoic acid, in alveolar macrophages [22]. However, the consequences of such RV-induced chemokine suppression on subsequent bacterial infection have not been demonstrated in vivo or in vitro. Further, the mechanism by which RV infection suppresses cytokine responses to subsequent TLR stimulation has not been determined.
In the present study, we show for the first time that RV promotes persistence of non-typeable Hemophilus influenzae (NTHi) by suppressing neutrophil-attracting chemokine responses. We also demonstrate that RV suppresses NTHi-induced IL-8 expression in airway epithelial cells and alveolar macrophages by inducing TLR2-dependent degradation of IRAK-1.
Major group rhinovirus, such as RV39, which binds to ICAM1 does not infect murine cells due to species specific variations in the ICAM-1 D1 extracellular Ig domain [23]. Previously, we have demonstrated the feasibility of infecting mice with RV1B, a minor group virus, which binds to low-density lipoprotein family receptors [24]. Therefore, in these experiments we used minor group virus, RV1B. Mice were infected with sham or RV1B by the intranasal route and two days later superinfected with NTHi by the intratracheal route. Chemokine expression and bacterial load in the lung were assessed 6 h and 1, 3 and 7 days post-NTHi infection. Although, there was no difference in the lung bacterial load between sham/NTHi and RV1B/NTHi groups at 6 and 24 h post-NTHi infection (Figure 1A), RV1B/NTHi group showed significantly less airway and interstitial neutrophils than sham/NTHi group at these time points (Figure 1B and 1C). While mice in sham/NTHi group cleared all bacteria by 72 h post-infection, RV/NTHi-infected animals showed bacteria in their lungs at low levels which were associated with increased number of airway and interstitial neutrophils. By 7 days post-NTHi infection, RV/NTHi-infected animals cleared all bacteria from their lungs and showed neutrophils counts similar to uninfected animals. Compared to sham-infected mice, RV-, sham/NTHi- and RV/NTHi-infected animals showed significant increases in both airway and interstitial lymphocyte counts 3 and 7 days post-NTHi infection (Supplemental Figure S1A and S1B). However, there was no difference between RV, sham/NTHi and RV/NTHi groups. Only the RV/NTHi group showed a significant increase in the number of macrophages/monocytes 3 and 7 days post-NTHi infection compared to sham-infected mice (Supplemental Figure S1C and S1D).
Histologic evaluation of lung sections, revealed diffuse neutrophilic inflammation in both the sham/NTHi and RV/NTHi groups 24 h post-NTHi infection, but the latter group showed comparatively less inflammation (Figure 2A and 2C). While sham/NTHi-infected animals resolved inflammation completely by 3 days post-NTHi infection, mice in the RV/NTHi group showed a persistence of neutrophilic inflammation (Figure 2B and 2D). RV/NTHi-infected mice showed a resolution of lung inflammation 7 days post-NTHi infection, correlating with clearance of bacteria from the lungs (Figure 2E).
Next we measured the levels of neutrophil-attracting chemokines in the lung homogenate supernatants. The RV/NTHi group showed a significant reduction in chemokine levels of KC/CXCL1 and MIP-2/CXCL2 compared to the sham/NTHi group (Figure 2F and 2G) at 6 and 24 h post-NTHi infection. However at 72 h post-NTHi infection, KC and MIP-2 levels were higher in the RV/NTHi group than the sham/NTHi group. Both KC and MIP-2 levels returned to normal levels by 7 days in both the sham/NTHi and RV/NTHi groups. These results imply that RV infection may suppress initial chemokine responses to subsequent bacterial challenge thereby reducing the neutrophil infiltration required for optimal bacterial clearance.
To understand the underlying mechanisms by which RV suppresses chemokine responses to subsequent NTHi infection, we performed in vitro studies using bronchial epithelial cells and confirmed key results in mouse alveolar macrophages. A human bronchial epithelial cell line (BEAS-2B cells) was infected with RV39 (a major group RV), RV1B or sham. Cells were then infected with NTHi and IL-8 protein in the cell culture supernatants was measured (Figure 3A and 3B). As observed previously, infection with RV or UV-RV stimulated IL-8 production in airway epithelial cells [25]. There was no difference between cells treated with RV and UV-RV. We have previously shown that RV binding/endocytosis is sufficient for stimulation of IL-8 expression in airway epithelial cells during the early phase of infection [25]. Cells infected with sham followed by NTHi showed significantly higher IL-8 levels compared to cells infected with sham or RV alone. In contrast, cells infected with RV39 or RV1B followed by NTHi showed IL-8 levels that were not significantly different from cells infected with RV alone. RV/NTHi-infected cells also showed significantly less IL-8 than sham/NTHi-infected cells, indicating that prior infection with RV suppresses NTHi-induced IL-8 production. Replication-deficient UV-irradiated RV had a similar effect. This was not a result of increased cell death, as there was no difference in LDH release or Annexin V/propidium iodide staining between sham/NTHi- and RV/NTHi-infected cells (data not shown).
To examine whether RV infection suppresses NTHi-induced IL-8 expression in well-differentiated primary airway epithelial cells, we infected these cells apically with RV39 or RV1B, incubated for 24 h and then infected with NTHi. Levels of IL-8 in the basolateral media were determined 6 h after NTHi infection. RV/NTHi-infected cells showed lower levels of IL-8 than the sham/NTHi-infected cells indicating that RV suppresses NTHi-stimulated IL-8 in primary cells (Figure 3C and 3D). Cells infected with UV-irradiated RV39 or RV1B also suppressed the IL-8 response to subsequent NTHi infection.
To investigate whether the suppressed IL-8 response to NTHi in RV-infected bronchial epithelial cells was due to decreased NTHi binding, we determined the binding of FITC-labeled NTHi to sham- or RV39-infected cells. Compared to uninfected cells, both sham/NTHi and RV39/NTHi infected cells showed a significant shift of histogram to the right and increased mean fluorescence intensity (MFI) indicating NTHi binding (Figure 3E and Figure 3F). Compared to sham/NTHi infected cells, RV39/NTHi infected cells showed further rightward shift of histogram and increased MFI, implying that prior infection with RV39 increases binding of bacteria to the cells. These results indicate that the suppressed IL-8 response in RV/NTHi-infected cells is not due to decreased bacterial binding.
Bacterial recognition by TLR2 and 5 contributes to IL-8 production in bronchial epithelial cells [26]. Therefore we examined whether RV suppresses TLR2- or TLR5-induced IL-8 responses in bronchial epithelial cells. BEAS-2B cells infected with sham and then challenged with a TLR2 (Pam3CSK4) or TLR5 (flagellin) agonist showed significant increases in IL-8 levels compared to sham alone infected cells (Figure 4A). Cells infected with RV39 followed by TLR2 or TLR5 challenge showed significantly lower IL-8 responses compared to similarly-challenged sham-infected cells.
RV-induced reductions in TLR-mediated IL-8 expression may relate to either reduced expression of TLRs or disruption of MyD88-dependent signaling, which is common for both TLR2 and 5. To assess the effect of RV infection on TLR expression, sham- or RV-infected cells were incubated with normal IgG or antibodies specific to TLR2 or TLR5 and analyzed by flow cytometry. BEAS-2B cells expressed both TLR2 and 5 on their surface, which were not significantly altered in RV39-infected cells (Figure 4B to 4D). These results suggest that suppression of TLR2 or TLR5 agonist-stimulated IL-8 in RV39-infected cells is not due to reduced TLR expression.
IRAK-M (also known as IRAK-3) is a well-described inhibitor of MyD88-dependent TLR signaling [27]. To investigate the possible role of IRAK-M in the suppression of NTHi-, or TLR2- or TLR5 agonist-induced IL-8, IRAK-M expression was assessed in RV39-infected BEAS-2B cells. RV significantly increased IRAK-M levels compared to sham controls (Figure 5A and 5B), suggesting that IRAK-M may be responsible for the observed suppression of NTHi- or TLR2 or 5 agonist-induced IL-8 expression in RV- infected cells. To examine this possibility, we transfected BEAS-2B cells with non-targeting or IRAK-M siRNA and then infected with either sham or RV. Western blot analysis showed that compared to NT–siRNA, IRAK-M siRNA transfection reduced IRAK-M expression by 83 and 77% in sham- and RV-infected cells respectively, an indication of efficient knockdown of IRAK-M in these cells (Figure 5C). However, knockdown of IRAK-M did not reverse the suppression of NTHi-stimulated IL-8 in RV-infected cells (Figure 5D). Similar results were observed when cells were challenged with TLR2 or TLR5 agonists instead of infecting with NTHi (data not shown). These data suggest that although RV increases IRAK-M expression, factors other than IRAK-M contribute to RV-induced suppression of NTHi-stimulated IL-8 response in bronchial epithelial cells.
It was recently shown that IL-1β and MyD88 are partially required for RV-induced IL-8 expression in BEAS-2B cells [28]. IRAK-1 undergoes proteosomal degradation following activation of the IL-1/TLR signaling pathway, thereby inducing a state of tolerance to subsequent TLR stimulation [29]. It is therefore conceivable that RV infection induces degradation of IRAK-1, leading to the attenuation of TLR-mediated IL-8 production, irrespective of IRAK-M status. To test this hypothesis, we determined the levels of IRAK-1 in BEAS-2B cells infected with either sham, RV39 or UV-irradiated RV39. Infection with RV39 and UV-RV39 each decreased IRAK-1 protein levels significantly 24 h after incubation (Figure 6A and 6B). RV39 infection decreased IRAK-1 levels as early as 4 h post-infection (Figure 6C and 6D). UV-RV39 also reduced IRAK-1 protein levels, indicating that the changes in IRAK-1 protein abundance do not require viral replication. Similar decreases in IRAK-1 were observed in BEAS-2B cells infected with RV1B (Figure 6E and 6F) and primary airway epithelial cells infected with RV39 or RV1B (Figure 6G and 6H).
To determine whether IRAK-1 is required for NTHi-induced IL-8 expression, BEAS-2B cells transfected with non-targeting or IRAK-1 siRNA were infected with NTHi and the IL-8 response was determined 3 h later. Immunoblotting showed complete knockdown of IRAK-1 in cells transfected with IRAK-1 but not in non-targeting siRNA (Figure 7A). NTHi-induced IL-8 was significantly reduced in cells transfected with IRAK-1 siRNA compared to NT-siRNA-transfected cells (Figure 7B). These results imply that IRAK-1 is required for NTHi-induced IL-8 expression, consistent with the notion that reduced IRAK-1 protein levels contribute to RV-induced suppression of NTHi-induced IL-8 expression.
To determine whether inhibition of IRAK-1 degradation in RV-infected cells restores NTHi-stimulated IL-8 expression, we infected cells with RV39 in the presence of proteosomal inhibitor lactacystin and examined IRAK-1 and IL-8 levels. In DMSO-treated cells (vehicle control), RV39 infection decreased IRAK-1 protein levels significantly (Figure 7C and 7D) and also suppressed NTHi-stimulated IL-8 (Figure 7E). By contrast, pre-treatment with lactacystin (5 µM) prevented RV-mediated reductions in IRAK-1 protein abundance and NTHi-induced IL-8 expression. These results show that IRAK-1 degradation caused by RV contributes to the reduced IL-8 response to subsequent NTHi infection.
Since RV has been shown to activate MyD88-dependent signaling via IL-1β [28], we hypothesized that IL-1β is required for the suppression of NTHi-stimulated IL-8 in RV-infected cells. To test this hypothesis, BEAS-2B cells were infected with sham or RV in the presence or absence of IL-1ra. Cells were then infected with NTHi and IL-8 measured in the cell culture supernatant. IL-1ra did not reverse the suppression of NTHi-stimulated IL-8 caused by RV, even at high concentration (Supplemental Figure S2A). In addition, IL-1ra failed to restore the levels of IRAK-1 in RV-alone infected cells (Supplemental Figure S2B). Together, these results suggest that RV-induced IRAK-1 degradation is triggered independently of IL-1β.
We examined the contribution of the MyD88-dependent TLRs TLR2, −4, −5, −7 and −8 to RV-induced IL-8 expression. BEAS-2B cells were transfected with non-targeting siRNA or gene specific siRNA to TLR2, TLR4, TLR5, TLR7 or TLR8 and infected with RV39. IL-8 mRNA and protein levels were determined 24 h post-infection. TLR2 siRNA-transfected cells infected with UV-RV or RV showed significantly lower IL-8 mRNA and protein than similarly-infected non-targeting siRNA-transfected cells (Figures 8A and 8B). TLR8 siRNA-transfected cells showed a small (20%) but significant reduction in RV-induced IL-8 mRNA, but not protein levels. Knockdown of TLR4 TLR5, or TLR7 had no effect on RV or UV-RV-induced IL-8 responses either at the mRNA or at protein levels. Next we assessed the knockdown of each TLR by quantifying mRNA by qPCR (Figure 8C) and protein by flow cytometry (Supplemental Figure S3). BEAS-2B cells showed expression of TLR 2, −4, and −5 at both the mRNA and protein level, and transfection of gene-specific siRNA knocked down the expression. BEAS-2B cells did not express TLR7. Cells expressed TLR8 at very low levels and expression was completely inhibited in cells transfected with TLR8 siRNA.
Since only knockdown of TLR2 and TLR8 reduced RV-induced IL-8 responses, we examined the levels of IRAK-1 protein in TLR2 or TLR8 siRNA-transfected cells after RV infection. Compared to non-targeting siRNA-transfected cells infected with RV39, similarly-infected TLR2 siRNA-transfected cells showed IRAK-1 levels similar to sham-infected cells (Figure 8D and 8E). In contrast TLR8 siRNA transfected cells infected with RV showed IRAK-1 degradation similar to cells transfected with NT siRNA. These results imply that TLR2, but not TLR8 is required for the RV-induced reduction in IRAK-1 protein levels.
To assess whether TLR2 or TLR8 knockdown restores NTHi-stimulated IL-8 expression in RV-infected cells, cells transfected with non-targeting, TLR2 or TLR8 siRNA were infected with sham or RV39 and then infected with NTHi. Not surprisingly, TLR2 knockdown reduced NTHi-stimulated IL-8 (Supplemental Figure S4), as NTHi-induced IL-8 expression is TLR2-dependent [30]. On the other hand, TLR8 knockdown had no effect on NTHi-stimulated IL-8 and did not reverse the suppressive effect of RV on NTHi-stimulated IL-8.
RV infection has been shown to desensitize TLR4-dependent signaling in alveolar macrophages [22]. We therefore assessed the effects of RV1B infection on NTHi-stimulated cytokine expression and IRAK-1 protein levels in MH-S mouse alveolar macrophages. MH-S cells were infected with sham, UV-RV1B or RV1B and incubated for 24 h. Cells were then infected with NTHi and levels of KC, MIP-2 and TNF-α in the medium were determined 6 h later. Both UV-RV1B/NTHi and RV1B/NTHi-infected cells showed significantly less KC, MIP-2 and TNF-α than the sham/NTHi-infected cells (Figure 9A–9C). On a similar note, RV1B-infected cells also showed decreased KC and TNF-α production in response to secondary challenge with TLR2 (Figure 9D–9F) or TLR5 agonists (Figure 9G–9I). Compared to sham-, UV-RV1B and RV1B-infected cells also showed reduced IRAK-1 protein levels (Figure 10A and 10B), indicating that RV-induced IRAK-1 degradation may be responsible for reduced cytokine production in alveolar macrophages stimulated with bacteria or TLR agonists.
In this study, we demonstrate for the first time that RV infection delays bacterial clearance in vivo. Delayed bacterial clearance was accompanied by attenuated expression of the neutrophil-attracting chemokines KC/CXCL1 and MIP-2/CXCL2. We also show that, despite increasing bacterial binding, RV suppresses chemokine responses to subsequent NTHi infection in cultured bronchial epithelial cells. RV infection also attenuated TLR2- and TLR5-stimulated chemokine responses in cultured bronchial epithelial cells and alveolar macrophages, suggesting that RV infection interferes with TLR-dependent innate immune defenses. Most importantly, we define a mechanism by which RV infection desensitizes TLR signaling in both bronchial epithelial cells and alveolar macrophages. We demonstrate that RV infection causes degradation of IRAK-1, a key adaptor protein in MyD88-dependent TLR signaling, thereby suppressing chemokine responses to subsequent NTHi challenge. Further, we show that IRAK-1 degradation was dependent on TLR2, but not on other TLRs which activate MyD88-dependent signaling. Based on these observations, we speculate that suppression of TLR-dependent innate immune responses increases the risk of acquiring secondary bacterial infection and promotes invasiveness of bacterial flora that is present in the nasopharynx/lower airways of CF and COPD patients.
Although there is clinical evidence indicating that RV infection precedes bacterial infection-associated disease exacerbations [3], [4], [5], until now there were no experimental data demonstrating an increased risk of bacterial infection following RV infection in vivo. Using a novel mouse model of RV infection that we developed [24], [31], we found that RV infection promotes NTHi persistence in the lungs, likely by attenuating KC and MIP-2 expression and impairing recruitment of neutrophils to the airways. Appropriate neutrophil recruitment is required for bacterial clearance, particularly under circumstances that exceed the antimicrobial capacity of airway epithelial cells and alveolar macrophages. Mice with a delayed neutrophil response fail to clear P. aeruginosa efficiently from their lungs [32]. Recently, influenza virus was also shown to suppress KC and MIP-2 responses to subsequent bacterial infection; suppression was dependent on virus-stimulated type I interferons [15]. Since RV infection also stimulates type I interferon responses both in vivo and in airway epithelial cells in vitro [24], [33], [34], [35], [36], it is plausible that RV suppresses chemokine responses via this mechanism. However, this is unlikely, because replication-deficient UV-irradiated RV, which does not stimulate expression of type I interferons, also attenuated chemokine response to NTHi infection in vitro.
Both bronchial epithelial cells and alveolar macrophages express TLRs. TLRs play a crucial role in the establishment of appropriate innate immune responses, including expression of chemokines, upon recognition of pathogen associated molecular patterns. Previously, RV infection was shown to suppress subsequent cytokine responses to LPS, a TLR4 ligand in human alveolar macrophages [22]. Suppression was dependent on viral replication. In contrast, we found that attenuation of TLR-dependent signaling in both bronchial epithelial cells and alveolar macrophages was independent of viral replication. These data suggest that suppression is initiated by an early event in the viral life cycle, for example, binding or endocytosis. This is not surprising, because RV binding itself has been shown to activate signaling mechanisms leading to phosphatidylinositol 3-kinase activation, IL-8 expression and generation of reactive oxygen species [25], [37].
In the present study, we found that RV infection not only attenuated chemokine expression due to subsequent challenge with NTHi infection, but also to TLR2 and TLR5 ligands. NTHi stimulates TLR2 signaling leading to pro-inflammatory cytokine expression [30]. In addition, both TLR2 and TLR5 initiate a MyD88-dependent signaling pathway leading to chemokine and pro-inflammatory cytokine expression. We therefore hypothesized that RV attenuates MyD88-dependent TLR signaling. Accordingly, we found that RV infection significantly increased IRAK-M while dramatically reducing the levels of IRAK-1 protein in both airway epithelial cells and murine alveolar macrophages. IRAK-M is thought to be a negative regulator of MyD88-dependent TLR signaling [38]. However, genetic silencing of IRAK-M did not restore NTHi-induced chemokine expression in RV-infected airway epithelial cells, suggesting that IRAK-M does not contribute to RV-induced attenuation of IL-8 expression. This is not entirely unexpected, as IRAK-M has been demonstrated to play a vital role in the suppression of a non-canonical TLR signaling pathway that relies on activation of NF-κB-inducing kinase rather than IκB kinase [39].
In contrast, blocking of RV-induced reduction in IRAK-1 by using a proteasomal inhibitor restored NTHi-induced IL-8 responses in RV-infected bronchial epithelial cells, indicating a role for IRAK-1. Previously, it has been shown that post-translational modification of IRAK-1, such as hyperphosphorylation and ubiquitination that occurs following MyD88-dependent TLR or IL-1 signaling, leads to IRAK-1 degradation [29]. IRAK-1 degradation constitutes a mechanism for limiting exaggerated innate immune responses. In the present study, we found that, in vitro, IRAK-1 is degraded as early as 4 h post-RV infection and the IRAK-1 protein levels remain low for at least 24 h. Pre-treatment with lactacystin prevented reductions in IRAK-1 protein abundance, suggesting that RV promotes proteasomal degradation of IRAK-1. Reduced IRAK-1 was also observed in mice infected with RV 48 h post-infection. These observations suggest that RV induces IRAK-1 degradation via the activation of MyD88-dependent IL-1/TLR signaling cascades. However, since lactacystin can also inhibit proteasomal degradation of other proteins of TLR pathway, further studies using IRAK-1 mutants which are not amenable to ubiquitation and proteasomal degradation are necessary to further confirm the specific role of IRAK-1 in RV-induced suppression of NTHi-stimulated chemokine responses.
Recently, RV infection of BEAS-2B cells was shown to elicit low levels of IL-1β release, which in turn induced MyD88-dependent potentiation of IL-8 production [28]. However, in the present study, inhibition of IL-1 receptor by IL-1ra neither blocked RV-induced degradation of IRAK-1 nor reversed suppression of NTHi-stimulated IL-8, suggesting that the IRAK-1 degradation that we observed is not due to RV-induced IL-1β signaling.
RV single-stranded RNA may activate MyD88-dependent TLR signaling directly via TLR7 and TLR8. In our studies we found that airway epithelial cells do not express TLR7 and express TLR8 at very low levels. Further, knockdown of TLR8 neither prevented RV-induced degradation of IRAK-1 nor reversed suppressive effects of RV on NTHi-stimulated IL-8. Similar to our results, recently it TLR7/8 agonist, imiquimod was found to be inefficient in activating airway epithelial cells suggesting that these cells may not express TLR7 or 8 [40]. On the other hand, it has been demonstrated that both intact and UV-irradiated RV-6 stimulate NF-κB transactivation in TLR2-transfected HEK-293 cells [41]. Since we observed that both replication-competent and replication-deficient RV were equally capable of causing IRAK-1 degradation, we posit that TLR2 plays a role in RV-induced IRAK-1 degradation. Consistent with this notion, we found that knockdown of TLR2 significantly reduced RV-induced IRAK-1. However, we could not determine the effect of TLR2 knockdown on suppression of NTHi-stimulated IL-8 in RV-infected cells, because NTHi also requires TLR2 to stimulate IL-8 [30]. Also, the precise mechanism by which RV activates TLR2 remains to be elucidated.
In summary, we have shown that RV suppresses NTHi-induced IL-8 expression in airway epithelial cells and alveolar macrophages by inducing TLR2-dependent degradation of IRAK-1. Our results suggest that RV may increase the risk of acquiring secondary bacterial infection by attenuating TLR-dependent innate immune responses.
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, United States. The protocol was approved by the Institutional Animal Care and Use Committee of the University of Michigan Medical School. All surgery was performed under sodium pentobarbital anesthesia, and all efforts were made to minimize suffering. Tracheal and bronchial trimmings from donor lungs obtained at the time of double lung transplantation (which are otherwise discarded) were used for isolation of airway epithelial cells and the protocol was approved by the University of Michigan Investigational Review Board. Since the tissues were not collected for the purpose of isolation of cells and there was no documentation linking the tissue to the donor, the University of Michigan Investigational Review Board provided waiver for obtaining consent.
RV39 and RV1B was obtained from ATCC and used to generate viral stocks via infection of HeLa cells as previously described [25]. Virus titer was determined by assessing 50% tissue culture infectivity dose (TCID50)/ml. For production of UV-irradiated (UV-RV) RV, samples were irradiated using a CL-1000 crosslinker (UVP, Upland, CA) at 10 mJ/cm2 for 10 min on ice as described previously [33]. RV inactivated by this method is fully capable of binding to epithelial cells and undergoing endocytosis and stimulates IL-8 protein levels similar to intact RV. At the same time it does not stimulate replication-dependent interferons and interferon-dependent genes.
Nontypeable Haemophilus influenzae (6P5H) was isolated from a COPD patient during exacerbation (Kindly provided by T. Murphy, University of Buffalo, Buffalo, NY). Bacteria was maintained as a glycerol stock at −80°C and cultured as described previously [42].
Normal 8 to 10 week-old BALB/C mice were infected with 50 µl of RV1B (1×108 TCID50T/ml) by intranasal route as described previously [24], [31]. Mice were then infected with 40 µl NTHi (1×109 colony forming units/ml) by intratracheal route [43] and sacrificed 6 h, 1 day or 3 days or 7 days post-NTHi infection. Lungs were collected under aseptic conditions, then homogenized in 2 ml sterile PBS. An aliquot of lung homogenate was 10 fold serially diluted and plated on chocolate agar plates to determine bacterial density. Lung homogenates were centrifuged and supernatants used for ELISA. Number of airway and tissue inflammatory cells was determined by counting cells in bronchoalveolar (BAL) fluid and lung digests respectively as described previously [34]. Briefly, after appropriate treatment, BAL was performed by instilling 1 ml PBS containing 5 mM EDTA 10 times. Cytospins prepared from BAL cells were stained with Diff-Quick and differential counts were determined by counting a minimum of 200 cells. To quantify the number of inflammatory cells in the tissues, lungs were digested with type IV collagenase for 1 h, lung digests were strained through 70-µm nylon mesh (BD Biosciences, San Jose, CA) and centrifuged. The cell pellet was treated with RBC lysis buffer (BD Biosciences) and subjected to density gradient centrifugation on 40% Percoll (Sigma-Aldrich, St. Louis, MO) to enrich for leukocytes. The total cell count was determined on a hemocytometer. Cytospins of these tissue leukocytes were stained with Diff-Quick and examined for differential cell counts. In some experiments, after appropriate treatment, lungs were inflation fixed and embedded in paraffin. Five micron thick sections were deparaffinized and stained with hematoxylin and eosin to evaluate lung inflammation.
Human bronchial epithelial cell line (BEAS-2B; American Type Culture Collection, Manassas, VA) were grown in collagen coated plates using bronchial epithelial cell growth medium (BEGM) (Lonza, Walkersville, MD)) as described previously [33]. Primary airway epithelial cells from normal subjects were cultured at air/liquid interface to promote differentiation into mucociliary phenotype [33], [42], [44]. Mouse alveolar macrophages (MH-S) (ATCC) cells were grown in RPMI media amended with L-glutamine and 10% serum.
BEAS-2B or MH-S cells grown to 90% confluence were infected with RV or UV-RV at multiplicity of infection (MOI) of 1 or equal volume of sham (media from uninfected HeLa cells) and incubated for 90 min at 33°C. Infection media was replaced with fresh media and the incubation continued for another 22 h. Cells were then infected with NTHi at MOI of 10 or treated with media alone, centrifuged at 500×g for 5 min and incubated at 37°C for 3 h. Media was collected for determination of IL-8 (for BEAS-2B cells) or KC, MIP-2 and TNF-α (for MH-S cells). In some experiments, cells were infected with RV in the presence of 5 µM lactacystin (Cayman chemicals, Ann Arbor, MI) or 10 to 100 ng/ml IL-1 receptor antagonist (PeproTech, Rocky Hills, NJ). Primary cells were infected apically with RV, UV-RV or sham at 1 MOI [45], incubated for 48 h and then cells were superinfected apically with NTHi at 10 MOI and basolateral media collected for IL-8 and cells for Western blot analysis.
BEAS-2B cells were infected with RV, UV-RV or sham as described above. Cells were then challenged with TLR2 agonist, Pam3CSK4 (InvivoGen, San Diego, CA), or TLR5 agonist flagellin (ENZO Life Sciences, Inc., Farmingdale, NY), incubated for 6 h at 37°C and IL-8 determined in the cell culture supernatant.
After relevant treatment, cells were washed with ice cold PBS and total cell lysates were prepared as described previously [46]. Briefly, cells were lysed in cold RIPA buffer containing complete protease inhibitors (Roche Diagnostics, Indianapolis, IN), lysates centrifuged and total protein determined in supernatants. Lungs harvested from mice were homogenized in 1 ml PBS conataining complete protease inhibitors and mixed with 2× RIPA buffer, sonicated for 3×10 seconds each, centrifuged and total protein determined in supernatants. Aliquots equivalent to equal amounts of protein were subjected to SDS-PAGE and protein transferred to nitrocellulose membrane. Membranes were blocked and probed with antibodies to IRAK-M, IRAK-1 (both from Santa Cruz Biotechnology, Santa Cruz, CA), or β-actin (Sigma-Aldrich, St. Louis, MO). Bound antibody was detected by using appropriate second antibody conjugated with horseradish peroxidase and chemiluminescent substrate. Band densities were normalized to β-actin or relevant total protein using NIH Image-J (NIH, Bethesda, MD).
Binding of bacteria to cells was determined using FITC-labeled NTHi as described previously [47]. Briefly, FITC-labeled NTHi was incubated with BEAS-2B cells infected with Sham, UV-RV or RV and incubated for 60 min. Cells were then washed and analyzed by flow cytometry (Becton-Dickinson, Franklin Lakes, NJ).
After relevant treatment BEAS-2B cells were incubated with blocking buffer followed by monoclonal antibodies to either TLR2, TLR4, TLR5, TLR7 or TLR8 (all from eBioscience, Inc., San Diego, CA) or normal mouse IgG (matched isotype control) conjugated with fluorescein isothiocyanate (FITC). For detection of TLR7 and TLR8 cells were permeabilized with 1% saponin for 30 min on ice prior to incubating with antibodies to TLR7 or TLR8. Cells were then washed and analyzed by flow cytometry [42].
Conditioned basolateral medium from cell cultures or lung homogenate supernatant was used to determine the protein levels of chemokines by ELISA (R&D systems, Minneapolis, MN) as described previously [45], [48].
BEAS-2B cells were reverse transfected with 10 picomoles of non-targeting (NT), or ON-TARGETplus SMART pool siRNA specific to TLR2, TLR4, TLR5, TLR7, TLR8, IRAK-1 or IRAK-M (Dharmacon, Inc., Chicago, IL) and incubated for 48 h. Cells were then infected with sham, RV or NTHi as appropriate (described in results section). Knockdown of gene expression was confirmed by qPCR or Western blot analysis.
Results are expressed as means ± SEM. Data were analyzed by SigmaStat statistical software (Systat Software, Inc., San Jose, CA). One-way analysis of variance (ANOVA) with Tukey-Kramer post-hoc analysis or two-way ANOVA was performed to compare more than two groups. To compare two groups, an unpaired t test with Welch's correction was used. If the data were not distributed normally data were expressed as range with median and analyzed by Mann-Whitney test. A value of p≥0.05 was considered significant.
|
10.1371/journal.pmed.1002019 | Estimating the Risk of Chronic Pain: Development and Validation of a Prognostic Model (PICKUP) for Patients with Acute Low Back Pain | Low back pain (LBP) is a major health problem. Globally it is responsible for the most years lived with disability. The most problematic type of LBP is chronic LBP (pain lasting longer than 3 mo); it has a poor prognosis and is costly, and interventions are only moderately effective. Targeting interventions according to risk profile is a promising approach to prevent the onset of chronic LBP. Developing accurate prognostic models is the first step. No validated prognostic models are available to accurately predict the onset of chronic LBP. The primary aim of this study was to develop and validate a prognostic model to estimate the risk of chronic LBP.
We used the PROGRESS framework to specify a priori methods, which we published in a study protocol. Data from 2,758 patients with acute LBP attending primary care in Australia between 5 November 2003 and 15 July 2005 (development sample, n = 1,230) and between 10 November 2009 and 5 February 2013 (external validation sample, n = 1,528) were used to develop and externally validate the model. The primary outcome was chronic LBP (ongoing pain at 3 mo). In all, 30% of the development sample and 19% of the external validation sample developed chronic LBP. In the external validation sample, the primary model (PICKUP) discriminated between those who did and did not develop chronic LBP with acceptable performance (area under the receiver operating characteristic curve 0.66 [95% CI 0.63 to 0.69]). Although model calibration was also acceptable in the external validation sample (intercept = −0.55, slope = 0.89), some miscalibration was observed for high-risk groups. The decision curve analysis estimated that, if decisions to recommend further intervention were based on risk scores, screening could lead to a net reduction of 40 unnecessary interventions for every 100 patients presenting to primary care compared to a “treat all” approach. Limitations of the method include the model being restricted to using prognostic factors measured in existing studies and using stepwise methods to specify the model. Limitations of the model include modest discrimination performance. The model also requires recalibration for local settings.
Based on its performance in these cohorts, this five-item prognostic model for patients with acute LBP may be a useful tool for estimating risk of chronic LBP. Further validation is required to determine whether screening with this model leads to a net reduction in unnecessary interventions provided to low-risk patients.
| A minority of patients who experience an episode of low back pain develop persistent (chronic) pain.
Offering tests and treatments to all these patients exposes high numbers of low-risk patients to unnecessary intervention, which is very costly and potentially harmful.
A tool to help healthcare practitioners accurately predict whether a patient with a recent episode of low back pain will develop persistent pain stands to greatly reduce the burden of low back pain on the health system and on patients.
We developed a five-item screening questionnaire using study data from 1,230 patients with a recent episode of low back pain.
We tested how well this screening questionnaire could predict the onset of persistent pain in a separate sample of 1,528 patients.
We found that the screening questionnaire could predict the onset of persistent pain with acceptable levels accuracy (area under the receiver operating characteristic curve = 0.66 [95% CI 0.63 to 0.69]; intercept = 0.55, slope = 0.89).
This brief, easy-to-use screening questionnaire could help healthcare practitioners and researchers make an early estimate of a patient’s risk of persistent low back pain.
The screening questionnaire predicted outcome more accurately in patients with low risk scores than in those with high risk scores.
Screening patients with a recent episode of low back pain could reduce the number of unnecessary interventions provided to low-risk patients.
| Low back pain (LBP) is a major global health problem that, compared to all other diseases and health conditions, is responsible for the most years lived with disability, an estimated 80 million years lived with disability in 2010 [1]. The costs of care, investigations, and lost productivity associated with LBP are a significant economic burden for industrialized nations [2]. For example, estimates for treatments alone are US$50 billion per annum in the United States [3] and US$4 billion in the United Kingdom [4]. The impact of LBP can be profound; in Australia, LBP is the leading cause of early retirement [5] and of income poverty in older adults [6].
Although most people with a new episode, or acute, LBP recover in a few weeks or months, around one-quarter of patients who present to primary care develop chronic LBP (pain lasting for longer than 3 mo) [7]. Chronic LBP is the most problematic type of LBP; its prognosis is poor [8], and it accounts for the majority of costs [4,9]. Between 1992 and 2006, the prevalence of chronic LBP in the United States more than doubled [10]. Managing patients with chronic LBP is difficult, and the effects of contemporary interventions are modest at best [11]. An alternative to costly and ineffective management of these patients is secondary prevention, where the goal is to prevent the onset of chronic LBP [12].
An important first step in secondary prevention is to estimate an individual patient’s risk of developing chronic LBP. The Prognosis Research Strategy (PROGRESS) group recently provided a framework for this step, which involves developing and validating prognostic models to determine risk profiles. For these models to be considered clinically useful, they must be easy to use, be able predict outcome with acceptable accuracy, and be validated in external samples. Risk estimates should be well matched to actual outcomes (calibration), higher for individuals who have a poor outcome than for those who do not (discrimination), and informative enough to justify screening compared to “treat all” or “treat none” approaches (net benefit). Estimates from validated models can add valuable information to the clinical decision-making process [13].
Early, accurate prognostic information also provides the opportunity for practitioners to counsel their patients on the necessity of further treatment [14]. Offering tests and treatments to all patients with acute LBP (“treat all” approach) is expensive and risks exposing high numbers of low-risk patients to unnecessary intervention [15]. Overtreatment of conditions such as LBP overburdens healthcare systems and diverts scarce resources away from where they are most needed [16]. Undertreatment of high-risk patients with acute LBP may also be harmful. A “treat none” approach to acute LBP guarantees that a significant proportion will develop chronic LBP and its long-term consequences [7], and wastes an opportunity to intervene early in primary care.
Targeting early intervention according to risk profile has been shown to be effective in breast cancer [17] and cardiovascular disease [18] and has been identified as a research priority for managing LBP [19]. There is preliminary evidence that a stratified approach improves disability in samples with predominantly chronic LBP [20], but it remains unknown whether such a prognostic approach can prevent the onset of chronic LBP. The absence of a valid prognostic model to inform risk-stratified management of acute LBP is therefore an important area of uncertainty [21]. Not having a validated prognostic model for acute LBP is also problematic for secondary prevention trials that are designed to target pain [22,23]; treat all approaches are unlikely to be efficient if the majority of included participants are at low risk of chronic LBP [24,25].
None of the commonly used screening tools in LBP are suited to this purpose. Tools such as the Start Back Tool (SBT) and the Orebro Musculoskeletal Pain Questionnaire (OMPQ) were either developed in samples that included patients with chronic LBP [26] or used to predict disability [26,27] or return to work [28] outcomes. When these tools were subsequently tested in acute LBP samples, they predicted chronic LBP with modest accuracy at best [27,29–31].
The primary aim of this study was to develop and validate a prognostic model to identify risk of chronic LBP in patients with acute LBP. Specifically, we aimed to develop a model that can provide an estimate for an individual patient’s risk of chronic LBP with acceptable levels of accuracy (calibration, discrimination, and net benefit). A secondary aim was to determine whether prognostic models varied by how chronic LBP was defined. Specifically, we aimed to develop two additional prognostic models using outcomes of high pain and chronic disability.
The protocol for this study has been published [32].
We used patient data from a prospective cohort study to develop the model (development sample) and patient data from a randomized trial to externally validate the model (external validation sample). Full details of these two studies have been published [23,33], and their key differences are summarized in Table 1. Both studies were conducted in Sydney, Australia. In short, the cohort study recruited consecutive patients with acute LBP presenting to their primary care provider (general practitioner, physiotherapist, chiropractor) between 5 November 2003 and 15 July 2005. The randomized trial recruited consecutive patients with acute LBP presenting to their primary care practitioner between 10 November 2009 and 5 February 2013 to test the effect of paracetamol on recovery. There was no difference in treatment effects between groups. Both studies followed a published protocol [34,35], and the trial was prospectively registered.
Baseline data were available on 20 predictors in six broad groups of putative prognostic factors that have been identified in previous studies [36–38]: sociodemographic factors, general health, work factors, current LBP characteristics, past LBP history, and psychological factors. Primary care clinicians collected these data at the first consultation. A full list of individual candidate predictors is provided in Table 2.
To develop the primary model, PICKUP (Predicting the Inception of Chronic Pain), we defined the main outcome as whether or not patients had chronic LBP, that is, ongoing LBP 3 mo after the initial consultation. In the development study, pain intensity was measured with a six-point Likert scale [39]. We classified patients as having “chronic LBP” if they reported greater than “mild” (2 on the Likert scale) pain intensity at 3-mo follow-up and had no periods of recovery [40].
To develop two secondary prognostic models (Models 2a and 2b), we used additional criteria to define chronic LBP. These secondary models allowed comparison of model performance to published models and to our primary prognostic model. Patients were classified as having “chronic LBP high pain” if they reported greater than “moderate” (3 on the Likert scale) pain intensity [39] at 3-mo follow-up (Model 2a). Patients were classified as having “chronic LBP disability” if they reported a score of 2 or more on a five-point Likert scale for disability [39] at 3-mo follow-up (Model 2b). Thresholds to define outcomes for all three models were determined a priori [32].
In the external validation sample, pain and disability scores were converted from an 11-point scale used to measure pain intensity and a 24-item scale used to measure disability to the six-point and five-point scales, respectively, used in the development sample. Both of the original studies assessed 3-mo outcomes over the phone, an approach that yields comparable results to in person assessment on pain-related outcomes [41].
The statistical analysis plan for this study was informed by recommendations from the PROGRESS group [13]. All preplanned analyses are outlined in our protocol published a priori [32].
Flow of patients in the development and external validation samples is shown in Fig 1. Eighteen patients (1.4%) in the cohort study (development sample) and 46 patients (2.7%) in the randomized trial (external validation sample) were un-contactable at 3-mo follow-up. Some patients were excluded from the external validation sample because they were not assessed for pain intensity (65 patients; 3.9%) or disability (87 patients; 5.2%) at 3-mo follow-up.
There were five missing predictor values in the development sample and 44 missing predictor values in the external validation sample. We found evidence against the hypothesis that predictor values were not missing completely at random (Little’s test, p > 0.05), and, because the number of missing values was small (<1%), we removed these cases from the primary analysis as per our protocol [32]. Imputing missing predictor values in the sensitivity analysis did not affect the results (S1 and S2 Tables).
Data were therefore available from 1,230 cases to develop the prognostic models. To externally validate the models, data were available from 1,528 complete cases to test PICKUP, 1,525 complete cases to test Model 2a, and 1,504 complete cases to test Model 2b.
Table 3 shows the characteristics of patients in the development and external validation samples. Patients were similar at baseline except for the proportion receiving disability compensation, which was higher in the development sample (18%) than in the external validation sample (7%).
At 3 mo, 30% of the patients in the development sample were classified as having chronic LBP. Table 4 shows predictors and regression coefficients for the primary model (PICKUP) and the two secondary models that were fitted in this sample. PICKUP contained five predictors. We did not detect significant non-linearity in any continuous predictor variables. Estimates for the predictive performance of each prognostic model in the development sample can be found in S2 Table. Recruitment setting (general practice, physiotherapy, chiropractic) did not affect performance estimates (S3 Table).
Table 5 summarizes the predictive performance of the prognostic models in the external sample. At 3 mo, 19% of the patients in the external validation sample were classified as having chronic LBP. The Nagelkerke R2 value was 7.7%, compared to 10.9% in the development sample, and the Brier score was 0.15, indicating a similar overall model fit. S2 Table shows the full results of performance testing for each prognostic model in the development and external validation samples. Discrimination performance for PICKUP fell within our prespecified acceptable range: the AUC was 0.66 (95% CI 0.63 to 0.69), the likelihood ratio in the high-risk group was 2.99 (95% CI 2.81 to 3.18), and the 95% confidence intervals did not overlap with between risk groups (S4 Table).
All models showed some miscalibration in the external validation sample (Fig 2). PICKUP demonstrated the best calibration and fell within our prespecified acceptable range in the lower seven of the ten risk groups, that is, predictions were within 5% of actual proportions of chronic LBP. In all three models, calibration was better for the low-risk patients than it was for the high-risk patients. After recalibration, slope and intercept estimates for each model were close to 1 and 0, respectively, which indicates near perfect calibration (S1–S3 Figs). Updating PICKUP with an additional prognostic factor (sleep quality) did not add significantly to the model (p > 0.10).
Fig 3 shows the results of the decision curve analysis. Treat all strategies assume that if all patients are treated, none will develop an unfavorable outcome. This may or may not be a reasonable assumption in LBP. Although there are effective treatments for acute LBP [54], evidence-based interventions to prevent the onset of chronic LBP are not yet available. The assumed outcome from treating all patients with acute LBP is that all high-risk patients are offered further intervention that could reduce their risk of chronic LBP. The assumed outcome from treating no patients with acute LBP is that all high-risk patients will develop an unfavorable outcome. In our external validation cohort, for example, if no high-risk patients were offered further intervention, one in five would develop chronic LBP.
Treat all strategies demonstrated the highest net benefit at threshold probabilities between 0% and 10%. At thresholds above the population risk (incidence rates were 19% for chronic LBP, 10% for chronic LBP with high pain, and 14% for chronic LBP with disability), the net benefit of treating all became negative (Fig 3). The net benefit of treating none was always assumed to be zero.
All prognostic models showed equal or higher net benefit than the treat all and treat none strategies. Using PICKUP and a cutoff set at 19% (i.e., only patients with a predicted risk higher than the population risk of 19% are recommended further intervention), the net number of cases of chronic LBP that would be detected through screening, without any increase in the number of patients unnecessarily recommended further intervention, would be four in every 100 patients.
Fig 4 shows the estimated net number of unnecessary interventions avoided through screening. Using PICKUP and a cutoff set at 30% (i.e., only patients with a predicted risk of 30% or higher are recommended further intervention) would lead to a net reduction of around 40 unnecessary interventions per 100 patients.
An individual score (ScoreCLBP) can be derived using the recalibrated logistic regression equation from PICKUP:
ScoreCLBP=−0.55 + 0.89*(−2.82 + [0.21*Pain + 0.44*Leg + 0.50*Comp + 0.06*Depress + 0.13*Risk])
where Pain = “How much low back pain have you had during the past week?” 1 = none, 2 = very mild, 3 = mild, 4 = moderate, 5 = severe, 6 = very severe; Leg = “Do you have leg pain?” 0 = no, 1 = yes; Comp = “Is your back pain compensable, e.g., through worker’s compensation or third party insurance?” 0 = no, 1 = yes; Depress = “How much have you been bothered by feeling depressed in the past week (0–10 scale)?” 0 = not at all, 10 = extremely; Risk = “In your view, how large is the risk that your current pain may become persistent (0–10 scale)?” 0 = none, 10 = extreme.
The predicted risk of developing chronic LBP (ProbCLBP) can then be calculated using the score and the following equation:
ProbCLBP = exp(ScoreCLBP)/(1 + exp[ ScoreCLBP ])
We have developed and tested the external validity of a prognostic model to identify the risk of chronic LBP in individuals with acute LBP. Values for discrimination and calibration fell within a prespecified [32] range of what we subjectively determined to be informative. Although the AUC values are modest (between 0.66 and 0.69), they suggest better predictive accuracy for pain outcomes than recently published values based on either clinician judgment alone (between 0.50 and 0.60) [29] or popular tools such as the SBT and OMPQ [29–31]. The results of our decision curve analysis indicate that, compared to treat all and treat none strategies, our model has the potential to substantially reduce harms associated with undertreating high-risk patients and overtreating low-risk patients with acute LBP.
The major strengths of this study are its preplanned methods, the use of large, high-quality datasets, and transparent reporting. To our knowledge, this is the largest “Type 3” study in LBP to have—in line with the PROGRESS initiative [13]—published a statistical analysis plan and reported results using the TRIPOD statement (see S1 TRIPOD Checklist). Type 3 studies build on foundational prognostic factor research (Type 1 and 2 studies) [55] by constructing prognostic models. Constructing accurate prognostic models is an essential step towards improving patient outcomes through stratified care (Type 4 studies) [56]. We used large samples of patients with acute LBP to develop and externally validate the models. The samples had a number of differences (Table 1), not least of which was the overall risk of developing chronic LBP (30% in the development sample versus 19% in the external validation sample). Despite these differences, the models made informative predictions in the external sample, which indicates favorable generalizability and suggests that further testing in additional samples is warranted. We have reported different aspects of model performance that can be interpreted for clinical and research applications.
This study has some limitations. First, we were restricted to the use of predictor variables measured in the original studies. We were therefore not able to directly compare our model or update existing models in this study, as is recommended by the PROGRESS framework (Recommendation 21) [13]. Moreover, we may not have included important prognostic variables in our models because they were not measured in the original studies. We attempted to overcome this limitation by updating the model at the external validation stage. Interestingly, when we updated the model with a recently identified prognostic factor, sleep quality [50,51], there was no improvement in any of our indices of predictive performance. Second, we used an automated stepwise approach to specify the models, principally because it is objective and generally results in smaller, clinically applicable models [57], but stepwise methods have well-known limitations such as unstable variable selection [58] and biased coefficient estimation [57]. It is therefore conceivable that our choice to use stepwise selection may have reduced the predictive performance of the models. Third, the overall model fit statistics indicate that the variance explained by our prediction models is modest. Perhaps some factors that are yet to be tested thoroughly in LBP, for example, structural pathology shown on imaging [59], explain additional variance in chronic LBP. However, tests involving imaging are onerous, costly, and potentially harmful for patients with acute nonspecific LBP [60]. Fourth, by prespecifying in our protocol that we would impute missing predictor values only if they were missing in more than 5% of the sample, we did not strictly adhere to the PROGRESS recommendation to impute values where reasonable (Recommendation 20). The complete case approach that we used in our primary analysis can be inefficient and is known to produce bias in prediction research [61]. However, the number of missing predictor values was small (<2%), and our post hoc sensitivity analysis showed no major differences in results when a post hoc imputation procedure was performed (S1 and S2 Tables). This suggests that our a priori decision to remove cases with missing predictor values did not bias the results. Finally, because our prognostic model is in the form of a logistic regression equation, this limits its ease of use. To address this limitation, we developed a calculator (based on the recalibrated PICKUP) that is freely available online at http://pickuptool.neura.edu.au/.
Deciding whether a model is useful or not depends both on its performance and its purpose. In the research setting, discrimination is an important consideration. When such a large number of patients recover with minimal or no intervention, treat all approaches to preventing chronic LBP are inevitably going to be inefficient. Some treatments for LBP, if applied to low-risk patients, may even be harmful. Our models can help discriminate between patients who experience poor outcomes and patients who experience good outcomes, with acceptable performance (AUC > 0.6, likelihood ratios not overlapping). In the external validation sample, patients allocated to the high-risk group (i.e., in the highest quartile of predicted risk) were three times more likely to develop chronic LBP than their medium- or low-risk counterparts (in the middle two and lowest quartiles of predicted risk, respectively). Including only patients with a predicted risk above a 30% in a secondary prevention trial would lead to a net reduction of 40 unnecessary episodes of care (for patients with good outcomes) per 100 patients (Fig 4).
In the clinical setting, calibration is important for providing accurate risk estimates to patients. Our primary prognostic model (PICKUP) demonstrated acceptable calibration (<5% difference between predicted risks and observed proportions of chronic LBP) in seven out of ten risk strata. However, we did observe some miscalibration in the higher risk strata—as predicted risk increased, accuracy decreased and the model overestimated risk (Fig 2). This, along with our negative predictive values above 90% (S4 Table), means that people with lower risk estimates are very unlikely to develop chronic pain, but those with higher risk estimates may still recover quickly. That is, the models are better at ruling out future chronic LBP than ruling it in. However, after recalibration the estimates were almost perfectly calibrated (S1–S3 Figs). With further testing and recalibration, these models have potential to be useful in other clinical settings.
Our decision curve analysis suggested that the primary model is likely to be useful for patients whose decision to pursue further intervention is based on a predicted risk between 12% and 35%. The question that remains is whether these thresholds are clinically relevant. For a range of thresholds under 50% to be considered clinically relevant, the assumption is that patients place more value on detecting an imminent problem (true positive rate) than undergoing unnecessary treatment (false positive rate) [62]. We would suggest that most patients with acute LBP would fall into this category: the consequences of undergoing, for example, an unnecessary course of physiotherapy, are outweighed by the prospect of missing a chance at preventing a long-term problem. However, this assumption rests on the nature of the treatment proposed. If the patient and their physician are considering invasive treatments such as spinal surgery, the patient might weigh the false positive rate more heavily, due to the higher risk of adverse events. In this case, a screening tool would need to yield a net benefit across a range of predicted risk cutoffs higher than 50%, and our model would not be considered useful [62]. We therefore speculate that our models are likely to be useful only for informing the choice between a wait-and-see approach and a course of conservative intervention.
Although several models have been developed in LBP, few have been externally validated [21], and none have been designed to predict the onset of chronic LBP. Pain is arguably the most important outcome to predict in LBP; it is clearly the most important issue for patients [63], and it is the slowest to recover [33]. The three available tools that have been tested in external samples of patients with acute LBP appear to predict pain outcomes at 3 and 6 mo with modest accuracy at best. Grotle et al. [31] tested the OMPQ in an acute LBP sample and reported an AUC for predicting pain at 6 mo of 0.62 (95% CI 0.51 to 0.73). Recent evaluations of SBT score in predicting ongoing pain at 6 mo in acute LBP samples reported AUC values of 0.50 [29] and 0.54 [30]. Williams et al. [64] reported an AUC of 0.60 (95% CI 0.56 to 0.64) for predicting recovery from pain (0 or 1/10 pain sustained for 7 d) at 3 mo. PICKUP appears to discriminate medium-term pain outcomes in patients with acute LBP more accurately than other validated models, and may be particularly useful for secondary prevention trials that target pain reduction. Because calibration performance has not been widely reported, we were unable to compare our model to others in these terms. Williams et al. [64] reported acceptable calibration for their model predicting outcomes in the first 2 wk but relatively poor calibration (more than 10% difference between predicted risks and observed proportions) for predicting pain outcomes at 3 mo. As suggested by PROGRESS, a formal comparison of our tool with other validated tools, for example, using a decision curve analysis, is a logical next step.
Based on its performance in these cohorts, this five-item prognostic model for patients with acute LBP may be a useful tool for estimating risk of chronic LBP. Further validation is required to determine whether screening with this model leads to a net reduction in unnecessary interventions provided to low-risk patients.
|
10.1371/journal.pntd.0000235 | Age-Related Alteration of Arginase Activity Impacts on Severity of Leishmaniasis | The leishmaniases are a group of vector-borne parasitic diseases that represent a major international public health problem; they belong to the most neglected tropical diseases and have one of the highest rates of morbidity and mortality. The clinical outcome of infection with Leishmania parasites depends on a variety of factors such as parasite species, vector-derived products, genetics, behaviour, and nutrition. The age of the infected individuals also appears to be critical, as a significant proportion of clinical cases occur in children; this age-related higher prevalence of disease is most remarkable in visceral leishmaniasis. The mechanisms resulting in this higher incidence of clinical disease in children are poorly understood. We have recently revealed that sustained arginase activity promotes uncontrolled parasite growth and pathology in vivo. Here, we tested the hypothesis that arginase-mediated L-arginine metabolism differs with age.
The age distribution of patients with visceral or cutaneous leishmaniasis was determined in cohorts of patients in our clinics in endemic areas in Ethiopia. To exclude factors that are difficult to control in patients, we assessed the impact of ageing on the manifestations of experimental leishmaniasis. We determined parasite burden, T cell responses, and macrophage effector functions in young and aged mice during the course of infection.
Our results show that younger mice develop exacerbated lesion pathology and higher parasite burdens than aged mice. This aggravated disease development in younger individuals does not correlate with a change in T helper cytokine profile. To address the underlying mechanisms responsible for the more severe infections in younger mice, we investigated macrophage effector functions. Our results show that macrophages from younger mice do not have an impaired capacity to kill parasites; however, they express significantly higher levels of arginase 1 than aged mice and promote parasite growth more efficiently. Thus, our results demonstrate that ageing differentially impacts on L-arginine metabolism and subsequent effector functions of physiologically distinct macrophage subsets.
Here, we show that arginase-mediated L-arginine metabolism is modulated with age and affects the capacity of macrophages to express arginase; the increased capacity to upregulate this enzyme in younger individuals results in a more permissive environment for parasite growth, increased disease severity and pathology. These results suggest that the difference in arginase-mediated L-arginine catabolism is likely to be an important factor contributing to the increased incidence of clinical cases in children. Thus, targeting L-arginine metabolism might be a promising therapeutic strategy against leishmaniasis, especially in children and young adults.
| It is well documented that ageing alters many aspects of immune responses; however, a causal relation between impaired immune functions in ageing individuals and the response to infection has not been established. Experimental leishmaniasis is an excellent model to analyse protective and pathological immune responses. Leishmania parasites are obligate intracellular pathogens and invade mainly macrophages, which have dual function: they can kill the parasites or promote their growth. We have recently shown that arginase, an enzyme induced in infected macrophages, is a key factor for parasite survival. Here, we show that ageing reduces the expression levels of arginase in macrophages, resulting in more efficient control of parasite growth. Our results suggest that age-related differences in the metabolism of arginase in macrophages might contribute to the higher susceptibility of children to leishmaniasis.
| Infections with protozoan parasites inflict an immense toll on the developing world; they are major causes of morbidity and mortality and impede economic development. Leishmaniases are vector-borne diseases, the parasites being transmitted by bites of blood feeding female sandflies and causing different disease manifestations in humans, ranging from the relatively benign self-healing cutaneous form through the disseminated and diffuse cutaneous, to the most severe visceral leishmaniasis. Leishmaniases belong to the most neglected diseases, yet they occur in five continents and are endemic in almost all tropical and subtropical areas [1]. The rising incidence of leishmaniasis around the world is an increasing concern for many countries.
A multitude of factors, including parasite and vector species, host immune responses, genetic and environmental factors influence the outcome of infection. The age of the infected individual also appears to be crucial, as a high proportion of the patients are children. This age-related higher prevalence of disease is most remarkable in visceral leishmaniasis [2],[3],[4],[5],[6],[7],[8],[9],[10]. The mechanisms resulting in this higher incidence of clinical cases in children however are poorly understood.
Leishmania are obligate intracellular parasites they survive and replicate predominantly in macrophages. Experimental infections of inbred strains of mice with Leishmania (L.) major have established the current paradigm of T helper (Th) subset involvement in infectious diseases. Control of infection and healing has been associated with a polarized Th1 response whereas non-healing is attributed to a dominant Th2 response [11]. However, the regulation of immune responses against Leishmania parasites is complex and Th2 dominance does not fully explain non-healing or reactivated forms of disease [12],[13].
Macrophages, the main effector cells in leishmaniasis, can be instructed to kill or to promote the growth of intracellular Leishmania parasites, depending on the balance of two inducible enzymes, nitric oxide synthase 2 (NOS2) and arginase. These two enzymes use a common substrate, L-arginine, and are competitively regulated by type1 and type 2 cytokines [14]. The fate of the intracellular parasite depends on the type of signal the macrophages receive: the type 1 cytokine interferon-γ (IFN-γ) induces classical activation of macrophages and expression of NOS2 that oxidizes L-arginine into nitric oxide (NO), a metabolite responsible for parasite killing [15]; the key type 2 cytokine IL-4 results in alternative activation of macrophages and the induction of arginase [16],[17]. Arginase initiates one of the classic pathways of arginine degradation and it regulates NO synthesis [18]. Two distinct arginase isoforms, encoded by different genes and differing in their cellular localization, have been identified in mammals: type 1 arginase is cytosolic and type 2 arginase is mitochondrial [19]. Arginase hydrolyzes L-arginine to urea and ornithine; the latter being the main intracellular source for synthesis of polyamines necessary for parasite growth. Indeed, we recently showed that sustained arginase activity promotes uncontrolled parasite growth and pathology in vivo [20].
In the current study we tested the hypothesis that arginase-mediated L-arginine metabolism is altered with age and thereby modulates the capacity of macrophages to control parasite growth.
3–4 and 6–8 week old female BALB/c mice (Charles River, UK) were kept in individually vented cages for up to 18 months. The animal colonies, screened regularly for mouse pathogens, consistently tested negative. Animal experiments were performed in accordance with Home Office and institutional guidelines.
Data from 91 patients diagnosed with cutaneous leishmaniasis (CL) at the University of Addis Ababa in the time period from July 2005–April 2007 were used for determination of age distribution of CL. Diagnosis was confirmed by smear and/or NNN culture and later confirmed to be due to L. aethiopica by biochemical typing of randomly selected strains. Typing of selected strains was done by a multi-locus enzyme electrophoresis (MLEE) technique at the Istituto Superiore di Sanita in Rome, Italy.
Data from 37 patients diagnosed from 1997–2000 with visceral leishmaniasis (VL) from an endemic region in Ethiopia were used to determine the age distribution of VL. All VL cases were due to L. donovani (determined by MLEE) and were diagnosed as described [21]. The presence of parasites in lymph node or splenic aspirates was assessed by direct smear or culture [21].
For infections, 2×106 L. major LV39 (MRHO/SU/59/P-strain) promastigotes were injected s.c. into the footpad and lesions were monitored as described [22].
The number of living L. major parasites in infected tissues was determined using the parasite limiting dilution assay [22].
Arginase activity was measured in macrophage lysates as previously described [14],[20]. One unit of enzyme activity was defined as the amount of enzyme that catalyzes the formation of 1 µmol of urea per min.
Arginase activity at the site of lesions was determined ex vivo using 1–10 µl of footpad homogenate, and the method described above.
Amastigotes were purified from the lesions as described in [23] and arginase activity was determined as described above.
Arginase 1 protein expression was determined by western blot as described [20], using anti-arginase 1, a rabbit polyclonal antibody raised against rat arginase 1, which cross-reacts with mouse arginase 1 but not arginase 2 [24].
Draining lymph nodes from individual mice (5×106 cells/ml) were restimulated with 1×106/ml L. major promastigotes. Forty-eight hours later, culture supernatants were harvested and cytokines determined by ELISA according to the supplier's protocol (Pharmingen). Detection limits are 20 pg/ml for IL-4, 20 pg/ml for IL-10 and 1 U/ml for IFN-gamma.
The frequency of CD4+ T cells expressing IL-4, IL-10 or IFN-γ was determined by flowcytometry as described [25]. Detection of intracellular cytokines was determined using an EPICS XL instrument (Beckman Coulter). Data were analyzed using Beckman Coulter Expo 32 software.
The proliferative capacity of CD4+ T cells was determined by flowcytometry, measuring BrdU incorporation as described [26].
Bone marrow obtained by flushing the femurs of naïve mice was cultured during 8 days in hydrophobic Teflon bags as described [14].
Macrophages were stimulated with 100 U/ml IFN-g (Peprotech) and 500 U/ml TNF-α (Peprotech) or with 20 U/ml IL-4 (Peprotech) for 48 hr. BMMΦ (5×105) were incubated with FcR blocking reagent (Pharmingen) for 5 min and anti-CD206 (Serotec), anti-CD40, anti-CD80 or anti-CD86 mAbs (eBioscience) were added for 20 min at 4°C. The expression of CD206, CD40, CD80 and CD86 was determined using an EPICS XL instrument (Beckman Coulter).
NO2− accumulation was used as an indicator of NO production and measured using Griess reagent [22].
Mature macrophages (5×105/ml) were plated and stimulated with either 100 U/ml IFN-γ (Peprotech) and 500 U/ml TNF-α (Peprotech) or indicated concentrations of IL-4 or IL-13 (Peprotech) and infected with 25×105/ml L. major parasites. After four days the macrophages were washed, lysed and the number of viable parasites was determined as described [22].
Statistical differences were determined using a two-tailed Mann-Whitney test. Differences were considered statistically significant at P<0.05.
To assess the impact of ageing on the development of experimental leishmaniasis, genetically susceptible young (6–8 week) and older (12 months) BALB/c mice were infected s.c. with L. major parasites and lesion development and parasite load were determined. The onset of lesion development was similar in both groups of mice. After three weeks of infection, the young BALB/c mice started to develop ulcers and 4 weeks later, the experiment had to be terminated owing to severity of the lesions. In contrast, older BALB/c mice had clearly reduced pathology; they did not develop ulcers and their lesions stabilized (Figure 1a). Consistent with the exacerbated pathology and increased lesion size, the parasite load at the site of the lesion was significantly higher in the young than in old BALB/c mice (Figure 1b). Similar results were obtained with mice aged 8, 12 and 18 months (data not illustrated). These results show that disease pathology is clearly exacerbated in young BALB/c mice.
Data obtained on leishmaniasis in our clinics in endemic regions of Ethiopia show that 67% of cases of cutaneous leishmaniasis occur in people younger than 26, and even more striking, 85% of cases of visceral leishmaniasis occur in children under 16 years (Figure 2). Thus in the human, the majority of the clinical cases appears to occur in younger individuals. Whereas many factors are likely to contribute to this higher prevalence of clinical cases in younger individuals, these data supports our finding that age is a major factor in the control of disease.
To assess whether the increased disease severity (Figures 1a and b) observed in young BALB/c mice correlated with modulation of the T helper phenotype, we determined antigen-specific cytokine production by lymph node cells. Interestingly in all experiments performed there was a consistent increase in the levels of antigen-specific IFN-γ, IL-4 and IL-10 in the aged mice, however, these differences were not statistically significant (P>0.05) (Figure 3 and Table 1). In agreement with these results, the frequencies of CD4+ T cells expressing IFN-γ, IL-4 or IL-10 were not significantly different between young and aged mice (Table 2). In addition, the capacity of CD4+ T cells to proliferate in response to L. major parasites was similar (Table 2).
These results indicate that increased disease severity in young BALB/c mice is unlikely to be due to alterations in the key Th cytokine profile.
We have recently shown that in L. major infected healer strains of mice (CBA, C57BL/6), arginase activity is low, only detectable during active disease. In contrast, increasing arginase activity was detected at the local site of lesion, and clearly correlated with uncontrolled parasite replication and lesion size in nonhealing BALB/c mice [20]. To assess whether the exacerbation of disease in young mice was also associated with higher arginase expression at the site of infection, we determined the activity of this enzyme. The data presented in Figure 4a do indeed show significantly higher arginase activity in the lesions of young BALB/c mice than in those of their older counterparts. We also determined arginase 1 protein expression by Western blot and found arginase to be higher in the lesions of young mice (Figure 4b).
Hormones such as estrogens can influence resistance to infections [27],[28],[29],[30],[31]; since estrogens are significantly higher in mice of reproductive age, in the next experiment we use pre-pubertal mice aged 3–4 weeks and determined whether they displayed an altered resistance to L. major infection. As shown in Figure 4c, the lesion size was slightly higher in 3–4 weeks old mice as compared to those of young adult mice (10 week old). Importantly lesion sizes from 3–4 and 10 week old mice were significantly higher then those of aged mice (40 week old). Accordingly, both parasite loads and arginase activities at the site of infection were significantly increased in both pre-pubertal and adult mice as compared to aged mice.
The increased arginase activity and arginase protein expression in the lesions of young L. major infected BALB/c mice support our hypothesis that arginase mediated L-arginine metabolism alters with age. These results further confirm our previous conclusion that the level of arginase activity correlates with disease severity [20].
We and others have shown that Leishmania parasites express their own arginase [20],[32],[33]. Therefore, to measure the contribution of parasite arginase to the total levels of arginase activity measured in the lesions of BALB/c mice, we measured arginase in the whole lesions, as well as in amastigotes purified from lesions. As shown in Figure 4d, we could detect low levels of arginase activity in the purified amastigotes, however it was 40.5-fold lower as compared to that detected in the parasitised lesions. No arginase 1 protein was detectable by Western blot in the purified amastigotes (data not shown). These results show that arginase activity from amastigotes only contributes minimally to arginase activity measured in the lesions.
Since macrophages are the ultimate effector cells, and arginase activity is increased in lesions of young mice, we reasoned that macrophages derived from old and young mice must differ in the efficiency of their effector functions. To test this hypothesis, we first examined the expression of activation markers on classically activated macrophages (CAMΦ) No differences were detected in the expression of CD86, CD40, CD80 and CD206 between CAMΦ from young and older mice (Figure 5a). We also assessed the capacity of CAMΦ from young and old mice to upregulate NOS2 and produce NO. MΦ from both groups produced similar levels of NO in response to IFN-γ/TNF-α in the absence or presence of parasites (Figure 5b). Furthermore, we determined NO production by CAMΦ derived from pre-pubertal, adult and aged mice. As shown in Table 3, NO production was similar in all age groups.
We conclude that ageing does not affect the capacity of CAMΦ to induce NOS2 or to generate NO and did not alter expression levels of activation markers in CAMΦ.
We next examined the capacity of alternatively activated (AA) MΦ from young and old mice to exert their effector functions. AAMΦ were obtained by differentiating mature bone marrow MΦ with IL-4 or IL-13 alone, as IL-10, which synergizes with IL-4 and strongly enhances the expression levels of arginase, acts on a different receptor [17]. We found no differences in the expression of CD206, CD40, CD80 and CD86 between AAMΦ derived from young and older mice (Figure 6a). We then assessed the capacity of macrophages from young and older mice to induce arginase in response to IL-4 or IL-13. As shown in Figure 6b, upregulation of arginase was observed in both age groups of AAMΦ, and infection with L. major parasites enhanced the arginase levels even further. Importantly, AAMΦ derived from young mice displayed significantly higher arginase activity in response to IL-4 or IL-13. We also examined the capacity of AAMΦ derived from pre-pubertal BALB/c mice (3–4 weeks) to express arginase; as shown in Table 4, similarly to adult mice, they express significantly more arginase then aged mice (40 weeks). These results show that ageing alters the capacity of AAMΦ to express arginase.
The polyamines resulting from the hydrolysis of L-arginine by arginase are essential for parasite growth [20],[34]. Here we hypothesize that lower arginase activity in AAMΦ from older mice (Figures 4a, b and 6b) is responsible for the decrease in parasite replication.
To determine whether AAMΦ from young and old mice support L. major replication differently, we examined the number of viable parasites by limiting dilution assay. There is a clearly higher parasite load in AAMΦ derived from young mice compared to those in older mice (Figure 7), indicating that AAMΦ from young mice promote parasite growth more effectively then AAMΦ from older mice. CAMΦ from both groups, however, are equally efficient in parasite killing (Figure 7), suggesting transport of L-arginine is sufficient in both age groups.
The results show that AAMΦ from young mice provide a more permissive environment for parasite growth, directly linking the higher arginase activity (Figures 1b, 4 and 6) to increased parasite replication (Figure 7).
Our results show that younger BALB/c mice develop clearly exacerbated disease as compared to older mice, as demonstrated by more pronounced lesions, pathology and higher parasite burdens at the site of parasite inoculation. This aggravation of disease correlates with a higher arginase activity expressed at the site of infection. This is due to altered macrophage effector functions, as AAMΦ derived from younger mice have an increased capacity to express arginase. To our knowledge, this is the first report showing an age-related modulation of arginase expression. Importantly, arginase activity in vivo was also increased at the site of lesion in L. major infected younger BALB/c mice. We have recently shown that sustained arginase activity promotes uncontrolled parasite growth in BALB/c mice and is pivotal for parasite growth [20]. Thus, increased arginase activity in macrophages from young mice results in increased L-arginine catabolism and consequently increased levels of polyamines, providing a more permissive environment for parasite replication.
The reduced parasite growth in macrophages from elderly mice could be due to enhanced parasite killing or reduced parasite growth or a combination of both. The results presented in Fig. 5b show that the potential of bone marrow derived classically activated macrophages to generate nitric oxide is comparable in cells derived from young, adult and elderly mice.
In contrast, the potential to induce arginase activity declined with age and the presence of L. major parasites further enhanced arginase activity (Fig. 6b). Both arginase and iNOS compete for utilizing L-arginine, however, this does not necessarily indicate that the increased susceptibility is due insufficient parasite killing. Improved control of parasite growth in the elderly is not necessarily due to a shift of the L-arginine metabolism towards the iNOS pathway and enhanced NO production. The expression of both of the L-arginine metabolizing enzymes in macrophages is induced by type 1 and type 2 cytokines, respectively and activation of macrophages results in increased expression of cationic amino acid transporters and increased transport of L-arginine into the macrophages; both enzyme levels and L-arginine transport determine the catabolic rate. Thus, the net-effect of unaltered generation of NO ( = constant rate of killing) and reduced induction of arginase activity ( = reduced nutrients, reduced growth) with age is likely to account for the improved disease control. Indeed, we previously showed that arginase directly regulates parasite growth without affecting NO synthesis [20]. In this context, it is important to note that macrophages synthesize proline via the arginase pathway and this macrophage-generated proline could also be used by the parasites. In agreement with our findings, arginase 1 has been shown to be limiting for polyamine synthesis without affecting NO levels in activated macrophages [35],[36].
We have previously shown that the levels of arginase 1 expressed by L. major parasites influence their virulence [32] and parasite arginase can catabolize radioactive labelled L-arginine [32]. Thus, parasite arginase activity is likely to influence the early immune response. Importantly, the levels of arginase expressed by intracellular L. major amastigotes are negligible as compared to macrophage arginase (Figure 4d) [32]. In agreement with these findings, a recent publication showed that arginase activity expressed by L. mexicana amastigotes is only a minute fraction compared to the macrophages arginase activity (0.418 in macrophages, 0.0235 by parasites [37]).
In conclusion, these in vitro results indicate that macrophages from aged mice provide a less favourable environment for parasite growth than those from younger mice due to reduced nutrient availability.
Ageing is generally associated with a decline of immune responses [38] and some responses can be improved by targeting L-arginine metabolism [39]. L-arginine administration is beneficial in conditions such as trauma, stress, burn or injury, it improves immune functions and facilitates wound healing [40].
Arginase plays a crucial role in wound healing, a process severely impaired in the elderly. During wound healing, NOS and arginase are induced in a time-coordinated manner [41], with NOS expression peaking first and to be particularly important in initiating the inflammatory phase. Arginase is expressed later and is of prime importance in actual wound healing; the conversion of L-arginine into urea and ornithine and its metabolism to polyamines initiates the repair phase of the inflammatory response [42]. The reduced expression of arginase in macrophages from older individuals is likely to result in lower levels of ornithine, thereby reducing collagen formation and the proliferation of cells, which might then result in delayed or ineffective wound healing.
In experimental leishmaniasis, uncontrolled disease development is generally associated with a strong Th2-type response [11], however, strong polarized Th2 responses are not sufficient to explain fully nonhealing [12],[13]. In agreement with the cited studies, our results show that amelioration of disease in older mice does not depend on a switch from a Th2- to a Th1-type response. The analysis of the Th1-Th2 cytokine balance during ageing has often resulted in contradictions; both impaired [43] and enhanced Th2 responses [44] have been reported. Factors such as the use of different strains of mice and parasites are likely to contribute to contradictory results.
In support of the increased incidence of leishmaniasis in children and young adults two studies, using mathematical models, have shown an association between age and rate of Leishmania infection and suggest that susceptibility to disease declines with age [45],[46]. However, the relationship between age and susceptibility to disease remains an open question. The development and manifestation of disease is the result of a dynamic interplay of a range of different environmental, host-, vector- and parasite-derived factors. Many factors such as nutrition, co-infections, behaviour, differences in metabolic pathways, and prolonged exposure to repeated bites by infected sandflies are important, but difficult to measure in patients. It still remains unclear why there are fewer clinical cases in older people; acquired immunity to re-infection is likely to be on of the main factors; indeed, unexposed adults display a higher susceptibility to infection as compared to the population living in endemic areas [47]. Furthermore, asymptomatic infections appear to provide natural immunisation [48]. In another study where VL was shown to be a disease of children, it was shown that there was a significantly higher prevalence of adults with positive Leishmanin Skin Test (LST), suggesting that they had acquired immunity [49] However, children remain the main risk group for VL [46] and in several outbreaks, children were the population most at risk [49],[50],[51],[52],[53] suggesting that acquired immunity alone might not be sufficient to explain age related higher prevalence of leishmaniasis. The higher level of protection against leishmaniasis with age could also be due to qualitative differences in cell-mediated immune responses in children and older people.
Based on our results, we propose that increased arginase expression and L-arginine catabolism contribute to exacerbation of leishmaniasis in younger individuals. The therapeutic regulation of L-arginine metabolism may provide a target for disease intervention. |
10.1371/journal.ppat.1007589 | Sensing of cell-associated HTLV by plasmacytoid dendritic cells is regulated by dense β-galactoside glycosylation | Human T Lymphotropic virus (HTLV) infection can persist in individuals resulting, at least in part, from viral escape of the innate immunity, including inhibition of type I interferon response in infected T-cells. Plasmacytoid dendritic cells (pDCs) are known to bypass viral escape by their robust type I interferon production. Here, we demonstrated that pDCs produce type I interferons upon physical cell contact with HTLV-infected cells, yet pDC activation inversely correlates with the ability of the HTLV-producing cells to transmit infection. We show that pDCs sense surface associated-HTLV present with glycan-rich structure referred to as biofilm-like structure, which thus represents a newly described viral structure triggering the antiviral response by pDCs. Consistently, heparan sulfate proteoglycans and especially the cell surface pattern of terminal β-galactoside glycosylation, modulate the transmission of the immunostimulatory RNA to pDCs. Altogether, our results uncover a function of virus-containing cell surface-associated glycosylated structures in the activation of innate immunity.
| Human T Lymphotropic virus type (HTLV) establishes persistent infections, leading to adult T-cell Leukemia, a life-threatening cancer in chronically-infected individuals. Viral persistence likely results from a failure of immune responses to eradicate viral replication, a least in part, by viral escape from innate immunity, and notably via decreased production of type I interferons (IFN-I) by infected cells. Plasmacytoid dendritic cells (pDCs) are known as robust producers of IFN-I in response to virus stimulation, thus bypassing the viral mechanisms to evade pathogen-sensing pathways in infected cells. However, HTLV particles are not detected in biological fluids of infected individuals, raising the question of the pDC-activating signal. Here, we demonstrate that pDCs produce IFN-I upon physical contacts with HTLV-infected cells. We show that pDCs sense surface associated-HTLV present with glycan-rich structure, referred to as HTLV-biofilm-like structure. Importantly, the sensing of infected cells by pDCs is modulated by the glycosylation pattern at the surface of infected cells. This newly ascribed regulation of innate immunity activation by cell surface-associated glycans might contribute to the differential activation levels of antiviral response to infected cells when their glycosylation profile is modified, such as for chronically infected cells or tumor cells.
| Human T-Lymphotropic Virus type 1 (HTLV-1) infects over an estimation of 5–10 million people. HTLV-1 is mainly present in Japan, central Africa, Caribbean and South America [1,2]. After a long period of clinical latency, HTLV-1 infection leads, in a fraction of infected individuals, either to Adult T-cell Leukemia/Lymphoma (ATL) [3] an uncontrolled CD4+ T–cell proliferation of very poor prognosis, or to an inflammatory disorder named HTLV-1 Associated Myelopathy / Tropical Spastic Paraparesis (HAM/TSP) [4]. In chronically infected individuals, HTLV-1 provirus is mainly found in CD4+ T-cells, yet infected dendritic cells (DCs) are also detected [5,6]. Their function is subsequently altered in vivo [6–8], thereby most likely contributing to viral pathogenesis.
Viral persistence leading to chronic infection and its associated diseases implies that innate and adaptive immune responses fail to eliminate HTLV-1 infected cells, possibly because HTLV-1 has evolved efficient strategies to escape immune pathways [9]. Type-I interferons (referred herein to as IFN-I, i.e., IFNα and β) are key mediators of innate immunity. They induce the expression of IFN-stimulated genes (ISGs) that suppress viral spread at different stages of the viral cycle, and stimulate the onset of adaptive immune responses. The IFN-I response is initiated via the recognition of pathogen-associated molecular patterns (PAMPs) by pattern-recognition receptors (PRRs), including the Toll like receptors (TLRs) [10]. Like virtually all viruses [11], HTLV-1 inhibits several steps of the PRR-induced pathways [12–14], and as a consequence, blunts IFN-I induction and signaling [15,16], leading to very limited production of IFN-I by infected cells.
Because the acute phase of the infection is asymptomatic, very little is known regarding host innate responses in HTLV-1-infected individuals. Nonetheless, indirect evidence infers that the IFN-I response exerts an antiviral action against HTLV-1. First, while not easily detectable in vivo [17], viral proteins expression is induced in T-lymphocytes isolated from infected patients when cultured ex vivo [18], likely as a result of the relief from in vivo repression. Consistently, culture of HTLV-1-infected cells with IFN-β-expressing stromal cells represses viral protein expressions [18]. Second, exogenous IFN-I decreases viral protein translation in vitro, and protects lymphocytes from de novo infection [19]. Thus, IFN-I-mediated antiviral control of HTLV-1 infection is likely to occur in vivo. Nonetheless, the cell type that produces IFN-I during infection remains enigmatic.
Plasmacytoid dendritic cells (pDCs) act as sentinels of viral infection, as they are the major IFN-I producers in vivo [20], being 1000-fold more potent for IFN-I production as compared to other cell types [20]. They predominantly recognize viral nucleic acids, i.e. single-stranded RNA and non-methylated CpG-containing DNA, by TLR7 and TLR9, respectively [21]. Cell-free HTLV-1 particles, when added at high concentration, were shown to induce IFN-I production by pDCs in vitro, in a TLR7-dependent manner [22]. Nonetheless, cell-free viruses are undetectable in the plasma of HTLV-1-infected individuals, which leaves open the question of the modality of pDC activation in vivo. Importantly, we and others recently revealed that cell contacts are required for efficient pDC activation by evolutionary divergent RNA viruses belonging to distinct families, such as Flaviviridae, Picornaviridae, Arenaviridae, Retroviridae, and Togoviridae [23–31]. Transfer of immunostimulatory viral RNAs from infected cells to pDCs was further shown to involve carriers in the form of non-infectious and/or non-canonical viral particles, including exosomes [25,27,29] and immature virus particles [24].
Interestingly, cell-cell transmission of viral material is reminiscent of HTLV cell-cell transmission [32], which is the only efficient way to infect new target cells. HTLV-1 viral transmission occurs through the transfer of neo-synthesized HTLV-1 virions via a virological synapse formed at the cell contact [33], and/or infectious viral particles embedded at the surface of infected cells within an extracellular matrix components (ECM)-rich structure [34]. The latter is referred to as the HTLV-1 biofilm-like structure [34]. This HTLV-1 biofilm-like structure has been further defined as the minimal infectious structure allowing viral transmission [35]. Importantly, the role of the cell surface associated virus within biofilm-like structure in the activation of the innate immune response is still unknown.
Here, we demonstrate that the pDC-mediated IFN-I response requires physical contacts with HTLV-infected cells. Moreover, we show that HTLV-1 biofilm-like structure represents the minimal virally induced-structure able to trigger an IFN-I response by pDC, and thus recapitulating pDC activation induced by contact with infected cells. Further, comparison of a panel of HTLV1/2 infected cells reveals that pDC-mediated IFN-I response inversely correlates with the ability of the HTLV-infected cells to transmit infectivity and with their surface glycosylation pattern. Indeed, we show that the density of terminal β-galactoside glycosylation at the surface of infected cells regulates IFN-I production by pDC. Altogether, our results highlight an unforeseen function of virus-containing cell surface-associated structures in the activation of pDCs by cell contacts, as well as its fine-tuning by the glycosylation pattern at the surface of the sensed infected cells.
We first determined the production of type I IFN (referred to as IFN-I) by PBMCs and pDCs upon recognition of infected cells as compared to cell-free virions present in supernatant (SN) of infected cell lines (Fig 1). pDCs, representing 0.2–0.5% of total PBMCs, were isolated from healthy blood donors with >91% of purity (Fig 1A, middle panel), consistently with our previous reports [24,25,30]. PBMCs or purified pDCs (referred to as responders) were co-cultured with HTLV-1 chronically infected cells, i.e., C91-PL cell line [36], (referred to as inducer). These HTLV-1-infected cells induced a potent IFN-I response by both PBMCs and purified pDCs, when in physical contact (Fig 1B). In sharp contrast, cell-free viruses present in the supernatant from HTLV-1-infected cells (approximately 10–25 ng/mL of the HTLV-1 capsid p19gag, i.e., representing the viral concentration reached in the supernatant of inducer cells at the time of coculture) failed to induce very low, or undetectable levels of IFN-I production (Fig 1B, around 5 U/mL when detected, or below the detection limit).
Next, we tested the contribution of pDCs relative to other PBMC cell types in the IFN-I response to HTLV-1 infected cells. Depletion of pDCs from PBMCs (Fig 1A, lower panel) abrogated the response to HTLV-1 infected cells (Fig 1B). We controlled that pDC-depleted PBMCs and PBMCs produced comparable levels of IL-6 after LPS stimulation, confirming that pDC depletion did not impair PBMC responsiveness (Fig 1C). pDCs obtained from 27 donors, reproducibly demonstrated robust IFN-I responses to HTLV-1 infected cells (Fig 1D; median value of 13 400 U/mL), albeit with some donor-to-donor variations. Of note, HTLV-1 infected cells alone did not produce IFN-I (Fig 1D). Together, these results indicate that pDCs are the main, if not exclusive, IFN-I producers among PBMCs in response to the contact with HTLV-1-infected cells.
Next, we tested whether pDC sensing of HTLV-1 infected cells involves TLR7, a sensor of single-stranded RNA. Inhibition of TLR7 recognition using a competitive inhibitor significantly decreased the IFN-I response to HTLV-1-infected cells (Fig 1E). The specificity of TLR7 inhibitor was validated by the inhibition of IFN-I production triggered by a TLR7 agonist but not by a TLR9 agonist, as expected (Fig 1E). These results suggested that pDCs sense HTLV-1-infected cells via TLR7, implying that HTLV-1 viral RNA is likely the immunostimulatory signal. Since IFN-I production by pDCs following incubation with cell-free viruses was not or barely detectable, we hypothesized that cell contacts are required for pDC activation. We thus measured IFN-I production when pDCs were physically separated from HTLV-1-infected cells by a 0.4μm permeable membrane (Fig 1F, TW). The absence of physical contact between inducer and responder cells abrogated IFN-I production (Fig 1F). We controlled that pDC responsiveness was maintained in this experimental setting, as pDCs produced similar amounts of IFN-I upon TLR7 agonist stimulation when cultured in transwell chambers or not (Fig 1F). This demonstrated that pDC contact with infected cells is required to trigger IFN-I production.
Exosomes have been involved in the transfer of immunostimulatory RNAs to pDCs for other viruses [25,27,29,37] and HTLV-1 infected cells are known to produce exosomes [38]. To test whether exosomes are involved in the transfer of the HTLV-1 immunostimulatory signals, we used the C8166 HTLV-1 cell line, which is impaired for expression of the structural proteins Gag and Env and thus do not produce infectious viral particles [39], as confirmed by absence of infectivity transmission to Jurkat-LTR-Luc reporter cell line (Fig 1G). While C8166 HTLV-1 cells retain the capacity to produce the Tax regulatory protein, and exosomes that contain several viral mRNAs [38], they failed to induce IFN-I production by co-cultured pDCs (Fig 1G). This inferred that the transmission of activating signal to pDCs likely requires Env gp46 and/or Gag mediated extracellular export of viral RNA, rather than exosomal export of viral RNAs. To address the importance of Env gp46 in pDC IFN-I response, we tested pDC activation upon co-culture with Jurkat cells transfected with the WT HTLV-1 molecular clone (i.e., pACH) or with the counterpart molecular clone lacking the envelope glycoprotein (i.e., pACH-ΔEnv). As expected, Env gp46 was not expressed when Jurkat cells were transfected with the ΔEnv molecular clone, while p19gag levels were similar (Figs 1 and S1A). Cells harboring WT but not ΔEnv molecular clone or only Tax expressing vector induced a robust IFN-I production by co-cultured pDCs (Fig 1H).
Next, we tested whether primary HTLV-1 infected cells from HAM/TSP patients were also able to induce IFN-I production by pDCs. As HTLV-1 infected cells isolated from the blood of patients do not express HTLV-1 [18], PBMCs from 3 HAM/TSP patients were first cultured in presence of IL-2 and PHA to induce viral re-expression. This was controlled by p19gag detection (S1B Fig). Viral re-expression was observed in all patient samples, with some donor-to-donor variation as expected (S1B Fig). These cells were then co-cultured with pDCs. PBMCs from the 3 independent HAM/TSP patients significantly induced pDC IFN-I production (Fig 1I), as opposed to the absence of response to PBMCs from healthy donors used as controls.
We then aimed at determining whether pDCs are susceptible to HTLV-1 infection as previously reported [5], in our experimental setup leading to IFN-I production (i.e., within 24h-incubation with HTLV-1 infected cells). The productive infection of pDCs at the end of co-culture with HTLV-1 infected cells was assessed by the detection of Tax, as we previously reported [32]. In contrast to monocytes-derived dendritic cells (MDDCs), Tax expression by pDCs was not readily detected 24h after co-culture with HTLV-1 infected cells (S2A Fig). Thus, this suggests that pDC IFN-I response to HTLV-1 infected cells does not involved a productive infection. Altogether, our results demonstrated that pDCs sense HTLV-1 infected cells by Env gp46-mediated transmission of pDC-activating signal by cell contact leading to robust IFN-I response via TLR7-induced signaling.
The capture of HTLV-1 cell-free virus by target cells involved binding of Env gp46 to NRP-1/BDCA-4 in cooperation with HSPG [40] and then to Glut-1 [41]. The latter also serves as the receptor mediating fusion of HTLV envelope with the cellular membrane [42]. NRP-1/BDCA-4, Glut-1 and HSPG are all readily expressed at the pDC surface (Fig 2A). We thus sought to determine the contribution of these receptors in the transfer of the activating signal from the infected cells to the pDCs. Previous reports showed that the binding of HTLV-1 Env gp46 to its receptors is mediated by the receptor binding domain (RBD; the first 215 amino acids of gp46), and can thus be out-competed by recombinant RBD [41]. Competition with recombinant RBD significantly reduced IFN-I production by pDCs (Fig 2B), viral binding to pDCs (Fig 2C and S2B Fig) and viral transmission to reporter cells (Fig 2D). This suggests that pDC sensing requires HTLV-1 Env binding to its receptor(s). RBD comprises residues that have been specifically involved in NRP-1/BDCA4 (i.e., at the position 90-to-94) [40] and the 94-to-101 stretch known to be pivotal for Glut-1 binding and subsequent viral fusion [43]. Thus, it does not allow to discriminate between binding to NRP-1/BDCA-4 versus Glut-1. Nonetheless, binding of Env gp46 to NRP-1/BDCA4 can be prevented by addition of recombinant VEGF165, a known ligand of NRP-1/BDCA4, that interacts directly through a peptide stretch similar to the 90–94 sequence found in Env gp46 but also using an HSPG dependent manner [40]. Thus VEGF165 does not allow discriminating binding to NRP-1 versus HSPG. Competition with recombinant VEGF165 did not prevent the IFN-I production by pDCs (Fig 2B), viral binding to pDCs (Fig 2C) nor cell-cell viral transmission to reporter cells (Fig 2D), suggesting that NRP-1/BDCA-4 HSPG-mediated and/or direct binding may not be involved in HTLV-1 transfer by cell-cell contact. The effectiveness of VEGF165 competition was confirmed by the expected reduction of the binding of cell-free HTLV-1 virion to reporter cells measured by flow cytometry detection of p19gag (Fig 2E), consistent with a previous report [40], whereas no competition by VEGF165 was observed in co-culture experiments with HTLV-1 infected cells (C91-PL) (Fig 2E and S2C Fig). VEGF165 and RBD treatment did not impair the pDC IFN-I response upon stimulation with a TLR7 agonist, thus ruling out non-specific effects of recombinant RBD and VEGF165 on pDC responsiveness (Fig 2B). Altogether, these results suggested that both transmission of infection to target cells after cell-cell contact as well as HTLV-1 sensing by pDCs require Env gp46 interaction with at least Glut-1.
Previous reports showed that HTLV-1 virions are present at the cell surface embedded within carbohydrate-rich elements, referred to as a viral biofilm-like structure [34], and involved in the infectivity transmission [34,35]. Since pDCs respond to HTLV-1-infected cells upon physical contact, we first determined whether HTLV-1 biofilm-like structure was present at the contact site between pDCs and HTLV-1-infected cells. Confocal microscopy analyses of pDC in contact with C91-PL cells revealed that HTLV-1 Env gp46 accumulates at the pDC/infected cell interface, together with carbohydrate-rich elements, known to be present in the viral biofilm-like structure [34], as revealed here by WGA lectin staining (Fig 3A). Of note, Env gp46 and WGA clusters were co-localized at the contact site for most of the analyzed pDC/infected cell contacts (Fig 3B, approx. 85%), suggesting that cell contacts are preferentially oriented toward these specific biofilm-like structures, or inversely that the biofilm-like structures are preferentially positioned at the contact site.
Next, we determined whether HTLV-1 biofilm-like structures could trigger pDC response. pDCs were cultured in the presence of HTLV-1 biofilm-like structures, isolated from infected cells as previously described [35]. Virus concentration in isolated biofilm-like structures ranged from 23.5 to 31.6 ng/mL of p19gag, a concentration similar to that found in the supernatant of infected cells. While cell-free virus-containing supernatants failed to induce IFN-I production by pDCs (Fig 1B), isolated biofilm-like structures significantly activated pDC IFN-I response (Fig 3C). This was specific to HTLV-1 infected cells, since similar isolation procedure from uninfected cells failed to activate the pDCs (Fig 3C, cont.), ruling out a putative non-specific activation by the experimental process (e.g., cellular debris). As expected, isolated biofilm-like structures from HTLV-1-infected cells transmitted infectious virions to Jurkat target cells, (Fig 3D). To further confirm that pDCs sense HTLV-1 biofilm-like structures, HTLV-1 biofilm-like structure was depleted using metalloprotease that digest the extracellular matrix [44] as previously described [34]. The metalloprotease treatment of HTLV-1-infected cells decreased the levels of surface envelope gp46 compared to untreated cells (Fig 3E), in association with a reduction in both IFN-I production by co-cultured pDCs and viral transmission to Jurkat-LTR-Luc reporter target cells (Fig 3F and 3G, and S3A and S3B Fig). This is shown for a representative experiment (i.e., pDCs from one blood donor co-cultured with biofilm-depleted HTLV-1 infected cells in Fig 3F and 3G) and for the means of independent experiments using pDCs from 3 blood donors (S3A and S3B Fig). Altogether, these results show that the HTLV-1 biofilm-like structure contains the immunostimulatory signal that triggers IFN-I production by pDCs.
Since HTLV-1 embedded in biofilm-like structure but not cell-free virons induced pDC IFN-I response, we next sought to examine the elements present in the biofilm-like structure that contribute to viral transmission and subsequent pDC activation. HTLV-1 biofilm-like structures contain ECM components and linkers, including collagen, agrin and heparan sulfate proteoglycans (HSPGs) [34]. As HSPGs are involved in cell-cell and cell-ECM interactions [45], we hypothesized that HSPGs present in HTLV-1 biofilm-like structure and/or in association with HSPGs at the pDC surface (Fig 2A, right panel) could favor cell-cell adhesion. To test this hypothesis, we used heparin, a polyanionic glycosaminoglycan that mimics the sulfate groups of HSPGs and that could thus act as a bridge to increase the pDC/infected cell contacts via the HTLV-1 biofilm-like structure. The impact of heparin on the frequency of cell conjugates formed between HTLV-1 infected cells and pDCs was analyzed by imaging flow cytometry (Image Stream X technology) (S4A Fig), as we previously established [24]. This quantitative analysis revealed that heparin significantly increased the frequency of pDC-HTLV-1-infected cell conjugates (Fig 4A). Consistently, heparin augmented pDC IFN-I production induced by HTLV-1 infected cells (Fig 4B). Importantly, heparin increased as well pDC activation induced by isolated HTLV-1 biofilm-like structures (Fig 4B). Similar results were obtained using blood samples from different donors (S4B Fig). HSPGs are known to act as attachment factors via an interaction with Env gp46 [46], thus heparin could compete for HTLV-1 binding and subsequent HTLV-1 infection. Nonetheless, similar heparin addition had no impact on the viral transmission to Jurkat reporter cells using either isolated HTLV-1 biofilm-like structures or HTLV-1-infected cells (Fig 4C). This contrasts with a previously reported impact of heparin in the context of distinct experimental procedure that showed that heparin, when added during biofilm isolation, reduced the ability of the isolated biofilm to infect Jurkat reporter cells [34]. The presence of heparin during biofilm isolation might have loosened the biofilm structure allowing a better exposure of the viral envelop to heparin competition. The lack of heparin competition both on infected cells and isolated biofilm (Fig 4C) demonstrated that under our experimental conditions, heparin did not compete with HSPGs for HTLV capture when HTLV-1 is embedded in an intact viral biofilm. Altogether, these experiments suggested that heparin increases pDC/infected cell contact as well as pDC transfer of the immunostimulatory signal from the isolated HTLV-1 biofilm-like structures and, likely as a consequence the potentiation of pDC IFN-I response, albeit additional effect can contribute as well. Importantly, the absence of modulation by these heparin treatments of infectivity transmission to target cells highlighted that the transfer of the immunostimulatory signal to pDCs features distinct regulatory mechanism(s) as compared to the infectivity transmission to other cell types.
Next, we sought to define the viral determinants and other cellular component(s) modulating the level of pDC response to infected cells, including the amounts of viral RNAs, cell contact efficiency and ability to transmit viral infectivity. To address these questions, we compared three HTLV-1 chronically infected cell lines (C91-PL, MT-2 and Hut102), known to produce different amounts of viral proteins [47] and two HTLV-2-infected cell lines (MO and C19). All HTLV-cell lines triggered IFN-I production by co-cultured pDCs, albeit at different levels ranking from lower to higher inducer cell lines, at the optimal pDC/infected cells ratios (S5D Fig), as follows: C19, MO, C91, Hut102 and MT-2 cell lines (Fig 5A). Neither differences in the amount of intracellular genomic RNA nor viral RNA released in the supernatant of infected cells was correlated to the observed differences in the induction of pDC IFN-I response (Fig 5B and 5C and S5A and S5B Fig). This suggests that the amount of RNA produced by the infected cells is not rate limiting for activation of pDCs.
Thus, we next assessed whether frequency of pDCs engaged in contacts with the different HTLV-infected cell lines regulated the intensity of pDC activation. The frequency of pDC conjugates with the different HTLV-infected cell lines was similar, except higher level for the HTLV-infected MO cell line (Fig 5D). Nonetheless, this higher frequency of cell conjugates with the HTLV-infected MO cell line did not translate into higher levels of IFN-I production (Figs 5D and S5C), suggesting that additional factor(s), other than pDC ability to establish contact with HTLV-infected cells, govern(s) the levels of pDC IFN-I response to HTLV-infected cells.
We thus tested whether variations of pDC induction by HTLV-infected cell lines might be explained by distinct mechanisms for viral capture by pDCs. To address this, we first evaluated viral binding and internalization in the pDCs upon co-culture with the different HTLV-1/2 cells lines by detection of intracellular p19gag in the CD123+ pDCs population by FACS (Fig 5E). Except pDCs co-cultured with C91-PL, we observed no differences in virus binding on pDCs. This suggests that the reduced IFN response induced by C19 cells does not result from diminished HTLV-2 capture by pDCs. Furthermore, consistent with results obtained using cell-free virus [22], HTLV-1 infected cells induced TRAIL expression by co-cultured pDCs (Fig 5F), as did HTLV-2 infected cells (Fig 5F). This suggests that the reduced pDC IFN response to C19 cells is not associated with other impairment link in their ability to respond to virus. Furthermore, using C19 cells that induced the lowest pDC IFN-I production (Fig 5A), we showed that both the pDC response and viral transmission were significantly out-competed by recombinant RBD, but not by recombinant VEGF165 (S6A–S6D Fig), suggesting that different HTLV viral receptor usages are not likely responsible for the difference in pDC IFN-I response to the various cell lines. Additionally, pDC sensing of C19 cells was specifically inhibited by TLR7 inhibitor (S6E Fig). This rules out the involvement of other PRR that would induced lower IFN-I induction upon HTLV-2 sensing, as suggested for other viruses [28].
As we showed that pDCs are activated by cell-cell contacts with infected cells and via viral biofilm-like structures, we then compared the viral accumulation at the surface of the panel of HTLV-infected cell lines (S6F Fig). The p19gag proteins were detected as patch/cluster at the surface of all infected cells, suggesting that the pDC activation is not directly linked to an absence of virus accumulation at the surface of the different HTLV-infected cell lines.
We next asked whether the level of pDC activation by the HTLV-infected cell lines correlate with their ability to transfer infectious virions to target cells. Regressive exponential correlation analysis revealed that IFN-I production by pDCs was inversely correlated with the ability of infected cells to transfer infectious virions (Fig 5G and 5H, p-value = 0,011). Altogether these results indicate that the sensing of HTLV-infected cells by pDCs is not strain-specific, and, importantly, inversely correlated to infectivity transmission to alternative target cells. It thus implies that pDC activation is likely modulated by other features of the infected cells.
Our results using heparin suggested that glycosylated proteins, including HSPGs are involved in the tethering of HTLV-1 stimulating signals to the pDC surface and/or its transfer, resulting in increased pDC IFN-I production. The density of surface glycosylation, including HSPGs [48], is known to be cell type specific [49]. We thus quantified cell surface glycosylation using various lectins known to bind different terminal glycosylation patterns (S7A Fig). The staining by Peanut agglutinin lectin (PNA), which bind to oligosaccharide structures with terminal β-galactose residues on the different HTLV-infected inducer cells (S7B Fig) revealed that the amount of this type of surface glycosylation was inversely correlated to the magnitude of IFN-I production by co-cultured pDCs (Fig 6A and 6B). As opposed, the levels of PNA lectin staining at the surface of infected cells positively correlated with their ability to transmit viral infection to target cells (Fig 6C). Consistent observations were obtained by confocal microscopy analysis of PNA lectin, displaying very weak PNA staining for MT2 and Hut102, the highest IFN-inducer cell lines (S7C Fig). Similar trend was observed for stained SBA-lectin (S7B Fig), thought to detect α- or β-linked N-acetylgalactosamine residues, albeit to with lower magnitudes of difference (S7B Fig), and without statistical correlation with IFN-I production (S7D Fig). In contrast, binding of the lectins UEIA, WGA and ConA, that recognized other glycosylated residues, did not demonstrate difference between the different cell lines (S7B Fig). Altogether, these results suggest that the composition of the terminal oligosaccharide residues, especially dense terminal β-galactose glycosylation, present at the surface of HTLV-1-infected cell lines might inversely govern both IFN-I response by co-cultured pDCs and viral transmission.
To further study the role of terminal β-galactose residues in IFN-I induction, we performed assay to mask such surface glycosylation by pretreating C19 cells with PNA-lectin prior contact with pDC. We controlled that the PNA concentrations used in the co-culture did not affect cell viability (S8A Fig). The presence of PNA-lectin at the surface of C19 cells significantly increased pDC-induced IFN-I production (Fig 6D). Conversely, the removal of sialic acid using neuraminidase treatment resulted in augmented exposure of terminal β-galactose at the surface of treated C91 cells (i.e., at levels similar to that of C19 cells, Fig 6E), and in significant decreased of pDC-induced IFN-I production (Fig 6D). Of note, the limited impact of neuraminidase treatment on pDC IFN-I production likely result from its short timeframe of impact on the exposure of β-galactoside residues, as revealed by the reduction overtime of PNA staining of neuraminidase-treated cells (S8B Fig). Thus, to strengthen the role of β-galactoside residues in the regulation of pDC IFN-I production, we determined whether viral expression in PBMCs from HAM/TSP patients was also associated with a modulation of PNA staining. In vitro culture of primary PBMCs from both healthy donors and HAM/TSP patients is enough to expose β-galactoside residues (S8C and S8D Fig). Consequently, β-galactoside was not preferentially/exclusively induced in PBMCs that expressed HTLV-1 (S8E Fig). However, we observed a higher proportion of infected PBMCs that expressed β-galactoside residues in the PBMCs from patient #1620 compared to the two others (S8E Fig). Interestingly, although the level of virus re-expression in PBMCs from patient #1620 was the highest (i.e. 70% of positive PBMCs compare to 40% for #1668 and 5% for #1485, see S1 Fig), this was not associated with a higher induction of type-I IFN by co-cultured with pDCs (Fig 1I, compare patients #1620 to #1668). Altogether, our results show that β-galactoside glycosylations at the surface of infected cells likely negatively regulate pDC activation by HTLV-infected cells.
Evidences suggested that IFN-I response is likely pivotal to repress HTLV-1 replication [18,19,50,51]. Yet, the persistent HTLV infection is thought to result from escape viral mechanisms and consequent failure of the immune detection and clearance of infection [52]. Along this line, HTLV-1 inhibits IFN-I induction and signaling [12–15] [16], leading to very limited production of IFN-I by infected cells. In this study, we elucidated an alternative sensing pathway mediated by the recognition of infected cells by pDCs, a sentinel cell type known to be a potent producer of IFN-I. We demonstrated that pDCs preferentially sense HTLV-1 infected cells via physical contact rather than HTLV-1 cell-free virions. This sensing pathway is thus congruent with the absence of, or very low, detection of cell-free virus in the blood of infected patients [53], a consequence of active repression of viral expression [54]. Recent report suggested that viral latency observed in vivo might be transiently relieved under changes in nutrients availability in the extra-cellular environment of the infected cells [55], thus potentially leading to sporadic viral expression in privileged sites such as lymphoid organs. Thus localized IFN-I response by pDCs upon contact with transiently reactivated infected cells would be in agreement with the absence of detection of IFN-I at the systemic level in infected carriers [56].
HTLV-1 viral transmission through physical contact likely compensates for the low levels of cell-free viruses found in the patients and/or their poor infectivity [35,53,57]. Upon cell contact, the virus is transmitted either through a virological synapse, in which virus assembly and budding are polarized toward the cell contact [58], or through the transfer of viral biofilm-like structures, an extracellular accumulation of viruses embedded in the infected-cell extracellular matrix (ECM) [34], both mechanisms being likely not mutually exclusive. Here, we showed that isolated viral biofilm-like structure is sufficient to trigger IFN-I production by pDCs. We further propose that the increased potential of biofilm-like structures for pDC activation, as compared to cell-free virus, could be due to components of these structures favoring the transmission of the pDC-activating signal, possibly by tethering the immunostimulatory RNA carrier to the pDC surface. Although pDCs are largely refractory to most viral infection [23], Jones et al. [5] observed viral production by pDCs exposed to cell-free HTLV-1 for at least 3 days. This timeframe is longer than our experimental setting (i.e., 24 hours, in accordance with previously reported pDC half-life limited to couple of days [59,60]), and during which we failed to detect productive HTLV-1 infection of pDCs. Thus, productive infection of pDCs by HTLV-1 is very unlikely needed for their rapid IFN-I production induced by contact with infected cells.
Our results suggested that pDC sensing of HTLV-infected cells is mediated via the HTLV entry receptor Glut-1, and NRP-1/BDCA-4 receptor seems dispensable. Previous report showed that infection of myeloid DC by cell-free HTLV-1 particles is independent of NRP-1/BDCA4, viral binding being ensured by the DC-SIGN lectin [61]. It is conceivable that pDC-expressed glycosylated surface factors, including lectins and HSPGs known as capture molecules for HTLV and other viruses, could act as cofactor for HTLV capture at the pDC surface via glycan-mediated interactions with HTLV before its delivery to Glut-1. Alternatively, virus delivery through cell-cell contact could bypass the need for attachment factors and virus concentration at the surface of target cells before virus interaction with its cognate receptor. Our results showed that heparin, an HSPG mimic, increased the frequency of pDC contacts with HTLV-1 infected cells and IFN-I production by pDCs in response to HTLV-1-infected cells, implying a putative function of HSPG in the stabilization of the pDC-infected cells interface required for the capture of the immunostimulatory RNA carrier present in the HTLV-1 biofilm-like structures by pDCs.
Our results uncovered differences of surface glycan pattern of HLTV-infected cells, including the composition of the terminal oligosaccharide residues at the cell surface (i.e., plasma membrane), with densification of certain residues being inversely correlated to the level of pDC IFN-I response. Further, enzymatic and pharmacological inhibition/modulation of the cell surface glycans impacted pDC response to infected cells. These observations support the proposition that the extracellular matrix and/or glycosylated proteins expressed at the plasma membrane of the infected cells likely govern both IFN-I production by co-cultured pDCs and viral transmission. Importantly, our results suggest that these two processes are inversely correlated.
By analogy to previously reported abilities of several viruses to modify ECM composition of the host cells to favor their own dissemination and/or immune escape [62–64], one might speculate that the chronic infection by HTLV modulates the cell surface glycan pattern [65]. We showed that terminal β-galactoside glycosylation density is inversely correlated with the ability of infected cells to promote contact-dependent pDC IFN-I production. Together with the absence of correlation between the levels of pDC IFN production, the amount of viral RNA production or capture of HTLV by pDCs, shown for different HTLV-infected cells, our results suggest that composition of the extracellular matrix and/or cell surface expression of glycosylated proteins that embed cell surface-attached viruses of the different HTLV-infected cells might regulate pDC activation. The results obtained using PBMCs from HAM/TSP patients suggest that β-galactoside residues induction at the surface of infected cells might not however be regulated by viral expression. In addition, we could not determine whether these residues are specific components of the viral biofilm, or whether β-galactoside-containing proteins surrounding the viral biofilm at the plasma membrane of the infected cells are enough to regulate IFN-I production.
Previous reports showed that pDC response to viral infections can be modulated by different cell surface factors including the regulatory receptors ILT7, BDCA2 or DCIR [66,67]. For example, ILT7 binds BST-2, an IFN-induced gene, initially described as an HIV restriction factor that impedes viral release from infected cells [68]. Since HIV Vpu protein counteracts viral tethering by BST2, virus-free surface BST-2 can readily interact with ILT7, and thereby inhibits pDC IFN-I production [69]. While BST-2 is also expressed upon HTLV-1 infection, as opposed to the negative regulation of HIV release, it participates in efficient HTLV-1 cell-cell transmission [70]. Other negative regulatory receptors e.g., BDCA2 and DCIR bind complex-type sugars chains (terminal β-galactoside containing complex sugars [71,72], and mannose/fucose containing complex sugars [73,74], respectively). Since pDC IFN-I response to HTLV is modulated by the available terminal sugar composition at the surface of infected cells, one might speculate that the density of specific glycans, via the interaction with negative regulators, e.g., BDCA2 might regulate the levels of pDC IFN-I production induced by HTLV-infected cells as previously shown in other contexts [75–77]. Interestingly, these dense specific glycans might not be part of the biofilm, but still might be engaged after contact with the pDC. Thus, along with the interaction of the viral envelop protein with the Glut-1 receptor, other cell surface factors, including heparan sulfate-containing proteins, and terminal galactoside-conjugated proteoglycans, at the pDC/infected-cell interface, could regulate the strength of pDC activation. Altogether our results provided an original illustration of the regulation of pDC IFN-I response by the surface glycan pattern of infected cells.
Jurkat cells (from ATCC, ref ACC 282) stably transfected with a plasmid encoding the luciferase (Luc) gene under the control of the HTLV-1 long terminal repeat (LTR) promoter and the HTLV-1 Tax-transactivator (Jurkat-LTR-Luc) [34] were maintained under hygromycin selection (450 μg/mL, Sigma) in culture RPMI medium: RPMI 1640 medium supplemented with 10% fetal calf serum (FCS; Gibco Life Technologies), L-Glutamine (2 mM, Gibco Life technologies) and penicillin-streptomycin (100 U/mL and 100 μg/mL respectively; Gibco Life Technologies). C91-PL (HTLV-1 infected T-cell line, Cellosaurus, ref CVCL_0197), MT-2 (HTLV-1 infected T-cell line, NIH, ref 237 and [78]), Hut102 (HTLV-1 infected T-cell line, Cellosaurus, ref CVCL_3526 and [79]) and C8166 (HTLV-1 infected T-cell line which does not produce infectious virus [39], ECACC ref 88051601) were maintained in culture RPMI medium. PBMCs from healthy blood donors or from HAM/TSP patients were cultured 18h in RPMI medium supplemented with 20% FCS supplemented with IL-2 (150 U/mL) and PHA (1μg/mL). C19 (HTLV-2 infected cell line, obtained from [80]) and MO (HTLV-2 infected cell line, ATCC ref CRL-8066) were maintained in culture RPMI medium supplemented with 20% FCS. The human fibrosarcoma cell line containing a plasmid encoding the luciferase gene under the control of the immediate early IFN-I inducible 6–16 promoter (HL116) (a kind gift from S. Pelligrini, Institut Pasteur, France) [81] was maintained under HAT selection in DMEM medium supplemented with 10% FCS and penicillin-streptomycin (100 U/mL and 100 μg/mL respectively). All cells were grown at 37°C in 5% CO2.
Ficoll-Hypaque (GE Healthcare Life Sciences); LPS, TLR7 agonist (R848) and TLR9 agonist (ODN2216) (Invivogen); TLR7 antagonist, IRS661 (5’-TGCTTGCAAGCTTGCAAGCA-3’) synthesized on a phosphorothionate backbone (MWG Biotech); Fc Blocking solution (MACS Miltenyi Biotec); BDCA-4-magnetic beads for selective isolation of pDCs (MACS Miltenyi Biotec); IL-6 ELISA kit (Affymetrix, eBioscience); Lipofectamine 2000 (Life Technologies); 96-well format transwell chambers (Corning); LabTek II Chamber Slide System, 96-Well Optical-Bottom Plates and Nunc UpCell 96F Microwell Plate (Thermo Fisher Scientific); Vibrant cell-labeling solution (CM-DiI, Life Technologies); rat anti-HSPG antibody (clone A7L6, Upstate Biotechnology) Hoescht and anti-mouse AlexaFluor 647-conjugated secondary antibody (Life Technologies); anti-mouse DyLight 488-conjugated secondary antibody (Vector); anti-rat APC-conjugated secondary antibody (SouthernBiotech); High Capacity cDNA reverse transcription kit (Applied Biosystems); Powerup Sybr Green Master Mix (Applied Biosystems); pDC specific markers: mouse PE or APC-conjugated anti-CD123 (clone AC145, Miltenyi), mouse APC-conjugated anti-BDCA-2 (AC144; Miltenyi); mouse PE-conjugated anti-TRAIL (ThermoFisher); Metalloproteinase 9 (Enzo Life Sciences); FITC-conjugated Peanut Agglutinin (PNA) (Sigma Aldrich); Alexa Fluor 680-conjugated Wheat Germ Agglutinin (WGA) (Thermo Fisher); FITC-conjugated WGA, Soy bean Agglutinin (SBA), Ulex europaeus agglutinin I (UEA-I) and Concanavalin A (ConA) were from Vectors; poly-L-lysine (Sigma, P4832), anti-HTLV-1 p19gag antibody (1:1000, Zeptometrix); recombinant VEGF165 protein (R&D); Glut-1.RBD.GFP (Metafora biosystem); recombinant IFN-α 2a (TEBU BIO PBL); anti-HTLV-1 Env gp46 antibody (1:1000, Zeptometrix), luciferase reporter activity assay (Promega); paraformaldehyde 20% (PFA; Electron Microscopy Sciences); saponin (Sigma); Heparin (Sigma). HTLV-1 molecular clone (pACH) and HTLV-1 molecular clone lacking the expression of the envelope protein (pACH ΔEnv, [82]) were provided by Dr. Pique (Institut Cochin, France). The Alexa Fluor 488-conjugated anti-Tax antibody (LT-4) was provided by Pr Tanaka (University of Ryukyus, Japan).
The pDCs and PBMCs were isolated from blood or cytapheresis units from healthy adult human volunteers which was obtained according to procedures approved by the “Etablissement Français du sang” (EFS) Committee. All donors provided informed consent to EFS. PBMCs from HAM/TSP patients were obtained in the context of a Biomedical Research Program approved by the Committee for the Protection of Persons, Ile-de-France II, Paris (2012-10-04 SC). All individuals gave informed consent.
PBMCs were isolated using Ficoll-Hypaque density centrifugation. pDCs were positively selected from PBMCs using BDCA-4-magnetic beads (MACS Miltenyi Biotec) and pDCs were depleted from PBMCs, as previously described [24,25]. The typical yields of PBMCs and pDCs were 800x106 and 2x106 cells, respectively, with a typical purity of >91% pDCs, as we previously reported [24,25].
After isolation, pDCs (2x104) were platted in 96-well round bottom plates and cultured at 37°C with HTLV-1 or HTLV-2 infected cell lines (2x104 or other count when indicated), or with PBMCs from healthy donors or from HAM/TSP patients (2x104), or with Jurkat cells microporated with the HTLV-1 molecular clone pACH or pACH ΔEnv (2x104), or with Jurkat cells (2x104) as negative control, or with isolated HTLV-1 biofilm-like structures (100μL), or with HTLV-1 biofilm-like structures depleted cells (2x104). When indicated, HTLV-2 infected cells (C19, 106 cells in RPMI culture medium) or HTLV-1 infected cells (C91, 106 in PBS) were treated with PNA (10 μg/ml, SIGMA) for 30 min at 4°C or neuraminidase (0,1U/ml, SIGMA) for 1h at 37°C respectively. Treated cells were then washed twice in RPMI culture medium prior to co-culture (2×104) with pDC (2×104). Culture with isolated pDCs or PBMCs were maintained in RPMI 1640 medium (Life Technologies) supplemented with 10% FCS, 10 mM HEPES, 100 units/mL penicillin, 100 μg/mL streptomycin, 2 mM L-glutamine, non-essential amino acids and 1 mM sodium pyruvate at 37°C/5% CO2. The supernatants were collected at 20-24h after co-culture. When indicated, infected cells or uninfected cells were co-cultured with pDCs in 96-well format transwell chambers separated by a 0.4 mm membrane (Corning), as previously [24,25].
HL116 cells were seeded at 2.104 cells/well in 96-well plate 24 h prior the assay, and incubated for 17 h with supernatant collected from pDC co-cultures (100 μL) or serial dilution of recombinant human IFN-α 2a (PBL Interferon Source), used for standard curve titration. Cells were then lysed and luciferase activity assayed. IFN-I levels were expressed as equivalent of IFN-α 2a concentration, in Unit/mL. The detection of IL-6 by ELISA was performed as previously[24] using kit (Affymetrix, eBioscience) and according to the manufacturer instructions.
Jurkat cells (8x104 cells) were transfected with 3 μg of pACH or pACH ΔEnv together with 1 μg of pSG5M-Tax1 [83] using the Neon Transfection System (ThermoFischer Scientific) following manufacter’s instructions. Cells were cultured 48h at 37°C before co-culture with pDCs.
HTLV-1 viral biofilm-like structure was prepared with a method that is slightly different from the original one [34] and as previously described [35]. Briefly C91-PL cells were platted (3x105 cells/mL) and cultured for 4 days. HTLV-1–infected cells were washed twice in RPMI-1640 serum-free medium and incubated at 1x106 cells/ml for 1 h at 37°C, with gentle agitation every 10 minutes. Then, FCS (10% final) and penicillin-streptomycin (100 μg/mL final) were added, and cells centrifuged. Supernatant containing biofilm-like structures preparation was collected and supplemented with Hepes (10 mM), non-essential amino acid (2.5 mM), sodium pyruvate (1 mM), β-mercaptoethanol (0.05 mM) before immediate use.
Cell-free viruses were also obtained from C91-PL cells (106 cells/mL) cultured for 24h at 37°C 5% CO2. Supernatant were clarified by centrifugation (5 minutes at 800g) and filtrated through a 0.45 μm-diameter pore filter (Millipore, MA) to eliminate cell debris. Virions were purified by ultracentrifugation through a 20% (wt/vol) sucrose cushion at 100,000g in SW32 (Beckman) for 1h30 at 4°C and stored at -80°C before use. Virus concentration was determined using Retrotek HTLV-1/2 p19gag Antigen ELISA kit (Zeptometrix) following manufacturer’s instructions and as previously described [19].
C91-PL cells (106 cells /mL) were treated with Metalloproteinase 9 (20 nM) in RPMI serum-free medium for 1h at 37°C 5% CO2. Cells were washed twice with culture RPMI medium, and immediately used. The efficacy of HTLV-1 viral biofilm-like structures shedding was controlled by analyzing gp46 viral envelope level by FACS and viral transmission to T-cells using Jurkat LTR-Luc reporter cells.
HTLV-1 (C91-PL, Hut102, MT-2), HTLV-2 (C19, MO) or uninfected (Jurkat) cell lines (103, 104 or 105) were co-cultured with Jurkat LTR-Luc cells (104). Different ratio of infected cells/target cells (1/10; 1/1 or 10/1) were incubated for 24 hours in round-bottom 96-wells plates at 37°C. Cells were washed once with PBS and stored at -80°C as dry pellets until assayed for luciferase reporter activity using manufacturer’s instructions (Promega). Luciferase results were normalized according to the amount of proteins determined by Bradford (Biorad).
Jurkat LTR-Luc cells (2x105) were incubated in culture RPMI medium with VEGF165 (80–100 ng/mL) or 10μL Glut-1.RBD.GFP at 4°C during 30 minutes before co-culture with C91-PL cells (2x104) or with cell-free viruses (50 ng/mL of p19gag equivalent as measured by ELISA) for 2 hours at 37°C. Cells were then harvested, washed with PBS, fixed in 4% PFA, permeabilized in PBS / 1% BSA / 0.05% Saponin and stained with an anti-p19gag antibody (1:1000) followed by FITC or Alexa Fluor 549-conjugated anti-mouse antibody. Fluorescence was acquired on at least 10 000 events with a FACSCanto II cytometer (BD Biosciences) and data analyzed on FlowJo software (Tree Star, Inc. Ashland, OR).
HTLV-1 (C91-PL, Hut102, MT-2), HTLV-2 (C19, MO) or uninfected (Jurkat) cell lines (2x105) were fixed with 4% PFA and stained with FITC-conjugated lectins (10 μg/ml) for 30 minutes at 4°C. The level of Env gp46 surface expression was determined on unfixed C91-PL cells or on Jurkat transfected cells using anti-HTLV-1 Env gp46 antibody (1:1000 in PBS-1% BSA) for 1h at 4°C followed by Alexa488-coupled anti-mouse antibody for 30 min at 4°C. pDC were surface-stained with mouse PE or APC-conjugated anti-CD123 and mouse APC-conjugated anti-BDCA-2, with Glu1.RBD.GFP protein (5μl/ 1x105 cells, Metafora), or with anti-NRP-1 (clone 12C2, Biolegend). Alternatively, pDCs were fixed in 4% PFA and stained with anti HSPG antibody (1:100) for 30 min at 4°C followed by APC-conjugated anti rat antibody (1:100) for 30 min at 4°C. Cells were then washed with PBS and fluorescence acquired using 20 000 events on a FACSCanto II cytometer, and analyzed with FlowJo software (Tree Star, Inc. Ashland, OR).
Transfected Jurkat cells or PBMCs from HAM/TSP patients were cultured 18h in presence or not of IL-2 and PHA and fixed with 4% PFA, permeabilized in PBS / 1% BSA / 0.05% Saponin, and stained with anti-p19gag antibody (1:1000) for 30 min at 4°C followed by DyLight488-conjugated anti-mouse antibody (1:1000). Cells were then washed and fluorescence acquired using at least 10 000 events on a FACSCanto II cytometer (BD Biosciences), and analyzed with FlowJo software.
After 24h co-culture with HTLV-1/2 infected cells, pDCs were collected, washed and stained with PE-conjugated anti-TRAIL and APC-conjugated anti-BDCA-2 antibodies. Cells were then washed and fixed in 4% PFA. Fluorescence was acquired using at least 10 000 events with a FACSCanto II cytometer (BD Biosciences) and analyzed with FlowJo.
pDCs (105) were co-cultured with C91-PL (HTLV-1 infected cells, 105) in the presence or not of Glu1.RBD.GFP (10μl) for 4h. Cells were then washed in PBS. For subsequent viral binding analyses, cells were surface-stained with anti-gp46 antibody (1:1000) followed by Alexa Fluor 647-conjugated anti-mouse antibody (1:1000). For viral capture analyses, cells were fixed in 4% PFA, permeabilized in PBS / 1% BSA / 0.05% Saponin, and stained with anti-p19gag antibody (1:1000) followed by Alexa Fluor 647-conjugated anti-mouse antibody (1:1000). After washing, pDCs were surface-stained with anti-CD123-Vioblue-conjugated antibody and fixed in 4% PFA.
pDCs (105) or MDDCs (2.5x105) were co-culture with C91-PL (105) for 24h or 72h respectively. For pDCs infection analysis, cells were washed, surface-stained with Vioblue-conjugated anti-CD123 antibody, fixed and permeabilized according to the manufacturer’s instructions (eBiosciences). For MDDCs infection analysis, cells were washed in PBS and in normal goat serum (7%, Sigma), fixed and permeabilized according to the manufacturer’s instructions (eBiosciences). pDCs or MDDCs were stained with biotin-coupled anti-Tax antibody (LT4) followed by streptavidin labeled with PE-Cy7 (BioLegend, Ozyme). After extensive washes, MDDCs were finally surface-stained with a V450-coupled anti-CD11c antibody. Fluorescence was acquired using at least 10 000 events with a FACSCanto II cytometer and data analyzed on FlowJo software.
RNAs were isolated from samples harvested in guanidinium thiocyanate citrate buffer (GTC; Sigma-Aldrich) by phenol/chloroform extraction procedure as previously described [25]. Reverse transcription was performed using the random hexamer-primed High Capacity cDNA reverse transcription kit (Applied Biosystems) and quantitative PCR was carried out using the Powerup SYBR Green Master Mix (Applied Biosystems). The absolute numbers of HTLV-1 transcripts were normalized to the total amount of RNA. For supernatant samples, qRT-PCR was controlled by the addition of exogenous carrier RNAs encoding xef1α (xenopus transcription factor 1α) in supernatant diluted in GTC buffer, as previously described [24,25].
For quantification of viral genomic RNA the following primers were used: HTLV-1 Forward (AAAGCGTGGAGACAGTTCAGG), HTLV-1 Reverse (CAAAGGCCCGGTCTCGAC), HTLV-2 Forward (CCTTGGGGATCCATCCTCTC), HTLV-2 Reverse (TCTCTAAAGACCCTCGGGGAG). For quantification of viral RNA present in the supernatant of infected cells, the following primers were used: Tax 1 Forward (GGATACCCAGTCTACGTGTTTGG), Tax 2 Forward (GGATACCCCGTCTACGTGTTTGG), Tax 1/2 Reverse (GGGGTAAGGACCTTGAGGGT).
HTLV-1 (C91-PL, Hut102, MT-2), HTLV-2 (C19, MO) cell lines and control uninfected Jurkat cells (5x105) were transduced in 24 well plate with lentiviral-based vector pseudotyped with VSV glycoprotein to stably express GFP. Briefly, 105 GFP-expressing HTLV infected cells and control cells were co-cultured with 4x104 pDCs in low-adherence micro-plate designed for cell harvesting by temperature reduction (Nunc UpCell 96F Microwell Plate from Thermo Scientific) for 4–5 hours at 37°C, and, as indicated, in presence or not of heparin. The co-cultured cells were detached by 20 minute-incubation at room temperature, harvested and fixed in 4% PFA. Cells were then washed twice with staining buffer (PBS 2% FBS), and pDCs were stained by the pDC-specific CD123 marker. Co-cultured cells were analyzed by Image Stream X technology (Amnis) at magnification x40 using IDEAS software, as previously described [24,25]. The cell population defined as pDC/HTLV-infected or uninfected cell conjugates comprises conjugates of at least one CD123 APC positive cell and at least one GFP positive cell among the total of CD123 APC-positive cells, GFP-positive cells and conjugates. As shown by representative images (S4A Fig), the population gated as pDC-HTLV-1-infected cell conjugates corresponded to GFP positive/CD123 positive cell conjugates, as expected, with 80–95% purity. The cell populations were sorted using masks (IDEAS software) to eliminate cells out of focus and/or with saturating fluorescent signal, and then selected based on cell size of the positive cells (i.e., fluorescent signal area).
For lectin/HTLV-1/2 virus localization analysis, HTLV-1 (C91-PL, Hut102, MT-2), HTLV-2 (C19, MO) or uninfected (Jurkat) cell lines cultured on Lab-tek chamber slides (Nunc) previously treated with 0.01% poly-L-lysine (Sigma, P4832) were surface-stained with FITC-conjugated PNA or WGA (10μg/ml), fixed in 4% PFA, then permeabilized and stained with antibodies against HTLV-1/2 matrix protein p19gag (1:1000) followed by Alexa Fluor549-conjugated anti mouse antibodies. Cells were counterstained with DAPI-Fluoromount G before analysis on Zeiss LSM 800 microscope. Images were acquired on ImageJ.
For pDC/HTLV-infected cells conjugates analysis, 4x104 pDCs were stained using 0.5 μM Vibrant cell-labeling solution as previously[24]. Labeled pDCs were washed twice with PBS and then co-cultured with 2x104 HTLV infected cells for 4–5 hours at 37°C, in a 96-well optical-bottom plate pre-coated with 8 μg/mL poly-L-lysin for 30 minutes at 37°C. Cells were then fixed in 4% PFA, washed with PBS and PBS-3%BSA, and stained with anti-HTLV-1 Env gp46 antibody (1:1000 in PBS—3%BSA) for one hour at room temperature. Prior antibody staining, cells were stained by WGA lectin coupled to Alexa 680 (Molecular Probes, ref W32465) diluted at 10μg/mL in HBSS, for 10 minutes at room temperature, then washed three times with HBSS. After three washes with PBS, cells were incubated with Alexa488-conjugated-anti-mouse antibody in 3% BSA-PBS and added to the cells along with Hoechst diluted at 1:500 (Molecular Probes) for 1 hour at room temperature. After three washes with PBS, cells were observed with a Zeiss LSM 710 laser scanning confocal microscope. The quantification of the phenotypes defined to as clusters at the contact were performed using Image J software package.
Statistical analysis was performed using PRISM v7.03 software (Graphpad). One-way analysis of variance (ANOVA) with Sidak’s multiple comparison test was used to determine statistically significant differences. Paired two-tail t-test was used to compare two groups from the same donor. Differences were considered significant if the p-value was < 0.05.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.